Integrating Retrieval-Augmented Generation and Thematic NLP for Vaccine Confidence Modeling in Alaska

Integrating Retrieval-Augmented Generation and Thematic NLP for Vaccine Confidence Modeling in Alaska | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Advisory Board Help Center Sign In Submit a Preprint Cite Share Download PDF Research Article Integrating Retrieval-Augmented Generation and Thematic NLP for Vaccine Confidence Modeling in Alaska Luay Abdeljaber, Sultan Alsarra, Latifur Khan, Renee F. Robinson, and 2 more This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-7368501/v1 This work is licensed under a CC BY 4.0 License Status: Posted Version 1 posted You are reading this latest preprint version Abstract Vaccine misinformation poses a significant public health threat, particularly in communities with varying levels of vaccine confidence. This study investigated vaccine hesitancy across Alaska’s diverse communities by triangulating public sentiment from social media with individual beliefs gathered through qualitative interviews. The aim was to explore how online discourse influences vaccine-related decision-making and to develop tools for real-time misinformation detection. We employed a mixed-methods approach, analyzing 1,300 Alaska-specific tweets and conducting 87 semi-structured interviews across urban and rural communities. A Retrieval-Augmented Generation (RAG) system was developed, integrating the context-rich LLaMA-2-7B model with the efficient T5-Base model to balance accuracy and computational performance. The system used sentence embeddings and FAISS-based similarity search to identify misinformation themes and generate context-aware responses grounded in real-world data. Sentiment analysis revealed that rural social media posts exhibited significantly higher negativity and misinformation (55.6% negative sentiment) compared to urban posts. In contrast, interview data reflected more balanced and nuanced attitudes toward vaccination. Thematic analysis identified systemic distrust and personal beliefs, particularly among Indigenous and rural populations, as key drivers of hesitancy. Model evaluation showed that LLaMA-2-7B outperformed T5-Base in contextual accuracy, while T5-Base offered faster but occasionally less accurate responses. By combining AI-driven insights with ethnographic data, this study highlights the divergence between online narratives and lived experiences. The proposed framework offers a scalable, real-time method for detecting misinformation and informing culturally responsive public health messaging. Future work will focus on optimizing system efficiency and collaborating with digital platforms to reduce the spread of viral misinformation. Misinformation Vaccination Sentiment NLP RAG Social Media Public Health Figures Figure 1 Figure 2 Highlights Retrieval-Augmented Generation system detects vaccine misinformation using real-world social media data. Combined Llama-2 and T5 models for balanced accuracy and response speed. Sentiment analysis revealed distrust-driven negativity, especially in rural areas. Interview sentiment diverged from social media and showed more neutral attitudes. Triangulating NLP and ethnographic data improves public health insights. Introduction Vaccine hesitancy continues to pose a critical public health challenge worldwide 1 , 2 , with rural and underserved populations like those in Alaska. 3 Here, geographic isolation, historical trauma, particularly among Indigenous populations, and distinct sociopolitical attitudes toward government and health autonomy shape divergent perceptions of vaccination. In addition, these communities often face intersecting barriers to vaccine uptake, including limited healthcare access, socio-cultural resistance, and deeply rooted institutional mistrust. 4 This has been compounded by the COVID-19 pandemic 5 , which brought a proliferation of misinformation through social media platforms that outpaced public health responses. 6 Social media algorithms are designed to maximize user interaction by amplifying emotionally charged and polarizing content, creating feedback loops that reward controversy over accuracy. 7 In this environment, vaccine myths, particularly those related to safety, efficacy, and side effects, can rapidly go viral, entrenching skepticism and distrust. 8 A wide range of factors contribute to vaccine hesitancy, including psychological concerns 9 , cultural and religious beliefs 10 , political ideologies 11 , and longstanding mistrust of health institutions. 12 In the United States (U.S.), these drivers are magnified in historically marginalized communities, where public health campaigns must contend with legacies of systemic discrimination and medical exploitation. 13 This multifaceted landscape calls for context-specific research that accounts for the structural, cultural, and informational dynamics at play. Artificial intelligence (AI) offers powerful tools to investigate and address vaccine hesitancy at scale. Natural language processing (NLP) models, in particular, have been used to analyze health-related discourse, detect misinformation, and assess public sentiment across large datasets. While lexicon-based tools like VADER offer speed and simplicity, transformer-based models, such as DistilBERT and CardiffNLP’s RoBERTa, enable more accurate sentiment classification in informal, short-form text like tweets. 13 – 15 Retrieval-Augmented Generation (RAG) models represent a recent advancement in NLP, combining document retrieval with generative modeling to produce context-aware responses grounded in external evidence. 15 , 16 These models have demonstrated effectiveness in applications such as fact-checking and misinformation detection, though their use in localized public health contexts remains limited. 17 This study applies a novel, AI-augmented framework to investigate vaccine hesitancy in Alaska. We combine sentiment analysis of geotagged Twitter data with qualitative thematic analysis of in-depth interviews, offering both macro-level insights into public discourse and micro-level understanding of individual beliefs. Our approach incorporates a dual-model RAG system, using T5-Base for efficient, factual outputs and LLaMA-2-7B for nuanced, context-rich generation, alongside FAISS vector indexing and transformer-based sentiment classification. This study aims to understand vaccine hesitancy within Alaska’s diverse communities by triangulating public sentiment (via social media) and individual beliefs (through qualitative interviews, combining technological insights with ethnographic data to create a multi-layered approach to vaccine misinformation detection. The analysis is framed within a novel Grounded Theory that integrates historical trauma, systemic distrust, and cultural beliefs as foundational drivers of vaccine hesitance. This theory was specifically suggested to address the unique sociopolitical and cultural dynamics present in Alaska, particularly among indigenous and rural populations. By emphasizing the local context and community-driven narratives, the study aims to offer a more comprehensive understanding of vaccine hesitancy that builds upon relevant public health models such as the Theory of Change, offering targeted insights for culturally informed public health messaging. Methods Sampling and inclusion criteria Participants were adult residents of Alaska (18+) with varying levels of vaccine hesitancy. Purposive sampling ensured representation across gender, race, geography (urban vs. rural), and hesitancy levels (hesitant, undecided, vaccinated). The sample included diverse ethnic groups, such as Native Alaskan, White, African American, Hispanic, and Asian residents. The novel grounded theory framework identifies systemic distrust as a central factor influencing vaccine hesitancy, particularly among indigenous communities, rooted in historical mistrust of government and healthcare institutions. 18 A total of 87 semi-structured interviews were conducted via Zoom between July and November 2024, each lasting 30–40 minutes. Data collection Social media data were sourced from a publicly available Kaggle dataset 19 and filtered for tweets geotagged within Alaska. Posts included were chosen based on keywords explicitly referencing vaccine-related topics, hesitancy, safety concerns, or COVID-19-related terms. Data analysis Interview transcripts were coded thematically to identify patterns in hesitancy and information sources across rural and urban populations. Approximately 1,300 tweets were preprocessed for sentiment and misinformation analysis. Preprocessing included removing URLs, mentions, special characters, and emojis, expanding contractions, lower casing text, and lemmatization using the NLTK library. Tweets were geotagged to distinguish rural and urban regions. Sentiment analysis was performed using the CardiffNLP/twitter-roberta-base-sentiment model 20 , fine-tuned for Twitter content. Tweets were tokenized using Hugging Face’s AutoTokenizer 21 and truncated to 512 tokens. Sentiments were classified into positive , neutral , or negative categories. For long-form interview transcripts, summaries capped at 500 tokens were generated using GPT-4-Turbo 22 prior to classification with the same model. We used three complementary methods to detect misinformation in tweets: Keyword filtering flagged posts with known false claims; Vector-based retrieval compared tweet embeddings to misinformation examples using FAISS indexing 23,24 ; and semantic classification grouped flagged content into key themes like vaccine safety fears, political misinformation, and alternative medicine claims. 25 Together, these approaches enabled accurate detection across varied misinformation types in vaccine-related discussions. The first RAG framework combined a T5-Base model 16 with dense retrieval, and the second other integrated the LLaMA-2-7B model 26 for greater generative depth. Sentence embeddings of social media content were stored in an FAISS index for retrieval. The top 100 relevant posts were used to condition the generation phase. T5-Base generated concise output efficiently, while LLaMA-2-7B produced more nuanced responses for complex queries. The integration of these two data sources (qualitative social media data and qualitative interview data) enabled triangulation of findings, providing a more comprehensive understanding of the multifaceted drivers of vaccine hesitancy. This mixed-method approach, guided by a novel grounded theory, combines AI-driven analysis with ethnographic data to explore public sentiment and lived experiences, offering a richer perspective on the factors shaping vaccine decisions in Alaska. Fig. 2 illustrates the architecture of our RAG model. Ethics Ethical approval was obtained from the Institutional Review Board at Rutgers University (IRB #Pro2023002010, dated 04.12.2024). Informed consent was obtained from all participants. Interview data were anonymized. Social media data were collected from publicly available Twitter posts per the platform's terms of service, with no personally identifiable information stored. Results Sentiment analysis findings Sentiment classification revealed clear differences in public discourse surrounding COVID-19 vaccination across Alaska’s rural and urban populations. Social media data consisted of approximately 1,300 geotagged tweets related to vaccines. Tweets were categorized into positive, neutral, or negative sentiment using the CardiffNLP/twitter-roberta-base-sentiment model. 20 Table 1 shows negative sentiment dominated in rural areas (55.56%), suggesting distrust-driven negativity, particularly in relation to vaccine mandates and governmental outreach. Rural tweets were more likely to express distrust, fear, or conspiracy-related concerns, often referencing vaccine passports or governmental mandates. In contrast, urban (48.23%) posts reflected a more polarized, politically charged narrative around the same themes. Neutral sentiment comprised 38.89% of rural and 41.79% of urban tweets, often referencing factual or news-related content. Positive sentiment appeared least frequently: 5.56% of rural posts and 8.96% of urban posts. Table 1: Sentiment classification of vaccine-related tweets by urban/rural location Sentiment Urban/Rural Percentage Key Phrases Positive Rural 5.56% ”new”, ”https”, ”vaccine”, ”today”, ”absolute”, ”variant” Urban 8.96% ”https”, ”vaccine mandate”, ”vaccine”, ”booster”, ”pfizer”, ”great” Neutral Rural 38.89% ”passports”, ”https”, ”vaccine”, ”da”, ”im”, ”pfizer” Urban 41.79% ”https”, ”vaccine mandate”,”vaccine”, ”mandate”, ”biden”, ”pfizer” Negative Rural 55.56% ”passports”, ”vaccine passports”, ”https”, ”vaccinated”, ”discriminatory”,”pfizer” Urban 48.23% ”https”, ”vaccinated”,”like”, ”moderna”, ”mandates”, ”vaccine mandates” Interview sentiment, however, revealed more moderate views. As shown in Table 2, neutral sentiment dominated both urban (50%) and rural (48.65%) interviews. Negative sentiment was nearly equal across rural (43.24%) and urban (43.75%) respondents, and rural interviews actually showed slightly higher positivity (8.11% vs. 6.25%). Table 2: Sentiment analysis of interview transcripts by urban/rural classification Sentiment Urban/Rural Percentage Negative Urban 43.75% Rural 43.24% Neutral Urban 50.00% Rural 48.65% Positive Urban 6.25% Rural 8.11% Thematic analysis findings In line with the grounded theory approach, thematic insights emerged from the qualitative data (interviews and social media analysis), revealing complex, multi-layered factors influencing vaccine hesitancy in rural and indigenous communities. These insights were informed by both systemic distrust and personal/cultural beliefs, reflecting the intricate interplay of historical sociopolitical and cultural influences on health decision-making. Systemic distrust and historical trauma One of the core themes that emerged from both interviews and social media analysis was systemic distrust, particularly among indigenous populations. Thematic analysis highlighted how historical events, including colonial exploitation and medical mistreatment, contributed to a deep mistrust of the government and governmental healthcare systems. 27 This resonates with grounded theory’s emphasis on contextual factors and historical understanding when constructing a theory. Thematic patterns in the interviews also align with grounded theory’s premise that data collection and analysis should highlight the structural forces affecting individual and group behaviors. The historical trauma influencing vaccine hesitancy points to a systemic distrust of public health systems, deeply embedded within the collective memory of indigenous communities. Grounded theories' emergent nature reveals this underlying theme without imposing pre-existing categories, showing that historical experiences shape modern-day vaccine decision-making. Cultural identity and collective narratives Another significant theme was the role of cultural identity in vaccine decision-making, which ties into personal and cultural beliefs. Rural participants, particularly those identifying as indigenous, viewed vaccine hesitancy not only as an individual decision but as a community-level issue. This collective perspective is crucial for understanding vaccine hesitancy within indigenous populations, where health decisions are framed in terms of community survival and sovereignty rather than individual autonomy (Table 3). In grounded theory, the researcher’s role is to remain open to emergent categories, such as collective versus individual decision-making. This insight is critical as it challenges the dominant individualistic framework often found in public health campaigns. The collective nature of health decisions in rural and indigenous communities reveals that public health messaging cannot simply focus on the individual’s choice to vaccinate but must consider the community narrative of trust and cultural survival. Misinformation and digital exposure A key theme that emerged, particularly in rural areas, was the impact of misinformation on vaccine attitudes. Social media platforms like Facebook were identified as frequent sources of misinformation. This aligns with grounded theory’s theoretical sampling approach as participants shared how misinformation spread through digital channels and influenced their vaccine beliefs. Thematic analysis grounded in theory suggests that misinformation isn’t just a product of technological platforms but is deeply entwined with cultural distrust of government and public healthcare institutions. This intertwined digital misinformation with systemic distrust creates a vicious cycle where misinformation feeds into already existing distrust, making it more challenging to influence vaccine uptake. Political polarization and government distrust The analysis also revealed that political polarization was a key influence on vaccine hesitancy, with differing views expressed in rural and urban settings. Rural participants tended to view government vaccine mandates as a form of political control, while urban participants tended to view government vaccine issues around personal freedoms. This polarization reinforces the grounded theory idea that political ideologies cannot be separated from public health behaviors. Grounded theory emphasizes how social contexts like political climate shape the emerging patterns from the data. In this study, divergent views on government mandates reflect sociopolitical dynamics that directly impact vaccine hesitancy. The study’s theoretical framework uncovers how political narratives are internalized within rural communities, creating a unique form of systemic distrust tied to the perception of governmental interference. Two primary dimensions that influence vaccine hesitancy emerged from qualitative data: institutional mistrust and political distrust. These themes reflected and expanded on the patterns observed in the sentiment data. Perceived exploitation and historical trauma fueling hesitancy Historical trauma, particularly among Indigenous rural communities, emerged as a prominent barrier to vaccine confidence. Participants referenced Western medical exploitation and colonial history. As a result, communities feel distrust of the federal government and worry that it is after profit, not their community’s well-being. Political polarization was a relevant issue for participants from rural and urban areas alike, although it is expressed in different ways. In urban areas, the discourse was framed around government mandates of vaccines infringing on personal freedom, while rural areas described this as being a way to control the population politically. Misinformation pathways were shaped by digital exposure, and rural interviewees described frequent encounters with misinformation on platforms like Facebook, whereas urban participants demonstrated greater media literacy but still expressed concern over biased reporting. Table 3: Themes related to trust and misinformation in vaccination Main Theme Subtheme Description Example Quote Trust in the healthcare system Historical mistrust in indigenous communities Concerns rooted in historical mistreatment by authorities ” So, Alaska, we’re spread out, difficult to access. And I'm curious if some of it is also a mistrust of government, especially when you’re in remote, primarily Alaska Native villages. I wouldn’t be surprised. There has been a lot of distrust in the federal government due to monetization. Western expansion. I think there’s a lot of pain and trauma there. So, I think building trust in communities to bring in strangers to do these big vaccination clinics, I don’t know if there’s a trust.” Influence of social media Spread of misinformation False claims about the vaccine safety and side effects “I think social media was a little too free with misinformation.” Political beliefs Government distrust Vaccines are seen as political control or an infringement of rights “This I didn’t trust, and I don’t have trust in all the government.” Misinformation impact Vaccine safety concerns/ Media influence False beliefs about vaccines causing severe side effects “What do you want to do? What do they want you to do? It depends on which platform you listen to, whether you have it, and whether you’re foreign. But the mainstream is pushing people to be for it and to get all these vaccines. And now it’s RSV, COVID, flu, whooping cough, and everything else.” Cultural values and safety fears as barriers to vaccination Safety concerns were the most common personal barrier, cutting across rural and urban lines. Many expressed apprehensions about potential side effects, the perceived speed of vaccine development, and uncertainty about vaccine ingredients. One participant explained, “I think that really all the risks from the code vaccine came because of how much they rushed through the process to actually get it released.” Others were skeptical of what goes into the vaccine. One interviewee said, “And not a lot of studies went into this vaccine, and I don't know all the ingredients in the vaccine, but this is just from what I hear, like online.” Natural immunity beliefs were more common among rural participants, with remarks like: “I feel like they could build a good natural immunity to it.” This contrasted with urban participants who, while skeptical, were more likely to acknowledge vaccination benefits. Cultural identity, especially among Indigenous respondents, shaped vaccine narratives as collective experiences of mistrust rather than individual decisions. One participant reflected, “For among native Alaskans, … some who worried well, the government is experimenting on us.” This framing puts vaccination decisions within a broader cultural and historical context. Table 4: Personal and cultural hesitancy themes from interviews Main Theme Subtheme Description Example Quote Personal beliefs and values Natural immunity preference Preference for natural infection over-vaccination “Do you believe that natural immunity acquired through infection is sufficient protection against COVID-19? Sometimes, yes.” Social media influence Viral misinformation The rapid spread of false narratives through online platforms “. . . during COVID, have you ever seen any kind of misinformation or some crazy information about vaccination, or could it be generally? Yeah, we would have. . . . . there were all kinds of crazy things on Facebook, and people were pointing to those going.” Vaccine concerns Safety and side effects Fears about adverse reactions and long-term effects ”So what do you think about vaccines generally . . . . I think they’re unsafe.” Cultural and social norms Indigenous perspectives Historical mistrust due to past medical exploitation “We think that the government may again test a new vaccine. It is on us, so we cannot trust it. I wanted to ask you how you feel about it. UM, yeah, I feel like it’s sort of like. Like we’re Lab Rats being tested, you know.” Table 5 : Common Twitter discourse themes and frequencies Theme Frequency of mentions Common hashtags/ Keywords Example post Misinformation 7.4% #democratsdeliver, #togetherdeclaration, #covid19, https, COVID, people EXCLUSIVE: Bucs receiver Antonio Brown obtained a fake COVID-19 vaccination card to avoid NFL protocols, according to hi... Vaccine advocacy 21.5% #democratsdeliver, #covid19, #omicron, https, covid, vaccinated Guess I won the vax-lottery (AZ+Moderna). But now, which booster should I take? Another Moderna, so I\’m matched up (for ... Fear-based content 9.2% #vaccinemandate, #healthcareworkers, #astrazeneca, https, COVID, people I know more people with serious adverse reactions to the vaccine than COVID-19.... Community support 7.0% #covid19, #togetherdeclaration, #covid 19, https, COVID, help This is strong. Thank you. Oklahoma National Guard & OK Governor. “Until a guardsman is activated under Title 10, they fo... Platform-specific discourse patterns Analysis of Twitter content revealed further distinctions in public sentiment and thematic framing, particularly around misinformation, advocacy, fear, and community narratives. 21.5% of tweets included vaccine advocacy content, often featuring hashtags like #democratsdeliver and references to vaccination success (Table 5). In contrast, 9.2% expressed fear-based content, typically focused on adverse reactions or perceived coercion. Misinformation-themed tweets (7.4%) included politicized content, exaggerated claims, and links to discredited sources. Notably, community-support narratives comprised 7.0% of tweets, including posts thanking National Guard clinics and highlighting local efforts, aligning with rural interview themes of trust through community action. These results indicate that social media narratives are often polarized or exaggerated compared to interview responses. While interviews explored distrust in nuanced ways, Twitter reduced concerns to binary hashtags and viral claims. This supports the study’s use of platform-aware modeling for misinformation detection and community response strategies. Theory of change The multi-layered analysis reveals a theory of change where vaccine hesitancy is shaped by interconnected structural, sociopolitical, and individual-level factors. At its foundation, historical trauma and systemic distrust, particularly in Indigenous and rural communities, play a pivotal role in undermining vaccine confidence. These communities’ collective memories of colonial exploitation and medical mistrust create deep-seated barriers to acceptance, which are further exacerbated by political polarization and misinformation. Social media amplifies these issues, fueling polarized narratives, misinformation, and fear-based content that shape public perception, especially in rural areas. On the individual level, personal fears, cultural beliefs, and a preference for natural immunity influence decisions, particularly among rural participants who frame vaccination as a collective rather than an individual choice. Cultural identity and collective trauma contribute to the perception of vaccines as external threats rather than tools for community health. To address vaccine hesitancy, it is essential to focus on long-term trust-building that acknowledges the historical, cultural, and sociopolitical realities of these communities. This requires not only platform-specific misinformation countermeasures but also culturally relevant messaging that resonates with both historical experiences and personal beliefs. An integrated, context-specific approach that combines AI-driven misinformation detection with ethnographic insights provides a pathway to improve vaccine uptake through targeted interventions that engage the structural, sociopolitical, and individual layers of influence. The thematic insights derived from this study, grounded in the novel theory of systemic distrust and historical trauma, underscore the necessity of moving beyond the traditional public health framework. Public health messaging must be culturally sensitive and historically informed, accounting for the collective nature of health decisions in rural and indigenous communities, as well as the impact of misinformation spread through digital platforms. The grounded theory approach provides a robust methodology for uncovering the underlying drivers of vaccine hesitancy and informs the development of targeted, contextually appropriate interventions. Discussion This study revealed a significant divergence between public discourse on social media and individual perspectives shared in interviews. Social media content, particularly from rural areas, tended to amplify polarized and negative sentiment, while interviews reflected more balanced, reflective, and contextually grounded views. By combining computational tools with qualitative inquiry, this mixed-methods approach provided a nuanced understanding of vaccine hesitancy across urban and rural Alaska. The integration of RAG models, sentiment analysis, and thematic coding enabled the detection of misinformation trends as well as culturally rooted vaccine narratives. These findings highlight critical gaps between online narratives and lived experiences, pointing to the complex interplay of sociocultural, informational, and geographic factors influencing vaccine decision-making. A key finding is the stark contrast between social media and interview sentiment. Twitter showed more polarization, especially in rural areas, while interviews revealed more moderate views, with neutral sentiment prevailing and rural participants slightly more likely to express positive perspectives. This disparity suggests that social media platforms, particularly Twitter, serve as amplifiers of extreme sentiment rather than accurate mirrors of community-level beliefs. While interviews revealed concerns about vaccine safety, side effects, and mistrust, these were often accompanied by thoughtful reflections, conditional acceptance, or pragmatic considerations. 28 Social media, on the other hand, tended to collapse these nuances into viral, often alarmist narratives. This underscores the importance of supplementing social media surveillance with qualitative data to avoid overestimating public hostility and mischaracterizing the complexity of vaccine decision-making. 29 Across both data sources, distrust in government and healthcare systems emerged as a dominant theme, particularly among rural and Indigenous populations. Thematic analysis highlighted that many rural interviewees, especially from Alaska Native communities, framed hesitancy within a broader historical context of colonization and systemic marginalization. References to being “lab rats” or skepticism toward outside health authorities underscore how the collective memory of exploitation and disempowerment continues to shape vaccine perceptions today. This aligns with previous research demonstrating that historical trauma is a persistent determinant of health behavior in Indigenous communities. 30 , 31 However, the present study adds granularity by showing how these legacies interact with real-time political and media environments. While urban participants were also distrustful, their concerns more frequently revolved around political polarization and media bias rather than historical injustice. 32 Thus, tailored public health messaging must not only address current misinformation but also contend with deeper-rooted narratives of marginalization and betrayal, particularly in rural and Indigenous settings. The influence of social media, particularly Facebook and Twitter, emerged as a significant driver of vaccine attitudes. 33 , 34 Rural interviewees frequently describe encountering and sometimes believing misinformation online. These narratives overlapped with tweets that referenced fake vaccine cards, exaggerated side effects, or political conspiracy theories. Twitter content analysis reinforced this divide, showing that misinformation-themed tweets, while only 7.4% of the sample, were often highly politicized and emotionally charged. The prevalence of hashtags like #togetherdeclaration and the conflation of unrelated health threats (e.g., RSV, flu, COVID-19) point to an ecosystem where fear and confusion are actively circulated. By contrast, the more reflective tone in interviews suggests that personal conversations may serve as a corrective to online distortion. Beyond institutional distrust, individual-level concerns, particularly around vaccine safety and natural immunity, were common. Both rural and urban interviewees expressed fears of side effects and distrust in the rapid development of COVID-19 vaccines. However, belief in natural immunity was notably more common among rural participants. Cultural identity also played a key role. Among Indigenous participants, vaccine hesitancy was framed not just as an individual choice but as a community-level issue tied to sovereignty, history, and cultural survival. This suggests that strategies to increase vaccine uptake must avoid individualistic framings and instead engage collective narratives, especially in Indigenous contexts. Public health campaigns that ignore these social dynamics risk reinforcing existing mistrust. These findings carry several important implications for public health planning and vaccine promotion. First, they highlight the risks of over-relying on social media analytics to gauge public opinion. While valuable for detecting trends, social media data often reflect the loudest or most extreme voices, especially in politicized contexts like vaccination [40,41]. Second, the persistence of historical and systemic distrust, particularly among rural and Indigenous communities, demands long-term engagement strategies rooted in cultural humility, respect, and relationship-building. Public health institutions must invest in sustained partnerships that foster local leadership and community ownership of health initiatives. Finally, the study underscores the need for platform-specific misinformation countermeasures. Community-driven media literacy initiatives, coupled with local storytelling and trusted messengers, may offer more effective pathways for addressing misinformation than top-down fact-checking alone. The RAG pipeline proved effective for misinformation detection and tailored response generation. LLaMA-2-7B consistently produced more nuanced, contextually rich outputs than T5-Base, especially for complex queries involving cultural or historical themes. However, it required significantly more computational resources. T5-Base, while faster, occasionally repeated misinformation when exposed to frequently retrieved themes, an issue partially mitigated by post-processing and evidence filtering. These findings align with recent work advocating hybrid NLP systems for misinformation detection. 17 Our integration of dense retrieval via FAISS 23 , sentiment classification using transformer models. 20 Culturally contextualized interviews are also a key component in our model. Beyond computational performance, this study emphasizes the importance of data triangulation. Sole reliance on social media analysis risks overlooking the nuance and lived experiences that interviews provide. By integrating social and cultural narratives with large-scale textual analysis, our approach addresses this gap. By employing grounded theory, this study allows thematic insights to emerge naturally from the data, ensuring that the analysis is context-specific and sensitive to the realities of Alaskan communities. The theory of systemic distrust is not imposed but is constructed based on the patterns revealed through both interviews and social media discourse. The emergent themes (historical trauma, cultural identity, and misinformation) align with grounded theory’s approach to data-driven theory development, offering a comprehensive understanding of vaccine hesitancy that incorporates both local contexts and digital influences. The study has several limitations. The LLaMA-2-7B model, while effective, is resource-intensive and may not support real-time applications in constrained environments. In contrast, T5-Base, though faster, sometimes lacked contextual sensitivity or echoed common misinformation phrases. Additionally, synthesizing multiple tweets occasionally resulted in conflicting inputs, requiring additional filtering to ensure coherence. Sentiment models, though strong on short-form text, may overlook sarcasm or ambiguity in nuanced narratives. 15 , 35 Despite these limitations, the study provides a scalable framework for vaccine misinformation tracking and response. Real-time RAG systems, integrated into public health messaging workflows, could help agencies detect emerging misinformation trends and issue timely, evidence-based responses. Tailoring these systems with culturally informed data, as demonstrated here, further enhances their relevance and equity. Conclusion Vaccine hesitancy in Alaska reflects a dynamic interplay between historical experience, personal belief, and digital information flows. Rural and urban communities differ not only in the content of their concerns but also in how those concerns are expressed and shaped by social context. The thematic insights derived from this study, grounded in the novel theory of systemic distrust and historical trauma, underscore the necessity of moving beyond traditional public health frameworks. Public health messaging must be culturally sensitive and historically informed, accounting for the collective nature of health decisions in rural and indigenous communities, as well as the impact of misinformation spread through digital platforms. The grounded theory approach provides a robust methodology for uncovering the underlying drivers of vaccine hesitancy and informs the development of targeted contextually appropriate interventions. Going forward, efforts to improve vaccine confidence must move beyond combating misinformation and address the deeper structural and cultural roots of distrust. Declarations Ethical approval was obtained from the Institutional Review Board at Rutgers University (IRB #Pro2023002010, dated 04.12.2024). Informed consent was obtained from all participants. Interview data were anonymized. Social media data were collected from publicly available Twitter posts per the platform's terms of service, with no personally identifiable information stored. Acknowledgement Supported in part by a research grant from Investigator-Initiated Studies Program of Merck Sharp & Dohme LLC. The opinions expressed in this paper are those of the authors and do not necessarily represent those of Merck Sharp & Dohme LLC References M. Brown, Vaccine misinformation outpaces efforts to counter it, Columbia Public Health (2024). URL https://www.publichealth.columbia.edu/news/vaccinemisinformation-outpaces-efforts-counter-it. WHO, Vaccine hesitancy: a growing challenge for immunization programs, World Health Organization: Geneve, Switzerland (2015). Robinson R, Nguyen E, Wright M, et al. Factors contributing to vaccine hesitancy and reduced vaccine confidence in rural underserved populations. Humanit Soc Sci Commun 2022; 9 (1): 416. Scharff DP, Mathews KJ, Jackson P, Hoffsuemmer J, Martin E, Edwards D. More than Tuskegee: understanding mistrust about research participation. J Health Care Poor Underserved 2010; 21 (3): 879–97. Tjaden J, Haarmann E, Savaskan N. Experimental evidence on improving COVID-19 vaccine outreach among migrant communities on social media. Sci Rep 2022; 12 (1): 16256. Wilson SL, Wiysonge C. Social media and vaccine hesitancy. BMJ Glob Health 2020; 5 (10). Cinelli M, De Francisci Morales G, Galeazzi A, Quattrociocchi W, Starnini M. The echo chamber effect on social media. Proc Natl Acad Sci U S A 2021; 118 (9). Loomba S, de Figueiredo A, Piatek SJ, de Graaf K, Larson HJ. Measuring the impact of COVID-19 vaccine misinformation on vaccination intent in the UK and USA. Nat Hum Behav 2021; 5 (3): 337–48. Nazli SB, Yigman F, Sevindik M, Deniz Ozturan D. Psychological factors affecting COVID-19 vaccine hesitancy. Ir J Med Sci 2022; 191 (1): 71–80. Lin F, Chen X, Cheng EW. Contextualized impacts of an infodemic on vaccine hesitancy: The moderating role of socioeconomic and cultural factors. Inf Process Manag 2022; 59 (5): 103013. J. Cao, C. M. Ramirez, R. M. Alvarez, The politics of vaccine hesitancy in the united states, Social Science Quarterly 103 (1) (2022) 42–54. Z. Wang, M. M. Rodriguez Morales, K. Husak, M. Kleinman, S. Parthasarathy, In communities we trust institutional failures and sustained solutions for vaccine hesitancy, Tech. rep. (2021). Laurencin CT, McClinton A. The COVID-19 Pandemic: a Call to Action to Identify and Address Racial and Ethnic Disparities. J Racial Ethn Health Disparities 2020; 7 (3): 398–402. V. Sanh, L. Debut, J. Chaumond, T. Wolf, Distilbert base uncased fine tuned on sst-2, https://huggingface.co/distilbert/distilbertbase-uncased-finetuned-sst-2-english, hugging Face (2019). F. Barbieri, J. Camacho-Collados, L. Espinosa Anke, L. Neves, Tweet-Eval: Unified benchmark and comparative evaluation for tweet classification, in: Findings of the Association for Computational Linguistics: EMNLP 2020, Association for Computational Linguistics, Online, 2020, pp. 1644–1650. doi:10.18653/v1/2020.findings-emnlp.148. C. Raffel, N. Shazeer, A. Roberts, K. Lee, S. Narang, M. Matena, Y. Zhou, W. Li, P. J. Liu, Exploring the limits of transfer learning with a unified text-to-text transformer, Journal of Machine Learning Research 21 (140) (2020) 1–67. Li H, Huang J, Ji M, Yang Y, An R. Use of Retrieval-Augmented Large Language Model for COVID-19 Fact-Checking: Development and Usability Study. J Med Internet Res 2025; 27 : e66098. Williams, David R., and Selina A. Mohammed. "Discrimination and racial disparities in health: evidence and needed research." Journal of behavioral medicine 32.1 (2009): 20-47. Kaggle Dataset. Available at [https://www.kaggle.com/datasets/rajugc/kaggle-dataset], last accessed on 08.13.2025. CardiffNLP, Cardiffnlp/twitter-roberta-base-sentiment, accessed: 2025-03-22 (2021). URL https://huggingface.co/cardiffnlp/twitter-robertabase-sentiment. H. Face, Auto classes, accessed: 2025-03-22 (2025). URL https://huggingface.co/docs/transformers/model_doc/auto#transformers.AutoTokenizer. OpenAI, Gpt-4 turbo, accessed: 2025-03-22 (2023). URL https://platform.openai.com/docs/models/gpt-4-turbo. H. J´egou, M. Douze, J. Johnson, Faiss: A library for efficient similarity search, accessed: 2025-03-22 (2017). URL https://engineering.fb.com/2017/03/29/datainfrastructure/faiss-a-library-for-efficient-similaritysearch. N. Reimers, I. Gurevych, Sentence-bert: Sentence embeddings using siamese bert-networks, in: Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing, 2019. URL https://arxiv.org/abs/1908.10084. Pinecone, Embeddings to identify fake newsAccessed: 2025-03-22 (2023). URL https://www.pinecone.io/learn/embeddings-identifyfake-news/. H. Touvron, L. Martin, K. Stone, P. Albert, A. Almahairi, Y. Babaei, N. Bashlykov, S. Batra, P. Bhargava, S. Bhosale, et al., Llama 2: Open foundation and fine-tuned chat models, arXiv preprint arXiv:2307.09288 (2023). Hiratsuka VY, Beans JA, Robinson RF, Shaw JL, Sylvester I, Dillard DA. Self-Determination in Health Research: An Alaska Native Example of Tribal Ownership and Research Regulation. Int J Environ Res Public Health 2017; 14 (11). Ledford CJW, Cafferty LA, Moore JX, et al. The dynamics of trust and communication in COVID-19 vaccine decision making: A qualitative inquiry. J Health Commun 2022; 27 (1): 17–26. Nowak SA, Chen C, Parker AM, Gidengil CA, Matthews LJ. Comparing covariation among vaccine hesitancy and broader beliefs within Twitter and survey data. PLoS One 2020; 15 (10): e0239826. Gone JP, Hartmann WE, Pomerville A, Wendt DC, Klem SH, Burrage RL. The impact of historical trauma on health outcomes for indigenous populations in the USA and Canada: A systematic review. Am Psychol 2019; 74 (1): 20–35. Heart MY, Chase J, Elkins J, Altschul DB. Historical trauma among Indigenous Peoples of the Americas: concepts, research, and clinical considerations. J Psychoactive Drugs 2011; 43 (4): 282–90. Li, Rui, et al. "A Tale of Two Cities: COVID-19 Vaccine Hesitancy as a Result of Racial, Socioeconomic, Digital, and Partisan Divides." ISPRS International Journal of Geo-Information 12.4 (2023): 158. Al-Regaiey KA, Alshamry WS, Alqarni RA, et al. Influence of social media on parents' attitudes towards vaccine administration. Hum Vaccin Immunother 2022; 18 (1): 1872340. Cascini F, Pantovic A, Al-Ajlouni YA, et al. Social media and attitudes towards a COVID-19 vaccination: A systematic review of the literature. EClinicalMedicine 2022; 48 : 101454. K. Roehrick, Valence aware dictionary and sentiment reasoner (vader), R-project. org/package= vader (2020). Additional Declarations The authors declare no competing interests. Cite Share Download PDF Status: Posted Version 1 posted You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {"props":{"pageProps":{"initialData":{"identity":"rs-7368501","acceptedTermsAndConditions":true,"allowDirectSubmit":true,"archivedVersions":[],"articleType":"Research Article","associatedPublications":[],"authors":[{"id":500207233,"identity":"b55dd8ab-e960-4376-b316-13bfc2d08189","order_by":0,"name":"Luay Abdeljaber","email":"","orcid":"","institution":"","correspondingAuthor":false,"prefix":"","firstName":"Luay","middleName":"","lastName":"Abdeljaber","suffix":""},{"id":500207234,"identity":"075427fc-c0bf-4078-8f5d-e33688e5d173","order_by":1,"name":"Sultan Alsarra","email":"","orcid":"","institution":"","correspondingAuthor":false,"prefix":"","firstName":"Sultan","middleName":"","lastName":"Alsarra","suffix":""},{"id":500207235,"identity":"8606ea05-70a0-4bca-aa1c-31f2fd384c33","order_by":2,"name":"Latifur Khan","email":"","orcid":"","institution":"","correspondingAuthor":false,"prefix":"","firstName":"Latifur","middleName":"","lastName":"Khan","suffix":""},{"id":500207236,"identity":"d3411eae-d3e4-406f-987b-2548cdf6385d","order_by":3,"name":"Renee F. Robinson","email":"","orcid":"","institution":"","correspondingAuthor":false,"prefix":"","firstName":"Renee","middleName":"F.","lastName":"Robinson","suffix":""},{"id":500207237,"identity":"dc4a521e-1d76-4af9-9eb3-e1979a722428","order_by":4,"name":"Ana Lorena Ruano","email":"","orcid":"","institution":"","correspondingAuthor":false,"prefix":"","firstName":"Ana","middleName":"Lorena","lastName":"Ruano","suffix":""},{"id":500207238,"identity":"aa45c08f-fe20-4495-8eb4-1bbe646b31a8","order_by":5,"name":"Ubydul Haque","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAA1klEQVRIiWNgGAWjYDACCRAqYDBgkGA+AOLKEKnFgMGAR4ItAcTlIUULjwGIT1iLwe0ew9tAxcb20j2fX92oseBhYD98dANeLXfOGFsDtZjxyJzdZp1zDOgwnrS0G3i13MgxkwZqseGRyN1mnMMG1CLBY0aslpxnxjn/SNBiBtTC/Di3jQgtkneOFVvOMZAw5rmRZsac2yfBw0bIL3y3mzfeeFNhY9g+I/nx55xvdXL87IeP4dXCwMABig4JEIsNQuJXDgLsD2As5g+EVY+CUTAKRsFIBACBtT48AWWfsgAAAABJRU5ErkJggg==","orcid":"","institution":"","correspondingAuthor":true,"prefix":"","firstName":"Ubydul","middleName":"","lastName":"Haque","suffix":""}],"badges":[],"createdAt":"2025-08-13 23:58:44","currentVersionCode":1,"declarations":{"humanSubjects":false,"vertebrateSubjects":false,"conflictsOfInterestStatement":false,"humanSubjectEthicalGuidelines":false,"humanSubjectConsent":false,"humanSubjectClinicalTrial":false,"humanSubjectCaseReport":false,"vertebrateSubjectEthicalGuidelines":false},"doi":"10.21203/rs.3.rs-7368501/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-7368501/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":89362855,"identity":"89184148-d79c-46fe-a9f8-b9c44ffc417e","added_by":"auto","created_at":"2025-08-19 08:39:15","extension":"png","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":81256,"visible":true,"origin":"","legend":"\u003cp\u003eFactors contributing to vaccine hesitancy, categorized into socio-cultural, political, social media, and misinformation influences.\u003c/p\u003e","description":"","filename":"1.png","url":"https://assets-eu.researchsquare.com/files/rs-7368501/v1/96a5a2f30c2270afbebe8a38.png"},{"id":89363198,"identity":"e9085a3f-a296-4221-8ffa-6d746e8554c5","added_by":"auto","created_at":"2025-08-19 08:47:15","extension":"png","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":64446,"visible":true,"origin":"","legend":"\u003cp\u003eCompact architecture of the RAG system for vaccine misinformation analysis with optimized spacing.\u003c/p\u003e","description":"","filename":"2.png","url":"https://assets-eu.researchsquare.com/files/rs-7368501/v1/d8753f3a207582fc184f7f9b.png"},{"id":89364177,"identity":"85e94520-aef3-4476-ad39-1bba57600358","added_by":"auto","created_at":"2025-08-19 08:55:15","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":915078,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-7368501/v1/9725088e-833f-4fb4-9727-d650c58e999e.pdf"}],"financialInterests":"The authors declare no competing interests.","formattedTitle":"\u003cp\u003eIntegrating Retrieval-Augmented Generation and Thematic NLP for Vaccine Confidence Modeling in Alaska\u003c/p\u003e","fulltext":[{"header":"Highlights","content":"\u003cul\u003e\n \u003cli\u003eRetrieval-Augmented Generation system detects vaccine misinformation using real-world social media data.\u003c/li\u003e\n \u003cli\u003eCombined Llama-2 and T5 models for balanced accuracy and response speed.\u003c/li\u003e\n \u003cli\u003eSentiment analysis revealed distrust-driven negativity, especially in rural areas.\u003c/li\u003e\n \u003cli\u003eInterview sentiment diverged from social media and showed more neutral attitudes.\u003c/li\u003e\n \u003cli\u003eTriangulating NLP and ethnographic data improves public health insights. \u003c/li\u003e\n\u003c/ul\u003e"},{"header":"Introduction","content":"\u003cp\u003eVaccine hesitancy continues to pose a critical public health challenge worldwide\u003csup\u003e\u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e1\u003c/span\u003e,\u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2\u003c/span\u003e\u003c/sup\u003e, with rural and underserved populations like those in Alaska.\u003csup\u003e\u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e3\u003c/span\u003e\u003c/sup\u003e Here, geographic isolation, historical trauma, particularly among Indigenous populations, and distinct sociopolitical attitudes toward government and health autonomy shape divergent perceptions of vaccination. In addition, these communities often face intersecting barriers to vaccine uptake, including limited healthcare access, socio-cultural resistance, and deeply rooted institutional mistrust.\u003csup\u003e\u003cspan citationid=\"CR4\" class=\"CitationRef\"\u003e4\u003c/span\u003e\u003c/sup\u003e This has been compounded by the COVID-19 pandemic\u003csup\u003e\u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e5\u003c/span\u003e\u003c/sup\u003e, which brought a proliferation of misinformation through social media platforms that outpaced public health responses.\u003csup\u003e\u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e6\u003c/span\u003e\u003c/sup\u003e Social media algorithms are designed to maximize user interaction by amplifying emotionally charged and polarizing content, creating feedback loops that reward controversy over accuracy.\u003csup\u003e\u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e7\u003c/span\u003e\u003c/sup\u003e In this environment, vaccine myths, particularly those related to safety, efficacy, and side effects, can rapidly go viral, entrenching skepticism and distrust.\u003csup\u003e\u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e8\u003c/span\u003e\u003c/sup\u003e\u003c/p\u003e\u003cp\u003eA wide range of factors contribute to vaccine hesitancy, including psychological concerns\u003csup\u003e\u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e9\u003c/span\u003e\u003c/sup\u003e, cultural and religious beliefs\u003csup\u003e\u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e10\u003c/span\u003e\u003c/sup\u003e, political ideologies\u003csup\u003e\u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e11\u003c/span\u003e\u003c/sup\u003e, and longstanding mistrust of health institutions.\u003csup\u003e\u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e12\u003c/span\u003e\u003c/sup\u003e In the United States (U.S.), these drivers are magnified in historically marginalized communities, where public health campaigns must contend with legacies of systemic discrimination and medical exploitation.\u003csup\u003e\u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e13\u003c/span\u003e\u003c/sup\u003e This multifaceted landscape calls for context-specific research that accounts for the structural, cultural, and informational dynamics at play.\u003c/p\u003e\u003cp\u003eArtificial intelligence (AI) offers powerful tools to investigate and address vaccine hesitancy at scale. Natural language processing (NLP) models, in particular, have been used to analyze health-related discourse, detect misinformation, and assess public sentiment across large datasets. While lexicon-based tools like VADER offer speed and simplicity, transformer-based models, such as DistilBERT and CardiffNLP\u0026rsquo;s RoBERTa, enable more accurate sentiment classification in informal, short-form text like tweets.\u003csup\u003e\u003cspan additionalcitationids=\"CR14\" citationid=\"CR13\" class=\"CitationRef\"\u003e13\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e15\u003c/span\u003e\u003c/sup\u003e Retrieval-Augmented Generation (RAG) models represent a recent advancement in NLP, combining document retrieval with generative modeling to produce context-aware responses grounded in external evidence.\u003csup\u003e\u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e15\u003c/span\u003e,\u003cspan citationid=\"CR16\" class=\"CitationRef\"\u003e16\u003c/span\u003e\u003c/sup\u003e These models have demonstrated effectiveness in applications such as fact-checking and misinformation detection, though their use in localized public health contexts remains limited.\u003csup\u003e\u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e17\u003c/span\u003e\u003c/sup\u003e\u003c/p\u003e\u003cp\u003eThis study applies a novel, AI-augmented framework to investigate vaccine hesitancy in Alaska. We combine sentiment analysis of geotagged Twitter data with qualitative thematic analysis of in-depth interviews, offering both macro-level insights into public discourse and micro-level understanding of individual beliefs. Our approach incorporates a dual-model RAG system, using T5-Base for efficient, factual outputs and LLaMA-2-7B for nuanced, context-rich generation, alongside FAISS vector indexing and transformer-based sentiment classification. This study aims to understand vaccine hesitancy within Alaska\u0026rsquo;s diverse communities by triangulating public sentiment (via social media) and individual beliefs (through qualitative interviews, combining technological insights with ethnographic data to create a multi-layered approach to vaccine misinformation detection. The analysis is framed within a novel Grounded Theory that integrates historical trauma, systemic distrust, and cultural beliefs as foundational drivers of vaccine hesitance. This theory was specifically suggested to address the unique sociopolitical and cultural dynamics present in Alaska, particularly among indigenous and rural populations. By emphasizing the local context and community-driven narratives, the study aims to offer a more comprehensive understanding of vaccine hesitancy that builds upon relevant public health models such as the Theory of Change, offering targeted insights for culturally informed public health messaging.\u003c/p\u003e\u003cp\u003e\u003c/p\u003e"},{"header":"Methods","content":"\u003cp\u003e\u003cstrong\u003eSampling and inclusion criteria\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eParticipants were adult residents of Alaska (18+) with varying levels of vaccine hesitancy. Purposive sampling ensured representation across gender, race, geography (urban vs. rural), and hesitancy levels (hesitant, undecided, vaccinated). The sample included diverse ethnic groups, such as Native Alaskan, White, African American, Hispanic, and Asian residents. The novel grounded theory framework identifies systemic distrust as a central factor influencing vaccine hesitancy, particularly among indigenous communities, rooted in historical mistrust of government and healthcare institutions.\u003csup\u003e18\u003c/sup\u003e A total of 87 semi-structured interviews were conducted via Zoom between July and November 2024, each lasting 30\u0026ndash;40 minutes.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eData collection\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eSocial media data were sourced from a publicly available Kaggle dataset\u003csup\u003e19\u003c/sup\u003e and filtered for tweets geotagged within Alaska. Posts included were chosen based on keywords explicitly referencing vaccine-related topics, hesitancy, safety concerns, or COVID-19-related terms.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eData analysis\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eInterview transcripts were coded thematically to identify patterns in hesitancy and information sources across rural and urban populations.\u003cstrong\u003e\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eApproximately 1,300 tweets were preprocessed for sentiment and misinformation analysis. Preprocessing included removing URLs, mentions, special characters, and emojis, expanding contractions, lower casing text, and lemmatization using the NLTK library. Tweets were geotagged to distinguish rural and urban regions.\u003cstrong\u003e\u0026nbsp;\u003c/strong\u003eSentiment analysis was performed using the CardiffNLP/twitter-roberta-base-sentiment model\u003csup\u003e20\u003c/sup\u003e, fine-tuned for Twitter content. Tweets were tokenized using Hugging Face\u0026rsquo;s AutoTokenizer\u003csup\u003e21\u003c/sup\u003e and truncated to 512 tokens. Sentiments were classified into \u003cem\u003epositive\u003c/em\u003e, \u003cem\u003eneutral\u003c/em\u003e, or \u003cem\u003enegative\u003c/em\u003e categories. For long-form interview transcripts, summaries capped at 500 tokens were generated using GPT-4-Turbo\u003csup\u003e22\u003c/sup\u003e prior to classification with the same model.\u003cstrong\u003e\u0026nbsp;\u003c/strong\u003eWe used three complementary methods to detect misinformation in tweets: Keyword filtering flagged posts with known false claims; Vector-based retrieval compared tweet embeddings to misinformation examples using FAISS indexing\u003csup\u003e23,24\u003c/sup\u003e; and semantic classification grouped flagged content into key themes like vaccine safety fears, political misinformation, and alternative medicine claims.\u003csup\u003e25\u003c/sup\u003e Together, these approaches enabled accurate detection across varied misinformation types in vaccine-related discussions. The first RAG framework combined a T5-Base model\u003csup\u003e16\u003c/sup\u003e with dense retrieval, and the second other integrated the LLaMA-2-7B model\u003csup\u003e26\u003c/sup\u003e for greater generative depth. Sentence embeddings of social media content were stored in an FAISS index for retrieval. The top 100 relevant posts were used to condition the generation phase. T5-Base generated concise output efficiently, while LLaMA-2-7B produced more nuanced responses for complex queries. The integration of these two data sources (qualitative social media data and qualitative interview data) enabled triangulation of findings, providing a more comprehensive understanding of the multifaceted drivers of vaccine hesitancy. This mixed-method approach, guided by a novel grounded theory, combines AI-driven analysis with ethnographic data to explore public sentiment and lived experiences, offering a richer perspective on the factors shaping vaccine decisions in Alaska. Fig. 2 illustrates the architecture of our RAG model.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eEthics\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eEthical approval was obtained from the Institutional Review Board at Rutgers University (IRB #Pro2023002010, dated 04.12.2024). Informed consent was obtained from all participants. Interview data were anonymized. Social media data were collected from publicly available Twitter posts per the platform\u0026apos;s terms of service, with no personally identifiable information stored.\u003c/p\u003e"},{"header":"Results","content":"\u003cp\u003e\u003cstrong\u003e\u003cem\u003eSentiment analysis findings\u003c/em\u003e\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eSentiment classification revealed clear differences in public discourse surrounding COVID-19 vaccination across Alaska\u0026rsquo;s rural and urban populations. Social media data consisted of approximately 1,300 geotagged tweets related to vaccines. Tweets were categorized into positive, neutral, or negative sentiment using the CardiffNLP/twitter-roberta-base-sentiment model.\u003csup\u003e20\u003c/sup\u003e Table 1 shows negative sentiment dominated in rural areas (55.56%), suggesting distrust-driven negativity, particularly in relation to vaccine mandates and governmental outreach. Rural tweets were more likely to express distrust, fear, or conspiracy-related concerns, often referencing vaccine passports or governmental mandates. In contrast, urban (48.23%) posts reflected a more polarized, politically charged narrative around the same themes. Neutral sentiment comprised 38.89% of rural and 41.79% of urban tweets, often referencing factual or news-related content. Positive sentiment appeared least frequently: 5.56% of rural posts and 8.96% of urban posts.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eTable 1:\u003c/strong\u003e Sentiment classification of vaccine-related tweets by urban/rural location\u003c/p\u003e\n\u003ctable border=\"1\" cellspacing=\"0\" cellpadding=\"0\"\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 12.5%;\"\u003e\n \u003cp\u003e\u003cstrong\u003eSentiment\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 13.4615%;\"\u003e\n \u003cp\u003e\u003cstrong\u003eUrban/Rural\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 11.5385%;\"\u003e\n \u003cp\u003e\u003cstrong\u003ePercentage\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 62.5%;\"\u003e\n \u003cp\u003e\u003cstrong\u003eKey Phrases\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 12.5%;\"\u003e\n \u003cp\u003e\u003cem\u003ePositive\u003c/em\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 13.4615%;\"\u003e\n \u003cp\u003e\u003cem\u003eRural\u003c/em\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 11.5385%;\"\u003e\n \u003cp\u003e\u003cem\u003e5.56%\u003c/em\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 62.5%;\"\u003e\n \u003cp\u003e\u003cem\u003e\u0026rdquo;new\u0026rdquo;, \u0026rdquo;https\u0026rdquo;, \u0026rdquo;vaccine\u0026rdquo;, \u0026rdquo;today\u0026rdquo;, \u0026rdquo;absolute\u0026rdquo;, \u0026rdquo;variant\u0026rdquo;\u003c/em\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 12.5%;\"\u003e\n \u003cp\u003e\u003cem\u003e\u0026nbsp;\u003c/em\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 13.4615%;\"\u003e\n \u003cp\u003e\u003cem\u003eUrban\u003c/em\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 11.5385%;\"\u003e\n \u003cp\u003e\u003cem\u003e8.96%\u003c/em\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 62.5%;\"\u003e\n \u003cp\u003e\u0026rdquo;https\u0026rdquo;, \u0026rdquo;vaccine mandate\u0026rdquo;, \u0026rdquo;vaccine\u0026rdquo;, \u0026rdquo;booster\u0026rdquo;, \u0026rdquo;pfizer\u0026rdquo;, \u0026rdquo;great\u0026rdquo;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 12.5%;\"\u003e\n \u003cp\u003e\u003cem\u003eNeutral\u003c/em\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 13.4615%;\"\u003e\n \u003cp\u003e\u003cem\u003eRural\u003c/em\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 11.5385%;\"\u003e\n \u003cp\u003e\u003cem\u003e38.89%\u003c/em\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 62.5%;\"\u003e\n \u003cp\u003e\u003cem\u003e\u0026rdquo;passports\u0026rdquo;, \u0026rdquo;https\u0026rdquo;, \u0026rdquo;vaccine\u0026rdquo;, \u0026rdquo;da\u0026rdquo;, \u0026rdquo;im\u0026rdquo;, \u0026rdquo;pfizer\u0026rdquo;\u003c/em\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 12.5%;\"\u003e\n \u003cp\u003e\u003cem\u003e\u0026nbsp;\u003c/em\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 13.4615%;\"\u003e\n \u003cp\u003e\u003cem\u003eUrban\u003c/em\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 11.5385%;\"\u003e\n \u003cp\u003e\u003cem\u003e41.79%\u003c/em\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 62.5%;\"\u003e\n \u003cp\u003e\u003cem\u003e\u0026rdquo;https\u0026rdquo;, \u0026rdquo;vaccine mandate\u0026rdquo;,\u0026rdquo;vaccine\u0026rdquo;, \u0026rdquo;mandate\u0026rdquo;, \u0026rdquo;biden\u0026rdquo;, \u0026rdquo;pfizer\u0026rdquo;\u003c/em\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 12.5%;\"\u003e\n \u003cp\u003e\u003cem\u003eNegative\u003c/em\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 13.4615%;\"\u003e\n \u003cp\u003e\u003cem\u003eRural\u003c/em\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 11.5385%;\"\u003e\n \u003cp\u003e\u003cem\u003e55.56%\u003c/em\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 62.5%;\"\u003e\n \u003cp\u003e\u003cem\u003e\u0026rdquo;passports\u0026rdquo;, \u0026rdquo;vaccine passports\u0026rdquo;, \u0026rdquo;https\u0026rdquo;, \u0026rdquo;vaccinated\u0026rdquo;, \u0026rdquo;discriminatory\u0026rdquo;,\u0026rdquo;pfizer\u0026rdquo;\u003c/em\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 12.5%;\"\u003e\n \u003cp\u003e\u003cem\u003e\u0026nbsp;\u003c/em\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 13.4615%;\"\u003e\n \u003cp\u003e\u003cem\u003eUrban\u003c/em\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 11.5385%;\"\u003e\n \u003cp\u003e\u003cem\u003e48.23%\u003c/em\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 62.5%;\"\u003e\n \u003cp\u003e\u003cem\u003e\u0026rdquo;https\u0026rdquo;, \u0026rdquo;vaccinated\u0026rdquo;,\u0026rdquo;like\u0026rdquo;, \u0026rdquo;moderna\u0026rdquo;, \u0026rdquo;mandates\u0026rdquo;, \u0026rdquo;vaccine mandates\u0026rdquo;\u003c/em\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n\u003c/table\u003e\n\u003cp\u003eInterview sentiment, however, revealed more moderate views. As shown in Table 2, neutral sentiment dominated both urban (50%) and rural (48.65%) interviews. Negative sentiment was nearly equal across rural (43.24%) and urban (43.75%) respondents, and rural interviews actually showed slightly higher positivity (8.11% vs. 6.25%).\u0026nbsp;\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eTable 2:\u003c/strong\u003e Sentiment analysis of interview transcripts by urban/rural classification\u003c/p\u003e\n\u003ctable border=\"1\" cellspacing=\"0\" cellpadding=\"0\"\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 208px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eSentiment\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 208px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eUrban/Rural\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 208px;\"\u003e\n \u003cp\u003e\u003cstrong\u003ePercentage\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd rowspan=\"2\" valign=\"top\" style=\"width: 208px;\"\u003e\n \u003cp\u003eNegative\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 208px;\"\u003e\n \u003cp\u003eUrban\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 208px;\"\u003e\n \u003cp\u003e43.75%\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 208px;\"\u003e\n \u003cp\u003eRural\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 208px;\"\u003e\n \u003cp\u003e43.24%\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd rowspan=\"2\" valign=\"top\" style=\"width: 208px;\"\u003e\n \u003cp\u003eNeutral\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 208px;\"\u003e\n \u003cp\u003eUrban\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 208px;\"\u003e\n \u003cp\u003e50.00%\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 208px;\"\u003e\n \u003cp\u003eRural\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 208px;\"\u003e\n \u003cp\u003e48.65%\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd rowspan=\"2\" valign=\"top\" style=\"width: 208px;\"\u003e\n \u003cp\u003ePositive\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 208px;\"\u003e\n \u003cp\u003eUrban\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 208px;\"\u003e\n \u003cp\u003e6.25%\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 208px;\"\u003e\n \u003cp\u003eRural\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 208px;\"\u003e\n \u003cp\u003e8.11%\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n\u003c/table\u003e\n\u003cp\u003e\u003cstrong\u003e\u003cem\u003eThematic analysis findings\u003c/em\u003e\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eIn line with the grounded theory approach, thematic insights emerged from the qualitative data (interviews and social media analysis), revealing complex, multi-layered factors influencing vaccine hesitancy in rural and indigenous communities. These insights were informed by both systemic distrust and personal/cultural beliefs, reflecting the intricate interplay of historical sociopolitical and cultural influences on health decision-making.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e\u003cem\u003eSystemic distrust and historical trauma\u003c/em\u003e\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eOne of the core themes that emerged from both interviews and social media analysis was systemic distrust, particularly among indigenous populations. Thematic analysis highlighted how historical events, including colonial exploitation and medical mistreatment, contributed to a deep mistrust of the government and governmental healthcare systems.\u003csup\u003e27\u003c/sup\u003e This resonates with grounded theory\u0026rsquo;s emphasis on contextual factors and historical understanding when constructing a theory.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eThematic patterns in the interviews also align with grounded theory\u0026rsquo;s premise that data collection and analysis should highlight the structural forces affecting individual and group behaviors. The historical trauma influencing vaccine hesitancy points to a systemic distrust of public health systems, deeply embedded within the collective memory of indigenous communities. Grounded theories\u0026apos; emergent nature reveals this underlying theme without imposing pre-existing categories, showing that historical experiences shape modern-day vaccine decision-making.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e\u003cem\u003eCultural identity and collective narratives\u003c/em\u003e\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eAnother significant theme was the role of cultural identity in vaccine decision-making, which ties into personal and cultural beliefs. Rural participants, particularly those identifying as indigenous, viewed vaccine hesitancy not only as an individual decision but as a community-level issue. This collective perspective is crucial for understanding vaccine hesitancy within indigenous populations, where health decisions are framed in terms of community survival and sovereignty rather than individual autonomy (Table 3).\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eIn grounded theory, the researcher\u0026rsquo;s role is to remain open to emergent categories, such as collective versus individual decision-making. This insight is critical as it challenges the dominant individualistic framework often found in public health campaigns. The collective nature of health decisions in rural and indigenous communities reveals that public health messaging cannot simply focus on the individual\u0026rsquo;s choice to vaccinate but must consider the community narrative of trust and cultural survival.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e\u003cem\u003eMisinformation and digital exposure\u003c/em\u003e\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eA key theme that emerged, particularly in rural areas, was the impact of misinformation on vaccine attitudes. Social media platforms like Facebook were identified as frequent sources of misinformation. This aligns with grounded theory\u0026rsquo;s theoretical sampling approach as participants shared how misinformation spread through digital channels and influenced their vaccine beliefs.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eThematic analysis grounded in theory suggests that misinformation isn\u0026rsquo;t just a product of technological platforms but is deeply entwined with cultural distrust of government and public healthcare institutions. This intertwined digital misinformation with systemic distrust creates a vicious cycle where misinformation feeds into already existing distrust, making it more challenging to influence vaccine uptake.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e\u003cem\u003ePolitical polarization and government distrust\u003c/em\u003e\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe analysis also revealed that political polarization was a key influence on vaccine hesitancy, with differing views expressed in rural and urban settings. Rural participants tended to view government vaccine mandates as a form of political control, while urban participants tended to view government vaccine issues around personal freedoms. This polarization reinforces the grounded theory idea that political ideologies cannot be separated from public health behaviors.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eGrounded theory emphasizes how social contexts like political climate shape the emerging patterns from the data. In this study, divergent views on government mandates reflect sociopolitical dynamics that directly impact vaccine hesitancy. The study\u0026rsquo;s theoretical framework uncovers how political narratives are internalized within rural communities, creating a unique form of systemic distrust tied to the perception of governmental interference.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eTwo primary dimensions that influence vaccine hesitancy emerged from qualitative data: institutional mistrust and political distrust. These themes reflected and expanded on the patterns observed in the sentiment data.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e\u003cem\u003ePerceived exploitation and historical trauma fueling hesitancy\u0026nbsp;\u003c/em\u003e\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eHistorical trauma, particularly among Indigenous rural communities, emerged as a prominent barrier to vaccine confidence. Participants referenced Western medical exploitation and colonial history. As a result, communities feel distrust of the federal government and worry that it is after profit, not their community\u0026rsquo;s well-being. Political polarization was a relevant issue for participants from rural and urban areas alike, although it is expressed in different ways. In urban areas, the discourse was framed around government mandates of vaccines infringing on personal freedom, while rural areas described this as being a way to control the population politically.\u003c/p\u003e\n\u003cp\u003eMisinformation pathways were shaped by digital exposure, and rural interviewees described frequent encounters with misinformation on platforms like Facebook, whereas urban participants demonstrated greater media literacy but still expressed concern over biased reporting.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eTable 3:\u0026nbsp;\u003c/strong\u003eThemes related to trust and misinformation in vaccination\u003c/p\u003e\n\u003ctable border=\"1\" cellspacing=\"0\" cellpadding=\"0\"\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 96px;\"\u003e\n \u003cp\u003eMain Theme\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 108px;\"\u003e\n \u003cp\u003eSubtheme\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 156px;\"\u003e\n \u003cp\u003eDescription\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 264px;\"\u003e\n \u003cp\u003eExample Quote\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 96px;\"\u003e\n \u003cp\u003eTrust in the\u003c/p\u003e\n \u003cp\u003ehealthcare\u003c/p\u003e\n \u003cp\u003esystem\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 108px;\"\u003e\n \u003cp\u003eHistorical mistrust\u003c/p\u003e\n \u003cp\u003ein indigenous\u003c/p\u003e\n \u003cp\u003ecommunities\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 156px;\"\u003e\n \u003cp\u003eConcerns rooted in historical\u003c/p\u003e\n \u003cp\u003emistreatment by\u003c/p\u003e\n \u003cp\u003eauthorities\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 264px;\"\u003e\n \u003cp\u003e\u0026rdquo; So, Alaska, we\u0026rsquo;re spread out, difficult to access. And I\u0026apos;m curious if some of it is also a mistrust of government, especially when you\u0026rsquo;re in remote, primarily Alaska Native villages. I wouldn\u0026rsquo;t be surprised. There has been a lot of distrust in the federal government due to monetization. Western expansion. I think there\u0026rsquo;s a lot of pain and trauma there. So, I think building trust in communities to bring in strangers to do these big vaccination clinics, I don\u0026rsquo;t know if there\u0026rsquo;s a trust.\u0026rdquo;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 96px;\"\u003e\n \u003cp\u003eInfluence\u003c/p\u003e\n \u003cp\u003eof social\u003c/p\u003e\n \u003cp\u003emedia\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 108px;\"\u003e\n \u003cp\u003eSpread of misinformation\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 156px;\"\u003e\n \u003cp\u003eFalse claims about the vaccine safety and side effects\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 264px;\"\u003e\n \u003cp\u003e\u0026ldquo;I think social media was a little\u003c/p\u003e\n \u003cp\u003etoo free with misinformation.\u0026rdquo;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 96px;\"\u003e\n \u003cp\u003ePolitical\u003c/p\u003e\n \u003cp\u003ebeliefs\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 108px;\"\u003e\n \u003cp\u003eGovernment\u003c/p\u003e\n \u003cp\u003edistrust\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 156px;\"\u003e\n \u003cp\u003eVaccines are seen as political\u003c/p\u003e\n \u003cp\u003econtrol or an infringement\u003c/p\u003e\n \u003cp\u003eof rights\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 264px;\"\u003e\n \u003cp\u003e\u0026ldquo;This I didn\u0026rsquo;t trust, and I don\u0026rsquo;t have trust in all the government.\u0026rdquo;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 96px;\"\u003e\n \u003cp\u003eMisinformation\u003c/p\u003e\n \u003cp\u003eimpact\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 108px;\"\u003e\n \u003cp\u003eVaccine safety\u003c/p\u003e\n \u003cp\u003econcerns/ Media\u003c/p\u003e\n \u003cp\u003einfluence\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 156px;\"\u003e\n \u003cp\u003eFalse beliefs about vaccines\u003c/p\u003e\n \u003cp\u003ecausing severe side\u003c/p\u003e\n \u003cp\u003eeffects\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 264px;\"\u003e\n \u003cp\u003e\u0026ldquo;What do you want to do? What do they want you to do? It depends on which platform you listen to, whether you have it, and whether you\u0026rsquo;re foreign. But the mainstream is pushing people to be for it and to get all these vaccines. And now it\u0026rsquo;s RSV, COVID, flu, whooping cough, and everything else.\u0026rdquo;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n\u003c/table\u003e\n\u003cp\u003e\u003cbr\u003e\u0026nbsp;\u003cstrong\u003e\u003cem\u003eCultural values and safety fears as barriers to vaccination\u0026nbsp;\u003c/em\u003e\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eSafety concerns were the most common personal barrier, cutting across rural and urban lines. Many expressed apprehensions about potential side effects, the perceived speed of vaccine development, and uncertainty about vaccine ingredients. One participant explained, \u0026ldquo;I think that really all the risks from the code vaccine came because of how much they rushed through the process to actually get it released.\u0026rdquo; Others were skeptical of what goes into the vaccine. One interviewee said, \u0026ldquo;And not a lot of studies went into this vaccine, and I don\u0026apos;t know all the ingredients in the vaccine, but this is just from what I hear, like online.\u0026rdquo;\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eNatural immunity beliefs were more common among rural participants, with remarks like: \u0026ldquo;I feel like they could build a good natural immunity to it.\u0026rdquo; This contrasted with urban participants who, while skeptical, were more likely to acknowledge vaccination benefits.\u003c/p\u003e\n\u003cp\u003eCultural identity, especially among Indigenous respondents, shaped vaccine narratives as collective experiences of mistrust rather than individual decisions. One participant reflected, \u0026ldquo;For among native Alaskans, \u0026hellip; some who worried well, the government is experimenting on us.\u0026rdquo; This framing puts vaccination decisions within a broader cultural and historical context.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eTable 4:\u003c/strong\u003e Personal and cultural hesitancy themes from interviews\u003c/p\u003e\n\u003ctable border=\"1\" cellspacing=\"0\" cellpadding=\"0\"\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 96px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eMain Theme\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 96px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eSubtheme\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 174px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eDescription\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 258px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eExample Quote\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 96px;\"\u003e\n \u003cp\u003ePersonal\u003c/p\u003e\n \u003cp\u003ebeliefs and\u003c/p\u003e\n \u003cp\u003evalues\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 96px;\"\u003e\n \u003cp\u003eNatural immunity\u003c/p\u003e\n \u003cp\u003epreference\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 174px;\"\u003e\n \u003cp\u003ePreference for natural\u003c/p\u003e\n \u003cp\u003einfection over-vaccination\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 258px;\"\u003e\n \u003cp\u003e\u0026ldquo;Do you believe that natural immunity acquired through infection is sufficient protection against COVID-19? Sometimes, yes.\u0026rdquo;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 96px;\"\u003e\n \u003cp\u003eSocial\u003c/p\u003e\n \u003cp\u003emedia influence\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 96px;\"\u003e\n \u003cp\u003eViral misinformation\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 174px;\"\u003e\n \u003cp\u003eThe rapid spread of false\u003c/p\u003e\n \u003cp\u003enarratives through online\u003c/p\u003e\n \u003cp\u003eplatforms\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 258px;\"\u003e\n \u003cp\u003e\u0026ldquo;. . . during COVID, have you ever seen any kind of misinformation or some crazy information about vaccination, or could it be generally? Yeah, we would have. . . . . there were all kinds of crazy things on Facebook, and people were pointing to those going.\u0026rdquo;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 96px;\"\u003e\n \u003cp\u003eVaccine\u003c/p\u003e\n \u003cp\u003econcerns\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 96px;\"\u003e\n \u003cp\u003eSafety and\u003c/p\u003e\n \u003cp\u003eside effects\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 174px;\"\u003e\n \u003cp\u003eFears about adverse reactions\u003c/p\u003e\n \u003cp\u003eand long-term effects\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 258px;\"\u003e\n \u003cp\u003e\u0026rdquo;So what do you think about vaccines generally . . . . I think they\u0026rsquo;re unsafe.\u0026rdquo;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 96px;\"\u003e\n \u003cp\u003eCultural\u003c/p\u003e\n \u003cp\u003eand social\u003c/p\u003e\n \u003cp\u003enorms\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 96px;\"\u003e\n \u003cp\u003eIndigenous\u003c/p\u003e\n \u003cp\u003eperspectives\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 174px;\"\u003e\n \u003cp\u003eHistorical mistrust due\u003c/p\u003e\n \u003cp\u003eto past medical exploitation\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 258px;\"\u003e\n \u003cp\u003e\u0026ldquo;We think that the government may again test a new vaccine. It is on us, so we cannot trust it. I wanted to ask you how you feel about it. UM, yeah, I feel like it\u0026rsquo;s sort of like. Like we\u0026rsquo;re Lab Rats being tested, you know.\u0026rdquo;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n\u003c/table\u003e\n\u003cp\u003e\u003cstrong\u003eTable 5\u003c/strong\u003e: Common Twitter discourse themes and frequencies\u003c/p\u003e\n\u003ctable border=\"1\" cellspacing=\"0\" cellpadding=\"0\"\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 114px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eTheme\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 84px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eFrequency of\u003c/strong\u003e\u003c/p\u003e\n \u003cp\u003e\u003cstrong\u003ementions\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 150px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eCommon hashtags/\u003c/strong\u003e\u003c/p\u003e\n \u003cp\u003e\u003cstrong\u003eKeywords\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 276px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eExample post\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 114px;\"\u003e\n \u003cp\u003eMisinformation\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 84px;\"\u003e\n \u003cp\u003e7.4%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 150px;\"\u003e\n \u003cp\u003e#democratsdeliver,\u003c/p\u003e\n \u003cp\u003e#togetherdeclaration,\u003c/p\u003e\n \u003cp\u003e#covid19, https,\u003c/p\u003e\n \u003cp\u003eCOVID, people\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 276px;\"\u003e\n \u003cp\u003eEXCLUSIVE: Bucs receiver Antonio Brown obtained a fake COVID-19 vaccination card to avoid NFL protocols, according to hi...\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 114px;\"\u003e\n \u003cp\u003eVaccine advocacy\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 84px;\"\u003e\n \u003cp\u003e21.5%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 150px;\"\u003e\n \u003cp\u003e#democratsdeliver,\u003c/p\u003e\n \u003cp\u003e#covid19, #omicron,\u003c/p\u003e\n \u003cp\u003ehttps, covid, vaccinated\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 276px;\"\u003e\n \u003cp\u003eGuess I won the vax-lottery (AZ+Moderna). But now, which booster should I take? Another Moderna, so I\\\u0026rsquo;m matched up (for ...\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 114px;\"\u003e\n \u003cp\u003eFear-based content\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 84px;\"\u003e\n \u003cp\u003e9.2%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 150px;\"\u003e\n \u003cp\u003e#vaccinemandate,\u003c/p\u003e\n \u003cp\u003e#healthcareworkers,\u003c/p\u003e\n \u003cp\u003e#astrazeneca, https,\u003c/p\u003e\n \u003cp\u003eCOVID, people\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 276px;\"\u003e\n \u003cp\u003eI know more people with serious adverse reactions to the vaccine than COVID-19....\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 114px;\"\u003e\n \u003cp\u003eCommunity\u003c/p\u003e\n \u003cp\u003esupport\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 84px;\"\u003e\n \u003cp\u003e7.0%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 150px;\"\u003e\n \u003cp\u003e#covid19, #togetherdeclaration,\u003c/p\u003e\n \u003cp\u003e#covid 19, https,\u003c/p\u003e\n \u003cp\u003eCOVID, help\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 276px;\"\u003e\n \u003cp\u003eThis is strong. Thank you. Oklahoma National Guard \u0026amp; OK Governor. \u0026ldquo;Until a guardsman is activated under Title 10, they fo...\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n\u003c/table\u003e\n\u003cp\u003e\u003cstrong\u003e\u003cem\u003ePlatform-specific discourse patterns\u003c/em\u003e\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eAnalysis of Twitter content revealed further distinctions in public sentiment and thematic framing, particularly around misinformation, advocacy, fear, and community narratives. 21.5% of tweets included vaccine advocacy content, often featuring hashtags like #democratsdeliver and references to vaccination success (Table 5). In contrast, 9.2% expressed fear-based content, typically focused on adverse reactions or perceived coercion.\u003c/p\u003e\n\u003cp\u003eMisinformation-themed tweets (7.4%) included politicized content, exaggerated claims, and links to discredited sources. Notably, community-support narratives comprised 7.0% of tweets, including posts thanking National Guard clinics and highlighting local efforts, aligning with rural interview themes of trust through community action.\u003c/p\u003e\n\u003cp\u003eThese results indicate that social media narratives are often polarized or exaggerated compared to interview responses. While interviews explored distrust in nuanced ways, Twitter reduced concerns to binary hashtags and viral claims. This supports the study\u0026rsquo;s use of platform-aware modeling for misinformation detection and community response strategies.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e\u003cem\u003eTheory of change\u003c/em\u003e\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe multi-layered analysis reveals a theory of change where vaccine hesitancy is shaped by interconnected structural, sociopolitical, and individual-level factors. At its foundation, historical trauma and systemic distrust, particularly in Indigenous and rural communities, play a pivotal role in undermining vaccine confidence. These communities\u0026rsquo; collective memories of colonial exploitation and medical mistrust create deep-seated barriers to acceptance, which are further exacerbated by political polarization and misinformation. Social media amplifies these issues, fueling polarized narratives, misinformation, and fear-based content that shape public perception, especially in rural areas. \u0026nbsp;\u003c/p\u003e\n\u003cp\u003eOn the individual level, personal fears, cultural beliefs, and a preference for natural immunity influence decisions, particularly among rural participants who frame vaccination as a collective rather than an individual choice. Cultural identity and collective trauma contribute to the perception of vaccines as external threats rather than tools for community health.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003e\u0026nbsp;To address vaccine hesitancy, it is essential to focus on long-term trust-building that acknowledges the historical, cultural, and sociopolitical realities of these communities. This requires not only platform-specific misinformation countermeasures but also culturally relevant messaging that resonates with both historical experiences and personal beliefs. An integrated, context-specific approach that combines AI-driven misinformation detection with ethnographic insights provides a pathway to improve vaccine uptake through targeted interventions that engage the structural, sociopolitical, and individual layers of influence.\u003c/p\u003e\n\u003cp\u003eThe thematic insights derived from this study, grounded in the novel theory of systemic distrust and historical trauma, underscore the necessity of moving beyond the traditional public health framework. Public health messaging must be culturally sensitive and historically informed, accounting for the collective nature of health decisions in rural and indigenous communities, as well as the impact of misinformation spread through digital platforms. The grounded theory approach provides a robust methodology for uncovering the underlying drivers of vaccine hesitancy and informs the development of targeted, contextually appropriate interventions.\u0026nbsp;\u003c/p\u003e"},{"header":"Discussion","content":"\u003cp\u003eThis study revealed a significant divergence between public discourse on social media and individual perspectives shared in interviews. Social media content, particularly from rural areas, tended to amplify polarized and negative sentiment, while interviews reflected more balanced, reflective, and contextually grounded views. By combining computational tools with qualitative inquiry, this mixed-methods approach provided a nuanced understanding of vaccine hesitancy across urban and rural Alaska. The integration of RAG models, sentiment analysis, and thematic coding enabled the detection of misinformation trends as well as culturally rooted vaccine narratives. These findings highlight critical gaps between online narratives and lived experiences, pointing to the complex interplay of sociocultural, informational, and geographic factors influencing vaccine decision-making.\u003c/p\u003e\u003cp\u003eA key finding is the stark contrast between social media and interview sentiment. Twitter showed more polarization, especially in rural areas, while interviews revealed more moderate views, with neutral sentiment prevailing and rural participants slightly more likely to express positive perspectives.\u003c/p\u003e\u003cp\u003eThis disparity suggests that social media platforms, particularly Twitter, serve as amplifiers of extreme sentiment rather than accurate mirrors of community-level beliefs. While interviews revealed concerns about vaccine safety, side effects, and mistrust, these were often accompanied by thoughtful reflections, conditional acceptance, or pragmatic considerations.\u003csup\u003e\u003cspan citationid=\"CR28\" class=\"CitationRef\"\u003e28\u003c/span\u003e\u003c/sup\u003e Social media, on the other hand, tended to collapse these nuances into viral, often alarmist narratives. This underscores the importance of supplementing social media surveillance with qualitative data to avoid overestimating public hostility and mischaracterizing the complexity of vaccine decision-making.\u003csup\u003e\u003cspan citationid=\"CR29\" class=\"CitationRef\"\u003e29\u003c/span\u003e\u003c/sup\u003e\u003c/p\u003e\u003cp\u003eAcross both data sources, distrust in government and healthcare systems emerged as a dominant theme, particularly among rural and Indigenous populations. Thematic analysis highlighted that many rural interviewees, especially from Alaska Native communities, framed hesitancy within a broader historical context of colonization and systemic marginalization. References to being \u0026ldquo;lab rats\u0026rdquo; or skepticism toward outside health authorities underscore how the collective memory of exploitation and disempowerment continues to shape vaccine perceptions today.\u003c/p\u003e\u003cp\u003eThis aligns with previous research demonstrating that historical trauma is a persistent determinant of health behavior in Indigenous communities.\u003csup\u003e\u003cspan citationid=\"CR30\" class=\"CitationRef\"\u003e30\u003c/span\u003e,\u003cspan citationid=\"CR31\" class=\"CitationRef\"\u003e31\u003c/span\u003e\u003c/sup\u003e However, the present study adds granularity by showing how these legacies interact with real-time political and media environments. While urban participants were also distrustful, their concerns more frequently revolved around political polarization and media bias rather than historical injustice.\u003csup\u003e\u003cspan citationid=\"CR32\" class=\"CitationRef\"\u003e32\u003c/span\u003e\u003c/sup\u003e Thus, tailored public health messaging must not only address current misinformation but also contend with deeper-rooted narratives of marginalization and betrayal, particularly in rural and Indigenous settings.\u003c/p\u003e\u003cp\u003eThe influence of social media, particularly Facebook and Twitter, emerged as a significant driver of vaccine attitudes.\u003csup\u003e\u003cspan citationid=\"CR33\" class=\"CitationRef\"\u003e33\u003c/span\u003e,\u003cspan citationid=\"CR34\" class=\"CitationRef\"\u003e34\u003c/span\u003e\u003c/sup\u003e Rural interviewees frequently describe encountering and sometimes believing misinformation online. These narratives overlapped with tweets that referenced fake vaccine cards, exaggerated side effects, or political conspiracy theories.\u003c/p\u003e\u003cp\u003eTwitter content analysis reinforced this divide, showing that misinformation-themed tweets, while only 7.4% of the sample, were often highly politicized and emotionally charged. The prevalence of hashtags like #togetherdeclaration and the conflation of unrelated health threats (e.g., RSV, flu, COVID-19) point to an ecosystem where fear and confusion are actively circulated. By contrast, the more reflective tone in interviews suggests that personal conversations may serve as a corrective to online distortion.\u003c/p\u003e\u003cp\u003eBeyond institutional distrust, individual-level concerns, particularly around vaccine safety and natural immunity, were common. Both rural and urban interviewees expressed fears of side effects and distrust in the rapid development of COVID-19 vaccines. However, belief in natural immunity was notably more common among rural participants.\u003c/p\u003e\u003cp\u003eCultural identity also played a key role. Among Indigenous participants, vaccine hesitancy was framed not just as an individual choice but as a community-level issue tied to sovereignty, history, and cultural survival. This suggests that strategies to increase vaccine uptake must avoid individualistic framings and instead engage collective narratives, especially in Indigenous contexts. Public health campaigns that ignore these social dynamics risk reinforcing existing mistrust.\u003c/p\u003e\u003cp\u003eThese findings carry several important implications for public health planning and vaccine promotion. First, they highlight the risks of over-relying on social media analytics to gauge public opinion. While valuable for detecting trends, social media data often reflect the loudest or most extreme voices, especially in politicized contexts like vaccination [40,41].\u003c/p\u003e\u003cp\u003eSecond, the persistence of historical and systemic distrust, particularly among rural and Indigenous communities, demands long-term engagement strategies rooted in cultural humility, respect, and relationship-building. Public health institutions must invest in sustained partnerships that foster local leadership and community ownership of health initiatives.\u003c/p\u003e\u003cp\u003eFinally, the study underscores the need for platform-specific misinformation countermeasures. Community-driven media literacy initiatives, coupled with local storytelling and trusted messengers, may offer more effective pathways for addressing misinformation than top-down fact-checking alone.\u003c/p\u003e\u003cp\u003eThe RAG pipeline proved effective for misinformation detection and tailored response generation. LLaMA-2-7B consistently produced more nuanced, contextually rich outputs than T5-Base, especially for complex queries involving cultural or historical themes. However, it required significantly more computational resources. T5-Base, while faster, occasionally repeated misinformation when exposed to frequently retrieved themes, an issue partially mitigated by post-processing and evidence filtering.\u003c/p\u003e\u003cp\u003eThese findings align with recent work advocating hybrid NLP systems for misinformation detection.\u003csup\u003e\u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e17\u003c/span\u003e\u003c/sup\u003e Our integration of dense retrieval via FAISS\u003csup\u003e\u003cspan citationid=\"CR23\" class=\"CitationRef\"\u003e23\u003c/span\u003e\u003c/sup\u003e, sentiment classification using transformer models. \u003csup\u003e\u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e20\u003c/span\u003e\u003c/sup\u003e Culturally contextualized interviews are also a key component in our model.\u003c/p\u003e\u003cp\u003eBeyond computational performance, this study emphasizes the importance of data triangulation. Sole reliance on social media analysis risks overlooking the nuance and lived experiences that interviews provide. By integrating social and cultural narratives with large-scale textual analysis, our approach addresses this gap.\u003c/p\u003e\u003cp\u003eBy employing grounded theory, this study allows thematic insights to emerge naturally from the data, ensuring that the analysis is context-specific and sensitive to the realities of Alaskan communities. The theory of systemic distrust is not imposed but is constructed based on the patterns revealed through both interviews and social media discourse. The emergent themes (historical trauma, cultural identity, and misinformation) align with grounded theory\u0026rsquo;s approach to data-driven theory development, offering a comprehensive understanding of vaccine hesitancy that incorporates both local contexts and digital influences.\u003c/p\u003e\u003cp\u003eThe study has several limitations. The LLaMA-2-7B model, while effective, is resource-intensive and may not support real-time applications in constrained environments. In contrast, T5-Base, though faster, sometimes lacked contextual sensitivity or echoed common misinformation phrases. Additionally, synthesizing multiple tweets occasionally resulted in conflicting inputs, requiring additional filtering to ensure coherence. Sentiment models, though strong on short-form text, may overlook sarcasm or ambiguity in nuanced narratives.\u003csup\u003e\u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e15\u003c/span\u003e,\u003cspan citationid=\"CR35\" class=\"CitationRef\"\u003e35\u003c/span\u003e\u003c/sup\u003e\u003c/p\u003e\u003cp\u003eDespite these limitations, the study provides a scalable framework for vaccine misinformation tracking and response. Real-time RAG systems, integrated into public health messaging workflows, could help agencies detect emerging misinformation trends and issue timely, evidence-based responses. Tailoring these systems with culturally informed data, as demonstrated here, further enhances their relevance and equity.\u003c/p\u003e"},{"header":"Conclusion","content":"\u003cp\u003eVaccine hesitancy in Alaska reflects a dynamic interplay between historical experience, personal belief, and digital information flows. Rural and urban communities differ not only in the content of their concerns but also in how those concerns are expressed and shaped by social context. The thematic insights derived from this study, grounded in the novel theory of systemic distrust and historical trauma, underscore the necessity of moving beyond traditional public health frameworks. Public health messaging must be culturally sensitive and historically informed, accounting for the collective nature of health decisions in rural and indigenous communities, as well as the impact of misinformation spread through digital platforms. The grounded theory approach provides a robust methodology for uncovering the underlying drivers of vaccine hesitancy and informs the development of targeted contextually appropriate interventions. Going forward, efforts to improve vaccine confidence must move beyond combating misinformation and address the deeper structural and cultural roots of distrust.\u003c/p\u003e"},{"header":"Declarations","content":"\u003cp\u003e\u003cspan\u003eEthical approval was obtained from the Institutional Review Board at Rutgers University (IRB #Pro2023002010, dated 04.12.2024). Informed consent was obtained from all participants. Interview data were anonymized. Social media data were collected from publicly available Twitter posts per the platform\u0026apos;s terms of service, with no personally identifiable information stored.\u003c/span\u003e\u003c/p\u003e\u003ch2\u003eAcknowledgement\u003c/h2\u003e\u003cp\u003eSupported in part by a research grant from Investigator-Initiated Studies Program of Merck Sharp \u0026amp; Dohme LLC. The opinions expressed in this paper are those of the authors and do not necessarily represent those of Merck Sharp \u0026amp; Dohme LLC\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\n\u003cli\u003eM. Brown, Vaccine misinformation outpaces efforts to counter it, Columbia Public Health (2024). URL https://www.publichealth.columbia.edu/news/vaccinemisinformation-outpaces-efforts-counter-it.\u003c/li\u003e\n\u003cli\u003eWHO, Vaccine hesitancy: a growing challenge for immunization programs, World Health Organization: Geneve, Switzerland (2015).\u003c/li\u003e\n\u003cli\u003eRobinson R, Nguyen E, Wright M, et al. Factors contributing to vaccine hesitancy and reduced vaccine confidence in rural underserved populations. \u003cem\u003eHumanit Soc Sci Commun\u003c/em\u003e 2022; \u003cstrong\u003e9\u003c/strong\u003e(1): 416.\u003c/li\u003e\n\u003cli\u003eScharff DP, Mathews KJ, Jackson P, Hoffsuemmer J, Martin E, Edwards D. More than Tuskegee: understanding mistrust about research participation. \u003cem\u003eJ Health Care Poor Underserved\u003c/em\u003e 2010; \u003cstrong\u003e21\u003c/strong\u003e(3): 879\u0026ndash;97.\u003c/li\u003e\n\u003cli\u003eTjaden J, Haarmann E, Savaskan N. Experimental evidence on improving COVID-19 vaccine outreach among migrant communities on social media. \u003cem\u003eSci Rep\u003c/em\u003e 2022; \u003cstrong\u003e12\u003c/strong\u003e(1): 16256.\u003c/li\u003e\n\u003cli\u003eWilson SL, Wiysonge C. Social media and vaccine hesitancy. \u003cem\u003eBMJ Glob Health\u003c/em\u003e 2020; \u003cstrong\u003e5\u003c/strong\u003e(10).\u003c/li\u003e\n\u003cli\u003eCinelli M, De Francisci Morales G, Galeazzi A, Quattrociocchi W, Starnini M. The echo chamber effect on social media. \u003cem\u003eProc Natl Acad Sci U S A\u003c/em\u003e 2021; \u003cstrong\u003e118\u003c/strong\u003e(9).\u003c/li\u003e\n\u003cli\u003eLoomba S, de Figueiredo A, Piatek SJ, de Graaf K, Larson HJ. Measuring the impact of COVID-19 vaccine misinformation on vaccination intent in the UK and USA. \u003cem\u003eNat Hum Behav\u003c/em\u003e 2021; \u003cstrong\u003e5\u003c/strong\u003e(3): 337\u0026ndash;48.\u003c/li\u003e\n\u003cli\u003eNazli SB, Yigman F, Sevindik M, Deniz Ozturan D. Psychological factors affecting COVID-19 vaccine hesitancy. \u003cem\u003eIr J Med Sci\u003c/em\u003e 2022; \u003cstrong\u003e191\u003c/strong\u003e(1): 71\u0026ndash;80.\u003c/li\u003e\n\u003cli\u003eLin F, Chen X, Cheng EW. Contextualized impacts of an infodemic on vaccine hesitancy: The moderating role of socioeconomic and cultural factors. \u003cem\u003eInf Process Manag\u003c/em\u003e 2022; \u003cstrong\u003e59\u003c/strong\u003e(5): 103013.\u003c/li\u003e\n\u003cli\u003eJ. Cao, C. M. Ramirez, R. M. Alvarez, The politics of vaccine hesitancy in the united states, Social Science Quarterly 103 (1) (2022) 42\u0026ndash;54.\u003c/li\u003e\n\u003cli\u003eZ. Wang, M. M. Rodriguez Morales, K. Husak, M. Kleinman, S. Parthasarathy, In communities we trust institutional failures and sustained solutions for vaccine hesitancy, Tech. rep. (2021).\u003c/li\u003e\n\u003cli\u003eLaurencin CT, McClinton A. The COVID-19 Pandemic: a Call to Action to Identify and Address Racial and Ethnic Disparities. \u003cem\u003eJ Racial Ethn Health Disparities\u003c/em\u003e 2020; \u003cstrong\u003e7\u003c/strong\u003e(3): 398\u0026ndash;402.\u003c/li\u003e\n\u003cli\u003eV. Sanh, L. Debut, J. Chaumond, T. Wolf, Distilbert base uncased fine tuned on sst-2, https://huggingface.co/distilbert/distilbertbase-uncased-finetuned-sst-2-english, hugging Face (2019).\u003c/li\u003e\n\u003cli\u003eF. Barbieri, J. Camacho-Collados, L. Espinosa Anke, L. Neves, Tweet-Eval: Unified benchmark and comparative evaluation for tweet classification, in: Findings of the Association for Computational Linguistics: EMNLP 2020, Association for Computational Linguistics, Online, 2020, pp. 1644\u0026ndash;1650. doi:10.18653/v1/2020.findings-emnlp.148.\u003c/li\u003e\n\u003cli\u003eC. Raffel, N. Shazeer, A. Roberts, K. Lee, S. Narang, M. Matena, Y. Zhou, W. Li, P. J. Liu, Exploring the limits of transfer learning with a unified text-to-text transformer, Journal of Machine Learning Research 21 (140) (2020) 1\u0026ndash;67.\u003c/li\u003e\n\u003cli\u003eLi H, Huang J, Ji M, Yang Y, An R. Use of Retrieval-Augmented Large Language Model for COVID-19 Fact-Checking: Development and Usability Study. \u003cem\u003eJ Med Internet Res\u003c/em\u003e 2025; \u003cstrong\u003e27\u003c/strong\u003e: e66098.\u003c/li\u003e\n\u003cli\u003eWilliams, David R., and Selina A. Mohammed. \u0026quot;Discrimination and racial disparities in health: evidence and needed research.\u0026quot; Journal of behavioral medicine 32.1 (2009): 20-47.\u003c/li\u003e\n\u003cli\u003eKaggle Dataset. Available at [https://www.kaggle.com/datasets/rajugc/kaggle-dataset], last accessed on 08.13.2025.\u003c/li\u003e\n\u003cli\u003eCardiffNLP, Cardiffnlp/twitter-roberta-base-sentiment, accessed: 2025-03-22 (2021). URL https://huggingface.co/cardiffnlp/twitter-robertabase-sentiment.\u003c/li\u003e\n\u003cli\u003eH. Face, Auto classes, accessed: 2025-03-22 (2025). URL https://huggingface.co/docs/transformers/model_doc/auto#transformers.AutoTokenizer.\u003c/li\u003e\n\u003cli\u003eOpenAI, Gpt-4 turbo, accessed: 2025-03-22 (2023). URL https://platform.openai.com/docs/models/gpt-4-turbo.\u003c/li\u003e\n\u003cli\u003eH. J\u0026acute;egou, M. Douze, J. Johnson, Faiss: A library for efficient similarity search, accessed: 2025-03-22 (2017). URL https://engineering.fb.com/2017/03/29/datainfrastructure/faiss-a-library-for-efficient-similaritysearch.\u003c/li\u003e\n\u003cli\u003eN. Reimers, I. Gurevych, Sentence-bert: Sentence embeddings using siamese bert-networks, in: Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing, 2019. URL https://arxiv.org/abs/1908.10084.\u003c/li\u003e\n\u003cli\u003ePinecone, Embeddings to identify fake newsAccessed: 2025-03-22 (2023). URL https://www.pinecone.io/learn/embeddings-identifyfake-news/.\u003c/li\u003e\n\u003cli\u003eH. Touvron, L. Martin, K. Stone, P. Albert, A. Almahairi, Y. Babaei, N. Bashlykov, S. Batra, P. Bhargava, S. Bhosale, et al., Llama 2: Open foundation and fine-tuned chat models, arXiv preprint arXiv:2307.09288 (2023).\u003c/li\u003e\n\u003cli\u003eHiratsuka VY, Beans JA, Robinson RF, Shaw JL, Sylvester I, Dillard DA. Self-Determination in Health Research: An Alaska Native Example of Tribal Ownership and Research Regulation. \u003cem\u003eInt J Environ Res Public Health\u003c/em\u003e 2017; \u003cstrong\u003e14\u003c/strong\u003e(11).\u003c/li\u003e\n\u003cli\u003eLedford CJW, Cafferty LA, Moore JX, et al. The dynamics of trust and communication in COVID-19 vaccine decision making: A qualitative inquiry. \u003cem\u003eJ Health Commun\u003c/em\u003e 2022; \u003cstrong\u003e27\u003c/strong\u003e(1): 17\u0026ndash;26.\u003c/li\u003e\n\u003cli\u003eNowak SA, Chen C, Parker AM, Gidengil CA, Matthews LJ. Comparing covariation among vaccine hesitancy and broader beliefs within Twitter and survey data. \u003cem\u003ePLoS One\u003c/em\u003e 2020; \u003cstrong\u003e15\u003c/strong\u003e(10): e0239826.\u003c/li\u003e\n\u003cli\u003eGone JP, Hartmann WE, Pomerville A, Wendt DC, Klem SH, Burrage RL. The impact of historical trauma on health outcomes for indigenous populations in the USA and Canada: A systematic review. \u003cem\u003eAm Psychol\u003c/em\u003e 2019; \u003cstrong\u003e74\u003c/strong\u003e(1): 20\u0026ndash;35.\u003c/li\u003e\n\u003cli\u003eHeart MY, Chase J, Elkins J, Altschul DB. Historical trauma among Indigenous Peoples of the Americas: concepts, research, and clinical considerations. \u003cem\u003eJ Psychoactive Drugs\u003c/em\u003e 2011; \u003cstrong\u003e43\u003c/strong\u003e(4): 282\u0026ndash;90.\u003c/li\u003e\n\u003cli\u003eLi, Rui, et al. \u0026quot;A Tale of Two Cities: COVID-19 Vaccine Hesitancy as a Result of Racial, Socioeconomic, Digital, and Partisan Divides.\u0026quot; ISPRS International Journal of Geo-Information 12.4 (2023): 158.\u003c/li\u003e\n\u003cli\u003eAl-Regaiey KA, Alshamry WS, Alqarni RA, et al. Influence of social media on parents\u0026apos; attitudes towards vaccine administration. \u003cem\u003eHum Vaccin Immunother\u003c/em\u003e 2022; \u003cstrong\u003e18\u003c/strong\u003e(1): 1872340.\u003c/li\u003e\n\u003cli\u003eCascini F, Pantovic A, Al-Ajlouni YA, et al. Social media and attitudes towards a COVID-19 vaccination: A systematic review of the literature. \u003cem\u003eEClinicalMedicine\u003c/em\u003e 2022; \u003cstrong\u003e48\u003c/strong\u003e: 101454.\u003c/li\u003e\n\u003cli\u003eK. Roehrick, Valence aware dictionary and sentiment reasoner (vader), R-project. org/package= vader (2020).\u003c/li\u003e\n\u003c/ol\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":true,"hideJournal":true,"highlight":"","institution":"","isAcceptedByJournal":false,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true},"keywords":"Misinformation, Vaccination, Sentiment, NLP, RAG, Social Media, Public Health","lastPublishedDoi":"10.21203/rs.3.rs-7368501/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-7368501/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003cp\u003eVaccine misinformation poses a significant public health threat, particularly in communities with varying levels of vaccine confidence. This study investigated vaccine hesitancy across Alaska\u0026rsquo;s diverse communities by triangulating public sentiment from social media with individual beliefs gathered through qualitative interviews. The aim was to explore how online discourse influences vaccine-related decision-making and to develop tools for real-time misinformation detection.\u003c/p\u003e\u003cp\u003eWe employed a mixed-methods approach, analyzing 1,300 Alaska-specific tweets and conducting 87 semi-structured interviews across urban and rural communities. A Retrieval-Augmented Generation (RAG) system was developed, integrating the context-rich LLaMA-2-7B model with the efficient T5-Base model to balance accuracy and computational performance. The system used sentence embeddings and FAISS-based similarity search to identify misinformation themes and generate context-aware responses grounded in real-world data.\u003c/p\u003e\u003cp\u003eSentiment analysis revealed that rural social media posts exhibited significantly higher negativity and misinformation (55.6% negative sentiment) compared to urban posts. In contrast, interview data reflected more balanced and nuanced attitudes toward vaccination. Thematic analysis identified systemic distrust and personal beliefs, particularly among Indigenous and rural populations, as key drivers of hesitancy. Model evaluation showed that LLaMA-2-7B outperformed T5-Base in contextual accuracy, while T5-Base offered faster but occasionally less accurate responses.\u003c/p\u003e\u003cp\u003eBy combining AI-driven insights with ethnographic data, this study highlights the divergence between online narratives and lived experiences. The proposed framework offers a scalable, real-time method for detecting misinformation and informing culturally responsive public health messaging. Future work will focus on optimizing system efficiency and collaborating with digital platforms to reduce the spread of viral misinformation.\u003c/p\u003e","manuscriptTitle":"Integrating Retrieval-Augmented Generation and Thematic NLP for Vaccine Confidence Modeling in Alaska","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2025-08-19 08:38:56","doi":"10.21203/rs.3.rs-7368501/v1","editorialEvents":[{"type":"communityComments","content":0}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"258f0e62-747a-4f27-8d15-8ee402d6d862","owner":[],"postedDate":"August 19th, 2025","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"posted","subjectAreas":[],"tags":[],"updatedAt":"2025-08-19T08:38:56+00:00","versionOfRecord":[],"versionCreatedAt":"2025-08-19 08:38:56","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-7368501","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-7368501","identity":"rs-7368501","version":["v1"]},"buildId":"XKTyCvWXoU3ODBz1xrDgd","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}