Empowering Doctoral Academic Research: Artificial Intelligence-driven Insights from Large Language Models

Empowering Doctoral Academic Research: Artificial Intelligence-driven Insights from Large Language Models | 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 Article Empowering Doctoral Academic Research: Artificial Intelligence-driven Insights from Large Language Models Yihuan Yuan, Jamalludin Harun This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-4337026/v1 This work is licensed under a CC BY 4.0 License Status: Under Review Version 1 posted 9 You are reading this latest preprint version Abstract The ever-expanding volume and complexity of academic research pose significant challenges for researchers, particularly doctoral students. In response to these challenges, utilizing Large Language Models (LLMs) has emerged as a promising alternative solution. Such LLMs as ChatGPT, Bing Chat and Google Bard are applied in academic research. This study conducted semi-structured interviews with 50 PhD students and used thematic analysis to explore the application of LLMs in academic research. The results indicate that LLMs assist literature reading by extracting main content, providing research topics, and making reading convenient; assist research design by generating research design ideas; assist academic writing by generating writing ideas, polishing writing, analyzing and visualizing data; assist knowledge construction by offering subject matter expertise and promoting science; assist admin works by writing admin emails. Based on these, a five-dimensional framework of AI-assisted academic research (AIAAR) has been established to explain the assistance of LLMs in academic research. This research not only sheds light on the practical benefits of integrating LLMs in academic research but also provides insights into optimizing their usage for enhanced scholarly productivity and knowledge advancement. Physical sciences/Mathematics and computing/Computational science Physical sciences/Mathematics and computing/Software Large language model artificial intelligence academic research doctoral students AIAAR Figures Figure 1 1. Introduction Artificial Intelligence (AI) has had a significant impact on various facets of contemporary society in recent years, offering numerous benefits that are transforming various aspects of modern life (Lund et al., 2023 ). Its development is breaking down barriers between different areas of application, extending its potential to a wide range. The integration of AI into education represents a continuous evolution in this field (Imran & Almusharraf, 2023 ). Large language models (LLMs) represent a notable advance in AI field (Kasneci et al., 2023 ). These models refer to deep learning architectures that have been trained on large textual datasets, enabling them to produce natural language text and facilitate the understanding of textual meaning. LLMs can handle a variety of natural language tasks, including text categorization, question answering and dialogue, making them an important component of artificial intelligence (Kandpal et al., 2023 ). Numerous LLMs have emerged in recent years, including GPT (Radford et al., 2018 ), BERT (Devlin et al., 2018 ), XLNet (Yang et al., 2019 ), T5 (Raffel et al., 2020 ), RoBERTa (Liu et al., 2019 ), GPT-3 (Floridi & Chiriatti, 2020 ; Scao et al., 2022 ), Bing Chat (Kelly et al., 2023 ), Google Bard (Urman & Makhortykh, 2023), and the widely recognized GPT-3.5 and GPT-4 (Rosenfeld & Lazebnik, 2024 ). ChatGPT, an AI-driven conversational agent powered by a large language models (LLMs) trained on rich Internet text data, is expected to address numerous limitations of previous chatbot technology and reshape learning dynamics (Heidt, 2023 ). Built on a transformer architecture, LLMs models are pre-trained on massive text datasets to mimic human text generation, respond to queries, assist with translation and summarization, and execute various natural language processing (NLP) functions via a unified pre-training and fine-tuning process. Large language models are currently used in many areas and have had positive results. Advancements in LLMs have paved the way for their integration into educational environments, providing opportunities to improve teaching and learning. A notable area is the creation and development of educational content, where scientists have used LLMs to create interactive learning resources such as quizzes and flashcards. These resources are expected to enhance student engagement and learning outcomes (Dijkstra et al., 20–22; Gabajiwala et al., 20–22). What’s more, LLMs offer several benefits, such as assisting in the generation of academic writing and research (Bin-Hady et al., 2023 ). For university students in particular, these models serve as valuable tools for research and writing assignments. They facilitate tasks such as creating summaries and outlines of texts, allowing students to quickly grasp the core ideas of a text and structure their thoughts for writing (Kasneci et al., 2023 ). The ever-increasing volume and intricacy of academic literature pose considerable challenges for researchers, particularly doctoral students, in navigating through vast amounts of information, designing robust research methodologies, and synthesizing knowledge effectively (Dempere et al, 2023 ). In response to these challenges, the integration of LLMs into doctoral research holds immense potential to enhance efficiency, productivity, and the quality of academic outputs. By harnessing LLMs, doctoral students can overcome research challenges more effectively and advance knowledge in their respective fields (Zhang et al., 2023 ; Jin et al., 2024 ). However, up to now, there are relatively few discussions about how doctoral students can effectively use LLMs for academic research. Namely, there is still a research gap in this field, and more in-depth investigation and exploration are still needed to reveal the practical application and potential value of LLMs in doctoral academic research. In response to this research gap, a qualitative investigation is conducted in this study. This exploratory study attempts to investigate how doctoral students use LLMs in academic research. Specifically, this study aims to establish a framework to explain the application of LLMs in academic research. The following research objectives guide the study: To explore how the large language models (LLMs) are used in doctoral academic research; To establish a framework of LLMs’ application in academic research to offer guidance for future research work. 2. Methodology This research adopts a qualitative approach, predominantly drawing insights from interviews with 50 PhD students. Grounded theory underpins the research design, as it holds significance in qualitative inquiry (Creswell, 2013). Grounded theory posits that data collection and analysis are iterative processes, driven by research objectives while aiming to achieve saturation. In line with it, semi-structured interviews are employed as the primary method of data collection. Interview questions: 1. How do you use LLMs in academic research? 2. What is the value of LLMs to doctoral students’ academic research? 2.1 Participants The study involved 50 PhD students (shown in Table 1 ) from a variety of academic backgrounds and countries who were actively using LLMs in their academic research, with the recruitment period spanning from February 1, 2023, to February 1, 2024. These participants were carefully selected to represent a wide range of disciplines and geographical regions, ensuring a broad understanding of the different ways in which LLMs were used in academic research in different contexts. Through the inclusion of a diverse group of students, the study aimed to capture a wide range of perspectives and experiences regarding the use of LLMs in academic environments, and to highlight their effectiveness and challenges in different academic disciplines and cultural contexts. Table 1 Information of participants Item Category Population Age 25–35 n = 40 35–45 n = 10 Gender Male n = 25 Female n = 25 Research area English n = 10 Education n = 10 Dig data and AI n = 10 Mechanical Engineering n = 10 Materials science n = 10 Study country Malaysia n = 30 Japan n = 5 UK n = 5 America n = 5 Italy n = 5 Participants were invited to participate in the interview via WeChat, accompanied by a cover letter. This letter explained the purpose of the study and assured that their responses would be kept anonymous and confidential. It further highlighted the voluntary nature of their participation, ensuring they knew they could opt-out at any time if they were uneasy during the interview. Written informed consent was secured from those who chose to take part. 2.2 Procedures and data analysis After selecting 50 PhD students, the researchers had individual interviews with each participant via Wechat. Each interview session lasted approximately 30 minutes, allowing sufficient time for in-depth discussion. Throughout the interviews, the researchers used audio recording to accurately capture the dialogue. When each interview was completed, transcription of the audio into text format was carried out using speech recognition technology. The transcription process ensured that all spoken contents were precisely preserved for subsequent analysis. For data analysis, thematic analysis was used to analyze the data from the interviews on NVivo. Braun and Clarke’s thematic analysis (2022) was also employed in this study, which is an iterative process consisting of six steps: (1) becoming familiar with the data, (2) generating codes, (3) generating themes, (4) reviewing themes, (5) defining and naming themes, and (6) locating exemplars. This analytical approach involved the systematic identification and analysis of recurring themes, patterns and insights within the interview data. By immersing themselves in the textual data, the researchers gained a nuanced understanding of the participants’ perspectives, experiences and practices in relation to their use of LLMs in academic research. 3. Results and Discussions This section provides a comprehensive analysis of the research findings and discussions, structured according to the predetermined research objectives. Through an in-depth analysis of the data collected, the following subsections explore the key findings of the study and their significance in addressing the research objectives. The themes found in the coding process are shown in Table 2 . Table 2 The coding process NO. Theme Code Exemplars 1 Literature reading Extract main content Interviewee 2 : Ask it (ChatGPT) about the main idea of the literature or ask him to summarize it, and I will get an answer immediately. Interviewee 26 : ChatGPT helps me summarize the academic paper in a way that is easy to understand… Provide research topics Interviewee 21 : …give the topic and time range of the paper, ChatGPT will automatically give some representative papers, based on the paper, I can get a general idea of the relevant field. Interviewee 35 : (ChatGPT) quickly and accurately grasp a large number of papers and background knowledge on a certain topic, based on which I can find my own research topic. Make reading convenient Interviewee 47 : Most of the time, I need to read several papers, especially when two of the same methods are compared, it is necessary to compare several papers at the same time. So I can use ChatGPT to read two articles at the same time and compare them. Interviewee17 : I can have multiple rounds of interaction with ChatGPT to discuss the issues in the paper. In this case, it is entirely possible to think of it as a supervisor, can ask questions, can answer, can discuss, can debate. 2 Research design Generate research design Interviewee 23 : ChatGPT can be used to generate experimental designs or research protocols. For example, a teaching and research department has designed two sets of training software and three training modes, claiming that the emotional intelligence of teenagers can be improved through practice. Now design an experiment to test its effectiveness. Interviewee 13 : In the research, I can use chatGPT and Bing Chat to generate interview questions, questionnaire questions and examination papers. Interviewee 40 : …through the conversation with chatGPT, the rationality of the experimental design was discussed, the experimental parameters were determined, and the experimental results were interpreted. 3 Academic writing Generate writing ideas Interviewee 30 : When I input the research topic, Google Bard can provide some initial ideas, arguments, paper structure, etc. Polish writing Interviewee 18 : ChatGPT helps me refine the details, such as adding specific information, data or examples to the article to support the argument and enrich the content. Interviewee 25 : As my native language is not English, there will be unprofessional expressions when writing papers. Using chatGPT can help me replace those unprofessional or too colloquial phrases and words, so as to improve the professionalism and formality of my writing. Analyse and visualize data Interviewee 44 : ChatGPT-4 can generate charts and make content visual by adding charts and visual elements. Graphic combination provides more clearly demonstrate the point of view. Interviewee 17 : ChatGPT-4 can receive various types of data and conduct data analysis through natural language. Such as extracting the information in the graph, conducting topic analysis, and generating data analysis reports. 4 Knowledge construction Offer expertise Interviewee 12 : ChatGPT has completely changed the way code is developed. It can empower programming, and when I encounter obstacles in programming, I can find answers in chatGPT. Interviewee 19 : In the research, I will encounter some unfamiliar research terms, and it will take a long time to search on the Internet. If I input the question into ChatGPT, I can get one answer quickly, greatly improving the learning efficiency. Promote subject area Interviewee 21 : If you can’t explain your research project clearly to the children, it means that your understanding is not deep enough. ChatGPT helps us understand scientific concepts and projects from the perspective of the general public. 5 Admin assistance provide communication assistance Interviewee 2 : ChatGPT can play an important auxiliary role in academic communication, especially in English mail. For students whose native language is not English, sending letters to foreign tutors has always been a major problem. Now, with the help of ChatGPT, an honest, well-structured letter can be generated by typing a single sentence. Interviewee 10 : Publishing a paper is a long process, and sometimes I don't receive a reply long after submitting it. I use ChatGPT to generate emails urging review. ChatGPT understands the writing style so much, which not only shows that reviewing the paper is very important to me but also considers the position of the reviewers. Table 3 Frequency of codes Code Frequency Number of participants Extract main content n = 43 n = 43 Provide research topics n = 27 n = 27 Make reading convenient n = 44 n = 40 Generate research design n = 35 n = 35 Generate writing ideas n = 38 n = 38 Polish writing n = 55 n = 50 Analyse and visualize data n = 20 n = 20 Offer expertise n = 15 n = 15 Promote subject area n = 19 n = 19 provide communication assistance n = 8 n = 8 Five themes presented in Table 2 explain how LLMs influence doctoral students’ academic research (RO1). Table 3 presents the frequency of mention during the interviews and the number of participants who highlighted specific aspects. The results show LLMs contribute to academic research in five key areas: literature review, research design, academic writing, knowledge construction, and admin assistance. These aspects are structured into a five-dimensional model for academic research (RO 2). 3.1 Literature reading LLMs assist literature reading from 3 aspects: extract main content, provide research topics, and make reading convenient. Firstly, LLMs can extract the main content of the literature immediately. Just like Interviewee 2 said, ChatGPT can summarize the main idea of the literature immediately . What’s more, Interviewee 6 mentioned ChatGPT provides an easy way to make academic papers understood . Secondly, LLMS can provide research topics for researchers. Just like interviewee 11 and 15 mentioned based on the paper or topics found by ChatGPT, they could have ideas about their own research topics. Third, LLMs can make reading convenient. According to Interviewee 7, ChatGPT provides the function of reading and comparing two papers at the same time . In addition, Interviewee 17 mentioned multiple rounds of interaction can be conducted on ChatGPT . This greatly improves the efficiency of reading. These findings are aligned with some previous research and enrich them in some degree. LLMs have become valuable tools for academic reading, providing comprehensive support throughout the research process. Antu and Richards (2024) found that ChatGPT, for example, can effectively organize literature, presenting a structured overview that expedites the reading process. LLMs excel at drawing out essential information from extensive sources, skilfully analyzing and summarising large volumes of text to extract key concepts, arguments and findings according to the needs of users. LLMs condense lengthy passages into succinct summaries, saving time and effort in reading. Eysenbach’s ( 2023 ) study highlights their capability to generate medical research summaries, aiding medical students’ comprehension of key findings and implications. This feature is particularly useful for academics who need to sift through a large amount of literature to inform their research or stay up-to-date with the latest developments in their field. Furthermore, LLMs are essential for proposing potential research directions by reviewing the existing literature. They analyze academic articles to identify gaps, controversies, or emerging themes within a particular field, providing valuable insights that inspire researchers to explore new avenues of inquiry or develop research questions to address unmet needs or unresolved issues. LLMs serve as a source of inspiration and guidance, allowing researchers to undertake innovative and meaningful research that contributes to the advancement of knowledge. This result supports the finding in Antu et al.(2023)’ research that ChatGPT may help researchers identify trends and gaps in current research. In a word, LLMs enhance the accessibility and convenience of the research reading process by offering a range of features and capabilities. This enables researchers to interact with literature more effectively and efficiently, accommodating their diverse requirements and inclinations in academic reading. 3.2 Research design LLMs are also helpful in research design. According to Interviewee 3, ChatGPT can be used to generate experimental designs or research protocols . According to Interviewee 13, ChatGPT and Bing Chat can generate interview questions, questionnaire questions and examination papers . Besides these, Interviewee 50 mentioned, through the conversation with ChatGPT, the rationality of the research design can be justified. LLMs can significantly support the research design process by assisting researchers at every stage. They can help to explore and refine research inquiries, identify relevant variables and factors, and select appropriate methodologies and approaches. Similarly, De Kok ( 2023 ) mentioned GPT-4 and ChatGPT were used to select an approach first and then choose an appropriate model so as to eliminate discrepancies in the research approach. Additionally, LLMs can simplify literature reviews by summarizing existing research, identifying areas lacking coverage, and suggesting potential paths for further exploration. Leveraging synthesized knowledge and insights from extensive textual data, LLMs can also assist in formulating hypotheses and defining research objectives. This finding can be interpreted by shedding light on ChatGPT is efficient in generating text, especially creative writing, and can produce human-like academic and professional tasks (Rasul et al., 2023 ; Suaverdez & Suaverdez, 2023 ). The transformative impact of AI on the research landscape is highlighted by the discovery that LLMs contribute to the design of research and make it possible to justify the rationality of research. By using LLMs, researchers can improve the accuracy, efficiency, and credibility of their research, leading to innovation and progress in various fields. 3.3 Academic writing LLMs can optimize academic writing, including generate outlines, polish writing as well as analyze and visualize data. For example, Google Bard can provide some initial ideas, arguments, paper structure (Interviewee 10). ChatGPT can be used to refine the details, such as adding specific information, data or examples to the article to support the argument and enrich the content (Interviewee 18). What’s more, for EFL (English as foreign) learners, ChatGPT can be used to replace those unprofessional or too colloquial phrases and words, so as to improve the professionalism and formality of writing (Interviewee 5). In addition, for data analysis, ChatGPT-4 can generate charts and make content visual by adding charts and visual elements (Interviewee 4). It can also receive various types of data and conduct data analysis through natural language (Interviewee 17). LLMs are adept at generating outlines for academic writing. They automatically arrange and organize complex information into coherent structures. Researchers can input their research objectives, main points, and supporting evidence into LLMs, and then generate well-structured outlines outlining the paper’s logical progression. This not only saves researchers’ time during the initial planning stages of writing but also ensures that their papers adhere to a clear and organized format. Generated outlines by LLMs can act as a guide for writing, steering researchers through the writing process and ensuring comprehensive coverage of essential aspects in their paper. This finding supports the idea that ChatGPT can assist students in generating new ideas for their writing tasks by suggesting topics, themes, and perspectives they may not have otherwise considered (Kasne et al., 2023). LLMs are proficient in providing recommendations to improve the clarity, coherence, and conciseness of academic writing. It is crucial to maintain objectivity and impartiality by avoiding biased or emotional language and clearly indicating subjective assessments. LLMs can refine language, strengthen arguments, and enhance the overall readability of academic texts. They can detect grammatical errors, improve sentence structure, suggest synonyms for repetitive terms, and propose stylistic enhancements to improve the quality of academic writing. It is imperative to ensure a clear and coherent structure, use precise language, and adhere to formal conventions. The creation of ChatGPT increases the potential for producing accurate and consistent content (Imran & Almusharraf, 2023 ). LLMs can be used by researchers to refine their writing and generate polished and professional manuscripts that captivate and resonate with readers. The substantial utility of ChatGPT in effectively addressing various writing tasks has been underscored by global academic and scientific discourse (Sallam, 2023). LLMs are valuable tools for data-driven research writing as they can analyze and visualize data. Researchers can input raw data or descriptive statistics into LLMs. LLMs can then generate insightful analyses, interpret findings, and create visualizations such as charts, graphs, and tables to illustrate key trends and patterns. LLMs can assist researchers in discovering concealed insights in their data, identifying correlations and relationships, and communicating complex findings clearly and compellingly. For example, Fink’s group (2023) found that the LLMs—ChatGPT and GPT-4 perform well in data mining and labeling oncologic phenotypes. Furthermore, LLMs can aid in producing data-driven narratives that contextualize research findings within the wider academic discourse, thereby enhancing the credibility and impact of research publications. Through the use of chatbots, researchers can optimize the writing process, increase the quality and effectiveness of their publications, and contribute to the advancement of scientific knowledge across disciplines. Consistent with this, Imran and Almusharraf ( 2023 ) also found that chatbots are useful tools for facilitating and supporting the academic process. 3.4 Knowledge Construction LLMs can assist in knowledge construction. In the process of research, it is very common to encounter unfamiliar or difficult professional questions. Asking others takes a long time and is not conducive to cultivating the exploration spirit of doctoral students. Searching on the Internet is the choice of many doctoral students, but the information on the Internet is too complex, and opinions on the same question or concept are varied, so it is difficult to find a definite answer in a short time. LLMs can help to solve this problem. Just like Interviewee 19 said If I input the question into ChatGPT, I can get one answer quickly, greatly improving the learning efficiency. On the other hand, LLMS can be used to promote science. According to Interviewee 1, If you can’t explain your research project clearly to the children, it means that your understanding is not deep enough. Fortunately, ChatGPT helps us understand scientific concepts and projects from the perspective of the general public. LLMs are trained on large amounts of text data from diverse sources, allowing them to acquire comprehensive knowledge across various disciplines, such as science, humanities, social sciences, technology, and medicine. LLMs possess a broad range of knowledge, making them a valuable resource for researchers, educators, students, and professionals seeking subject matter expertise. Imran and Almusharraf’s ( 2023 ) findings support the notion that ChatGPT efficiently comprehends and interprets user input commands to generate appropriate answers, including examples related to any subject. Namely, LLMs can generate relevant and informative responses based on their understanding of the subject matter. This instantaneous access to information enables users to obtain insights, explanations, and perspectives on complex topics in real time, facilitating research, decision-making, and problem-solving processes. LLMs possess a vast knowledge base and demonstrate a deep understanding of specific domains. Their exposure to specialized literature and texts within particular fields enables them to develop detailed insights, understand terminological nuances, and comprehend contexts deeply within those subject areas. This profound understanding equips LLMs to provide detailed explanations, analyses, and interpretations on complex subjects, similar to experts in those respective domains. Yang et al.(2023) also agreed that LLM could provide a detailed and understandable explanation. In conclusion, the proficiency of LLMs in knowledge construction marks a substantial transformation in how information is accessed, synthesized, and shared in the digital era. Leveraging the capabilities of LLMs enables individuals and institutions to tap into extensive expertise spanning various fields, expedite research and creativity, and tackle intricate societal issues. 3.5 Admin assistance LLMs can serve as an Admin assistant for every doctoral student. For example, it can be used to write E-mails. For students whose native language is not English, sending letters to foreign supervisors has always been a major problem. With the help of ChatGPT, a well-structured letter can be generated by typing a single sentence (Interviewee 2). In addition, LLMs can generate well-formulated emails that help authors ask about the review progress of publishing (Interviewee 9). LLMs can be of great assistance to doctoral students in managing their academic communication more effectively. They can help with organizing emails, scheduling meetings, and prioritizing tasks related to research, collaborations, and academic commitments. LLMs can save time and reduce cognitive load by providing students with concise overviews of important information and action items. According to Cotton et al. ( 2023 ), LLMs have the potential to enhance work pace and efficiency in daily routines. LLMs can assist doctoral students in preparing for academic conferences and presentations by generating presentation slides, summarising key points, and providing speaking prompts. They can also help organize conference schedules, coordinate travel arrangements, and facilitate networking with other researchers and scholars in the field. In addition, LLMs can generate answers to frequently asked questions and anticipate potential questions during Q&A sessions. In conclusion, the invention of ChatGPT has the potential to streamline academic research processes by significantly reducing time and effort. Its capacity to generate coherent and well-structured text across various subjects can alleviate the workload for students and educators, allowing them to allocate more time to other tasks (Lund et al., 2023 ; Yan, 2023 ). The idea of employing LLMs as administrative aids for doctoral students in academic communication shows promise in enhancing efficiency, productivity, and collaboration throughout the research journey. With the help of LLMs, doctoral candidates can benefit from personalized assistance, access up-to-date information, and utilize advanced communication tools to navigate academic pursuits more efficiently. 3.6 Five-dimension framework of AI-assisted Academic Research (AIAAR) Based on the above findings, a five-dimension framework of AIAAR (shown in Fig. 1) is established, which explains LLMs’ application in academic research to offer guidance for future research work (RO2). The five dimensions are literature reading, research design, academic writing, knowledge construction and admin assistance. 4. Limitations The limitations of this study are that the results drawn from this study should be considered preliminary and subject to further investigation. These conclusions are based on the subjective views of 50 PhD students who provided their perspectives on the use of LLMs in academic research, each potentially influenced by their individual backgrounds, motivations and expectations of ChatGPT in its early stages of development. Students were not involved in the data collection as ChatGPT is still in its early stages and students need time to become familiar with the technology. 5. Implication and Conclusion This study conducted semi-structured interviews with 50 PhD students, using thematic analysis to explore how they use LLMs in academic research. The results show that LLMs can assist literature reading, generate research design, optimize academic writing, improve knowledge construction and offer admin assistance. Based on these, a five-dimension framework of AIAAR is established to explain LLMs’ application in academic research to offer guidance for future research work. In general, it is of great significance and value to explore how doctoral students use LLMs for academic research, which can help doctoral students improve research efficiency, broaden research horizons, and enhance research quality. By using LLMs, doctoral students can more quickly access relevant literature, summarise research results, write papers, etc., thus improving research efficiency. Through LLMs, doctoral students can easily access research results and viewpoints from different disciplines, broaden their research horizons and promote interdisciplinary integration. By using LLMs, doctoral students can express their research ideas more precisely, avoid vagueness and imprecision in language, and thus improve the quality of research. The findings of this study can contribute to the understanding of the use of LLMs in academic research and to the future development of LLMs in academic research. First of all, it should be made clear that LLMs are designed to facilitate research, not to replace researchers, especially in paper writing. Over-reliance on AI for writing will undermine researchers’ creativity and hinder their learning process. Second, completely blocking the application of LLMs in academic research is not wise. There are still many scholars who hope that ChatGPT can develop into an academic research tool. Therefore, it is necessary to use LLMs rationally to improve the efficiency of research and expand the depth and breadth of academic research. What’s more, relevant departments should set up relevant policies and regulations to standardize the use of LLMs. The insights garnered from this research not only contribute to a deeper understanding of LLMs’ utilization in doctoral academic research but also provide a structured framework for optimizing their application. This framework is invaluable for researchers, educators, and policymakers seeking to harness the potential of AI-driven technologies to advance scholarly productivity, innovation, and knowledge dissemination in academia. Declarations Author Contribution Yihuan Yuan wrote the main manuscript text,Jamalludin Harun revised the manuscript. Data Availability This study involved conducting interviews with individuals and recording the interviews. To protect the privacy of the interviewees, the interview recordings are not publicly available. However, researchers may provide interview summaries or excerpts upon specific request for academic or research purposes. Please contact Yihuan Yuan ( [email protected] ) for more information and requests for access to interview data. References Antu, S. A., Chen, H., & Richards, C. K. (2023). Using LLMS (Large Language Model) to Improve Efficiency in Literature Review for Undergraduate Research. Bin-Hady, W. R. A., Al-Kadi, A., Hazaea, A., & Ali, J. K. M. (2023). Exploring the dimensions of ChatGPT in English language learning: A global perspective. Library Hi Tech , (ahead-of-print). Cotton, D. R., Cotton, P. A., & Shipway, J. R. (2023). Chatting and cheating: Ensuring academic integrity in the era of ChatGPT. Innovations in Education and Teaching International . De Kok, T. (2023). Generative LLMs and Textual Analysis in Accounting:(Chat) GPT as Research Assistant?. Available at SSRN . Dempere, J., Modugu, K., Hesham, A., & Ramasamy, L. K. (2023, September). The impact of ChatGPT on higher education. Frontiers in Education, 8,1206936. Devlin, J., Chang, M.-W., Lee, K., & Toutanova, K. (2018). BERT: Pre-training of deep bidirectional transformers for language understanding. arXiv. preprint arXiv: 1810.04805. Dijkstra, R., Genç, Z., Kayal, S., & Kamps, J. (2022). Reading comprehension quiz generation using generative pre-trained transformers. https://e.humanities.uva.nl/publications/2022/dijk_read22.pdf . Eysenbach, G. (2023). The role of ChatGPT, generative language models, and artificial intelligence in medical education: a conversation with ChatGPT and a call for papers. JMIR Medical Education, 9 (1), e46885. Fink, M. A., Bischoff, A., Fink, C. A., Moll, M., Kroschke, J., Dulz, L., … Weber, T. F. (2023). Potential of ChatGPT and GPT-4 for data mining of free-text CT reports on lung cancer. Radiology, 308 (3), e231362. Floridi, L., & Chiriatti, M. (2020). GPT-3: Its nature, scope, limits, and consequences. Minds and Machines, 30 (4), 681–694. Gabajiwala, E., Mehta, P., Singh, R., & Koshy, R. (2022). Quiz maker: Automatic quiz generation from text using NLP. In Futuristic trends in networks and computing technologies , 523–533. Heidt, A. (2023). ‘Arms race with automation’: Professors fret about AI-generated coursework. Nature. Imran, M., & Almusharraf, N. (2023). Analyzing the role of ChatGPT as a writing assistant at higher education level: A systematic review of the literature. Contemporary Educational Technology, 15 (4), ep464. Jin, H., Zhang, Y., Meng, D., Wang, J., & Tan, J. (2024). A Comprehensive Survey on Process-Oriented Automatic Text Summarization with Exploration of LLM-Based Methods. arXiv preprint arXiv:2403.02901. Kandpal, N., Deng, H., Roberts, A., Wallace, E., & Raffel, C. (2023). Large language models struggle to learn long-tail knowledge. In International Conference on Machine Learning , 15696–15707). PMLR. Kasneci, E., Seßler, K., Küchemann, S., Bannert, M., Dementieva, D., Fischer, F., … Kasneci, G. (2023). ChatGPT for good? On opportunities and challenges of large language models for education. Learning and individual differences, 103 , 102274. Kelly, D., Chen, Y., Cornwell, S. E., Delellis, N. S., Mayhew, A., Onaolapo, S., & Rubin, V. L. (2023). Bing Chat: The Future of Search Engines?. Proceedings of the Association for Information Science and Technology , 60 (1), 1007–1009. Lametti, D. (2022). AI could be great for college essays. slate.com. https://slate.com/technology/2022/12/chatgpt-college-essay-plagiarism.html Liu, G., & Ma, C. (2023). Measuring EFL learners’ use of ChatGPT in informal digital learning of English based on the technology acceptance model. Innovation in Language Learning and Teaching, 1–14. Liu, Y., Ott, M., Goyal, N., Du, J., Joshi, M., Chen, D., Levy, O., Lewis, M., Zettlemoyer, L., & Stoyanov, V. (2019). Roberta: A robustly optimized bert pretraining approach. arXiv. preprint arXiv :1907.11692. Lund, B. D., Wang, T., Mannuru, N. R., Nie, B., Shimray, S., & Wang, Z. (2023). ChatGPT and a new academic reality: Artificial Intelligence-written research papers and the ethics of the large language models in scholarly publishing. Journal of the Association for Information Science and Technology, 74 (5), 570–581. Radford, A., Narasimhan, K., Salimans, T., Sutskever, I., et al. (2018). Improving language understanding by generative pre-training Accessed: 2023-01-22. Raffel, C., Shazeer, N., Roberts, A., Lee, K., Narang, S., Matena, M., Zhou, Y., Li, W., Liu, P. J., et al. (2020). Exploring the limits of transfer learning with a unified textto-text transformer. Journal of Machine Learning Research, 21 (140), 1–67. Rasul, T., Nair, S., Kalendra, D., Robin, M., de Oliveira Santini, F., Ladeira, W. J., Sun, M., Day, I., Rather, R. A., & Heathcote, L. (2023). The role of ChatGPT in higher education: Benefits, challenges, and future research directions. Journal of Applied Learning and Teaching, 6 (1) Rosenfeld, A., & Lazebnik, T. (2024). Whose LLMS is it Anyway? Linguistic Comparison and LLMS Attribution for GPT-3.5, GPT-4 and Bard. arXiv preprint arXiv:2402.14533 . Scao, T. L., Fan, A., Akiki, C., Pavlick, E., Ili´c, S., Hesslow, D., Castagn´e, R., Luccioni, A. S., Yvon, F., Gall´e, M., et al. (2022). BLOOM: A 176B-parameter openaccess multilingual language model. arXiv. preprint arXiv:2211.05100. Suaverdez, J. B., & Suaverdez, U. V. (2023). Chatbots impact on academic writing. Global Journal of Business and Integral Security , (2). Urman, A., & Makhortykh, M. (2023). The Silence of the LLMs: Cross-Lingual Analysis of Political Bias and False Information Prevalence in ChatGPT, Google Bard, and Bing Chat, OSF Preprints . Yan, D. (2023). Impact of ChatGPT on learners in a L2 writing practicum: An exploratory investigation. Education and Information Technologies . Yang, J., Li, H. B., & Wei, D. (2023). The impact of ChatGPT and LLMs on medical imaging stakeholders: perspectives and use cases. Meta-Radiology, 100007. Yang, Y., Tang, Y., & Tam, K. Y. (2023). Investlm: A large language model for investment using financial domain instruction tuning. arXiv preprint arXiv:2309.13064 . Yang, Z., Dai, Z., Yang, Y., Carbonell, J., Salakhutdinov, R., & Le, Q. V. (2019). XLNet: Generalized Autoregressive Pretraining for Language Understanding. Advances in neural information processing systems, 32. preprint arXiv :1810.04805. Zhang, X., Yu, B., Yu, H., Lv, Y., Liu, T., Huang, F., … Li, Y. (2023). Wider and deeper llm networks are fairer llm evaluators. arXiv preprint arXiv:2308.01862. Additional Declarations No competing interests reported. Cite Share Download PDF Status: Under Review Version 1 posted Reviews received at journal 11 Jul, 2024 Reviewers agreed at journal 02 Jul, 2024 Reviews received at journal 07 Jun, 2024 Reviewers agreed at journal 28 May, 2024 Reviewers invited by journal 26 May, 2024 Editor assigned by journal 26 May, 2024 Editor invited by journal 07 May, 2024 Submission checks completed at journal 06 May, 2024 First submitted to journal 28 Apr, 2024 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-4337026","acceptedTermsAndConditions":true,"allowDirectSubmit":false,"archivedVersions":[],"articleType":"Article","associatedPublications":[],"authors":[{"id":299889315,"identity":"61e1d45b-70e8-4f9e-b128-cfc467a2eec8","order_by":0,"name":"Yihuan Yuan","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAAxUlEQVRIiWNgGAWjYBACfv7mAwcSDCTq29gbiNQiOeNY4oEPBTaMfTwHiNRicCDH+OCMD2mM8yQSiNZyxuAwj8FhZjbJxxtvMNTYRBN22OG2ApAWNjbptGILhmNpuQ2EtPAdOLwBpIWHTTrHTIKx4TBhLQzA4AJpkWCTPEOkFoEDKQYHZxikGbBJ8BCpBRjICQc+GNgksPEA/ZJAjF+AUXn4Q8IfiQT59sMbb3yosSHCL0jAgOioQdJCqo5RMApGwSgYGQAA5c5D1/qM9VoAAAAASUVORK5CYII=","orcid":"","institution":"University of Technology Malaysia","correspondingAuthor":true,"prefix":"","firstName":"Yihuan","middleName":"","lastName":"Yuan","suffix":""},{"id":299889317,"identity":"d4f5799a-d141-469d-9a42-58795dcfff98","order_by":1,"name":"Jamalludin Harun","email":"","orcid":"","institution":"University of Technology Malaysia","correspondingAuthor":false,"prefix":"","firstName":"Jamalludin","middleName":"","lastName":"Harun","suffix":""}],"badges":[],"createdAt":"2024-04-28 08:40:06","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-4337026/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-4337026/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":56229511,"identity":"65097a48-ea01-4f29-85ad-76b035d9d46c","added_by":"auto","created_at":"2024-05-10 06:56:10","extension":"png","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":75545,"visible":true,"origin":"","legend":"\u003cp\u003eFive-dimension framework of AIAAR\u003c/p\u003e","description":"","filename":"1.png","url":"https://assets-eu.researchsquare.com/files/rs-4337026/v1/10a5e3a99cd232bd9cd07321.png"},{"id":56229512,"identity":"bb5e6dfa-61ed-44bd-a8a5-95aa37ac0829","added_by":"auto","created_at":"2024-05-10 06:56:15","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":549950,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-4337026/v1/d767cca2-a5ad-4260-8aa2-35ea8cbec792.pdf"}],"financialInterests":"No competing interests reported.","formattedTitle":"Empowering Doctoral Academic Research: Artificial Intelligence-driven Insights from Large Language Models","fulltext":[{"header":"1. Introduction","content":"\u003cp\u003eArtificial Intelligence (AI) has had a significant impact on various facets of contemporary society in recent years, offering numerous benefits that are transforming various aspects of modern life (Lund et al., \u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e2023\u003c/span\u003e). Its development is breaking down barriers between different areas of application, extending its potential to a wide range. The integration of AI into education represents a continuous evolution in this field (Imran \u0026amp; Almusharraf, \u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e2023\u003c/span\u003e).\u003c/p\u003e \u003cp\u003eLarge language models (LLMs) represent a notable advance in AI field (Kasneci et al., \u003cspan citationid=\"CR16\" class=\"CitationRef\"\u003e2023\u003c/span\u003e). These models refer to deep learning architectures that have been trained on large textual datasets, enabling them to produce natural language text and facilitate the understanding of textual meaning. LLMs can handle a variety of natural language tasks, including text categorization, question answering and dialogue, making them an important component of artificial intelligence (Kandpal et al., \u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e2023\u003c/span\u003e).\u003c/p\u003e \u003cp\u003eNumerous LLMs have emerged in recent years, including GPT (Radford et al., \u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e2018\u003c/span\u003e), BERT (Devlin et al., \u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e2018\u003c/span\u003e), XLNet (Yang et al., \u003cspan citationid=\"CR32\" class=\"CitationRef\"\u003e2019\u003c/span\u003e), T5 (Raffel et al., \u003cspan citationid=\"CR23\" class=\"CitationRef\"\u003e2020\u003c/span\u003e), RoBERTa (Liu et al., \u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e2019\u003c/span\u003e), GPT-3 (Floridi \u0026amp; Chiriatti, \u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e2020\u003c/span\u003e; Scao et al., \u003cspan citationid=\"CR26\" class=\"CitationRef\"\u003e2022\u003c/span\u003e), Bing Chat (Kelly et al., \u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e2023\u003c/span\u003e), Google Bard (Urman \u0026amp; Makhortykh, 2023), and the widely recognized GPT-3.5 and GPT-4 (Rosenfeld \u0026amp; Lazebnik, \u003cspan citationid=\"CR25\" class=\"CitationRef\"\u003e2024\u003c/span\u003e). ChatGPT, an AI-driven conversational agent powered by a large language models (LLMs) trained on rich Internet text data, is expected to address numerous limitations of previous chatbot technology and reshape learning dynamics (Heidt, \u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e2023\u003c/span\u003e).\u003c/p\u003e \u003cp\u003eBuilt on a transformer architecture, LLMs models are pre-trained on massive text datasets to mimic human text generation, respond to queries, assist with translation and summarization, and execute various natural language processing (NLP) functions via a unified pre-training and fine-tuning process.\u003c/p\u003e \u003cp\u003eLarge language models are currently used in many areas and have had positive results. Advancements in LLMs have paved the way for their integration into educational environments, providing opportunities to improve teaching and learning. A notable area is the creation and development of educational content, where scientists have used LLMs to create interactive learning resources such as quizzes and flashcards. These resources are expected to enhance student engagement and learning outcomes (Dijkstra et al., 20\u0026ndash;22; Gabajiwala et al., 20\u0026ndash;22).\u003c/p\u003e \u003cp\u003eWhat\u0026rsquo;s more, LLMs offer several benefits, such as assisting in the generation of academic writing and research (Bin-Hady et al., \u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2023\u003c/span\u003e). For university students in particular, these models serve as valuable tools for research and writing assignments. They facilitate tasks such as creating summaries and outlines of texts, allowing students to quickly grasp the core ideas of a text and structure their thoughts for writing (Kasneci et al., \u003cspan citationid=\"CR16\" class=\"CitationRef\"\u003e2023\u003c/span\u003e).\u003c/p\u003e \u003cp\u003eThe ever-increasing volume and intricacy of academic literature pose considerable challenges for researchers, particularly doctoral students, in navigating through vast amounts of information, designing robust research methodologies, and synthesizing knowledge effectively (Dempere et al, \u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e2023\u003c/span\u003e). In response to these challenges, the integration of LLMs into doctoral research holds immense potential to enhance efficiency, productivity, and the quality of academic outputs. By harnessing LLMs, doctoral students can overcome research challenges more effectively and advance knowledge in their respective fields (Zhang et al., \u003cspan citationid=\"CR33\" class=\"CitationRef\"\u003e2023\u003c/span\u003e; Jin et al., \u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e2024\u003c/span\u003e).\u003c/p\u003e \u003cp\u003eHowever, up to now, there are relatively few discussions about how doctoral students can effectively use LLMs for academic research. Namely, there is still a research gap in this field, and more in-depth investigation and exploration are still needed to reveal the practical application and potential value of LLMs in doctoral academic research.\u003c/p\u003e \u003cp\u003eIn response to this research gap, a qualitative investigation is conducted in this study. This exploratory study attempts to investigate how doctoral students use LLMs in academic research. Specifically, this study aims to establish a framework to explain the application of LLMs in academic research.\u003c/p\u003e \u003cp\u003eThe following research objectives guide the study:\u003c/p\u003e \u003cp\u003e \u003col\u003e \u003cspan\u003e \u003cli\u003e \u003cp\u003eTo explore how the large language models (LLMs) are used in doctoral academic research;\u003c/p\u003e \u003c/li\u003e \u003c/span\u003e \u003cspan\u003e \u003cli\u003e \u003cp\u003eTo establish a framework of LLMs\u0026rsquo; application in academic research to offer guidance for future research work.\u003c/p\u003e \u003c/li\u003e \u003c/span\u003e \u003c/ol\u003e \u003c/p\u003e"},{"header":"2. Methodology","content":"\u003cp\u003eThis research adopts a qualitative approach, predominantly drawing insights from interviews with 50 PhD students. Grounded theory underpins the research design, as it holds significance in qualitative inquiry (Creswell, 2013). Grounded theory posits that data collection and analysis are iterative processes, driven by research objectives while aiming to achieve saturation. In line with it, semi-structured interviews are employed as the primary method of data collection.\u003c/p\u003e \u003cp\u003eInterview questions:\u003c/p\u003e\n\u003cp\u003e1. How do you use LLMs in academic research?\u003c/p\u003e\n\u003cp\u003e2. What is the value of LLMs to doctoral students’ academic research?\u003c/p\u003e\n\u003cdiv id=\"Sec5\" class=\"Section2\"\u003e \u003ch2\u003e2.1 Participants\u003c/h2\u003e \u003cp\u003eThe study involved 50 PhD students (shown in Table\u0026nbsp;\u003cspan refid=\"Tab1\" class=\"InternalRef\"\u003e1\u003c/span\u003e) from a variety of academic backgrounds and countries who were actively using LLMs in their academic research, with the recruitment period spanning from February 1, 2023, to February 1, 2024. These participants were carefully selected to represent a wide range of disciplines and geographical regions, ensuring a broad understanding of the different ways in which LLMs were used in academic research in different contexts. Through the inclusion of a diverse group of students, the study aimed to capture a wide range of perspectives and experiences regarding the use of LLMs in academic environments, and to highlight their effectiveness and challenges in different academic disciplines and cultural contexts.\u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab1\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 1\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003eInformation of participants\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"3\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003eItem\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003eCategory\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003ePopulation\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\" morerows=\"1\" rowspan=\"2\"\u003e \u003cp\u003eAge\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e25\u0026ndash;35\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003en\u0026thinsp;=\u0026thinsp;40\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e35\u0026ndash;45\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003en\u0026thinsp;=\u0026thinsp;10\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\" morerows=\"1\" rowspan=\"2\"\u003e \u003cp\u003eGender\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eMale\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003en\u0026thinsp;=\u0026thinsp;25\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eFemale\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003en\u0026thinsp;=\u0026thinsp;25\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\" morerows=\"4\" rowspan=\"5\"\u003e \u003cp\u003eResearch area\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eEnglish\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003en\u0026thinsp;=\u0026thinsp;10\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eEducation\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003en\u0026thinsp;=\u0026thinsp;10\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eDig data and AI\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003en\u0026thinsp;=\u0026thinsp;10\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eMechanical Engineering\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003en\u0026thinsp;=\u0026thinsp;10\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eMaterials science\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003en\u0026thinsp;=\u0026thinsp;10\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\" morerows=\"4\" rowspan=\"5\"\u003e \u003cp\u003eStudy country\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eMalaysia\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003en\u0026thinsp;=\u0026thinsp;30\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eJapan\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003en\u0026thinsp;=\u0026thinsp;5\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eUK\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003en\u0026thinsp;=\u0026thinsp;5\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eAmerica\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003en\u0026thinsp;=\u0026thinsp;5\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eItaly\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003en\u0026thinsp;=\u0026thinsp;5\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003cp\u003eParticipants were invited to participate in the interview via WeChat, accompanied by a cover letter. This letter explained the purpose of the study and assured that their responses would be kept anonymous and confidential. It further highlighted the voluntary nature of their participation, ensuring they knew they could opt-out at any time if they were uneasy during the interview. Written informed consent was secured from those who chose to take part.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec6\" class=\"Section2\"\u003e \u003ch2\u003e2.2 Procedures and data analysis\u003c/h2\u003e \u003cp\u003eAfter selecting 50 PhD students, the researchers had individual interviews with each participant via Wechat. Each interview session lasted approximately 30 minutes, allowing sufficient time for in-depth discussion. Throughout the interviews, the researchers used audio recording to accurately capture the dialogue.\u003c/p\u003e \u003cp\u003eWhen each interview was completed, transcription of the audio into text format was carried out using speech recognition technology. The transcription process ensured that all spoken contents were precisely preserved for subsequent analysis.\u003c/p\u003e \u003cp\u003eFor data analysis, thematic analysis was used to analyze the data from the interviews on NVivo. Braun and Clarke\u0026rsquo;s thematic analysis (2022) was also employed in this study, which is an iterative process consisting of six steps: (1) becoming familiar with the data, (2) generating codes, (3) generating themes, (4) reviewing themes, (5) defining and naming themes, and (6) locating exemplars. This analytical approach involved the systematic identification and analysis of recurring themes, patterns and insights within the interview data. By immersing themselves in the textual data, the researchers gained a nuanced understanding of the participants\u0026rsquo; perspectives, experiences and practices in relation to their use of LLMs in academic research.\u003c/p\u003e \u003c/div\u003e"},{"header":"3. Results and Discussions","content":"\u003cp\u003eThis section provides a comprehensive analysis of the research findings and discussions, structured according to the predetermined research objectives. Through an in-depth analysis of the data collected, the following subsections explore the key findings of the study and their significance in addressing the research objectives.\u003c/p\u003e \u003cp\u003eThe themes found in the coding process are shown in Table\u0026nbsp;\u003cspan refid=\"Tab2\" class=\"InternalRef\"\u003e2\u003c/span\u003e.\u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab2\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 2\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003eThe coding process\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"4\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003eNO.\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003eTheme\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003eCode\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c4\"\u003e \u003cp\u003eExemplars\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\" morerows=\"2\" rowspan=\"3\"\u003e \u003cp\u003e1\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\" morerows=\"2\" rowspan=\"3\"\u003e \u003cp\u003eLiterature reading\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eExtract main content\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e\u003cb\u003eInterviewee 2\u003c/b\u003e: \u003cem\u003eAsk it (ChatGPT) about the main idea of the literature or ask him to summarize it, and I will get an answer immediately.\u003c/em\u003e\u003c/p\u003e \u003cp\u003e\u003cb\u003eInterviewee 26\u003c/b\u003e: \u003cem\u003eChatGPT helps me summarize the academic paper in a way that is easy to understand\u0026hellip;\u003c/em\u003e\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eProvide research topics\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e\u003cb\u003eInterviewee 21\u003c/b\u003e: \u003cem\u003e\u0026hellip;give the topic and time range of the paper, ChatGPT will automatically give some representative papers, based on the paper, I can get a general idea of the relevant field.\u003c/em\u003e\u003c/p\u003e \u003cp\u003e\u003cb\u003eInterviewee 35\u003c/b\u003e: \u003cem\u003e(ChatGPT) quickly and accurately grasp a large number of papers and background knowledge on a certain topic, based on which I can find my own research topic.\u003c/em\u003e\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eMake reading convenient\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e\u003cb\u003eInterviewee 47\u003c/b\u003e: \u003cem\u003eMost of the time, I need to read several papers, especially when two of the same methods are compared, it is necessary to compare several papers at the same time. So I can use ChatGPT to read two articles at the same time and compare them.\u003c/em\u003e\u003c/p\u003e \u003cp\u003e\u003cb\u003eInterviewee17\u003c/b\u003e: \u003cem\u003eI can have multiple rounds of interaction with ChatGPT to discuss the issues in the paper. In this case, it is entirely possible to think of it as a supervisor, can ask questions, can answer, can discuss, can debate.\u003c/em\u003e\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e2\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eResearch design\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eGenerate research design\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e\u003cb\u003eInterviewee 23\u003c/b\u003e: \u003cem\u003eChatGPT can be used to generate experimental designs or research protocols. For example, a teaching and research department has designed two sets of training software and three training modes, claiming that the emotional intelligence of teenagers can be improved through practice. Now design an experiment to test its effectiveness.\u003c/em\u003e\u003c/p\u003e \u003cp\u003e\u003cb\u003eInterviewee 13\u003c/b\u003e: \u003cem\u003eIn the research, I can use chatGPT and Bing Chat to generate interview questions, questionnaire questions and examination papers.\u003c/em\u003e\u003c/p\u003e \u003cp\u003e\u003cb\u003eInterviewee 40\u003c/b\u003e:\u003cem\u003e\u0026hellip;through the conversation with chatGPT, the rationality of the experimental design was discussed, the experimental parameters were determined, and the experimental results were interpreted.\u003c/em\u003e\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\" morerows=\"2\" rowspan=\"3\"\u003e \u003cp\u003e3\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\" morerows=\"2\" rowspan=\"3\"\u003e \u003cp\u003eAcademic writing\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eGenerate writing ideas\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e\u003cb\u003eInterviewee 30\u003c/b\u003e: \u003cem\u003eWhen I input the research topic, Google Bard can provide some initial ideas, arguments, paper structure, etc.\u003c/em\u003e\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003ePolish writing\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e\u003cb\u003eInterviewee 18\u003c/b\u003e: \u003cem\u003eChatGPT helps me refine the details, such as adding specific information, data or examples to the article to support the argument and enrich the content.\u003c/em\u003e\u003c/p\u003e \u003cp\u003e\u003cb\u003eInterviewee 25\u003c/b\u003e: \u003cem\u003eAs my native language is not English, there will be unprofessional expressions when writing papers. Using chatGPT can help me replace those unprofessional or too colloquial phrases and words, so as to improve the professionalism and formality of my writing.\u003c/em\u003e\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eAnalyse and visualize data\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e\u003cb\u003eInterviewee 44\u003c/b\u003e: \u003cem\u003eChatGPT-4 can generate charts and make content visual by adding charts and visual elements. Graphic combination provides more clearly demonstrate the point of view.\u003c/em\u003e\u003c/p\u003e \u003cp\u003e\u003cb\u003eInterviewee 17\u003c/b\u003e: \u003cem\u003eChatGPT-4 can receive various types of data and conduct data analysis through natural language. Such as extracting the information in the graph, conducting topic analysis, and generating data analysis reports.\u003c/em\u003e\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\" morerows=\"1\" rowspan=\"2\"\u003e \u003cp\u003e4\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\" morerows=\"1\" rowspan=\"2\"\u003e \u003cp\u003eKnowledge construction\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eOffer expertise\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e\u003cb\u003eInterviewee 12\u003c/b\u003e: \u003cem\u003eChatGPT has completely changed the way code is developed. It can empower programming, and when I encounter obstacles in programming, I can find answers in chatGPT.\u003c/em\u003e\u003c/p\u003e \u003cp\u003e\u003cb\u003eInterviewee 19\u003c/b\u003e: \u003cem\u003eIn the research, I will encounter some unfamiliar research terms, and it will take a long time to search on the Internet. If I input the question into ChatGPT, I can get one answer quickly, greatly improving the learning efficiency.\u003c/em\u003e\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003ePromote\u003c/p\u003e \u003cp\u003esubject area\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e\u003cb\u003eInterviewee 21\u003c/b\u003e: \u003cem\u003eIf you can\u0026rsquo;t explain your research project clearly to the children, it means that your understanding is not deep enough. ChatGPT helps us understand scientific concepts and projects from the perspective of the general public.\u003c/em\u003e\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e5\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eAdmin assistance\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eprovide communication\u003c/p\u003e \u003cp\u003eassistance\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e\u003cb\u003eInterviewee 2\u003c/b\u003e: \u003cem\u003eChatGPT can play an important auxiliary role in academic communication, especially in English mail. For students whose native language is not English, sending letters to foreign tutors has always been a major problem. Now, with the help of ChatGPT, an honest, well-structured letter can be generated by typing a single sentence.\u003c/em\u003e\u003c/p\u003e \u003cp\u003e\u003cb\u003eInterviewee 10\u003c/b\u003e: \u003cem\u003ePublishing a paper is a long process, and sometimes I don't receive a reply long after submitting it. I use ChatGPT to generate emails urging review. ChatGPT understands the writing style so much, which not only shows that reviewing the paper is very important to me but also considers the position of the reviewers.\u003c/em\u003e\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab3\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 3\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003eFrequency of codes\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"3\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003eCode\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003eFrequency\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003eNumber of participants\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eExtract main content\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003en\u0026thinsp;=\u0026thinsp;43\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003en\u0026thinsp;=\u0026thinsp;43\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eProvide research topics\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003en\u0026thinsp;=\u0026thinsp;27\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003en\u0026thinsp;=\u0026thinsp;27\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eMake reading convenient\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003en\u0026thinsp;=\u0026thinsp;44\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003en\u0026thinsp;=\u0026thinsp;40\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eGenerate research design\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003en\u0026thinsp;=\u0026thinsp;35\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003en\u0026thinsp;=\u0026thinsp;35\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eGenerate writing ideas\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003en\u0026thinsp;=\u0026thinsp;38\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003en\u0026thinsp;=\u0026thinsp;38\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003ePolish writing\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003en\u0026thinsp;=\u0026thinsp;55\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003en\u0026thinsp;=\u0026thinsp;50\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eAnalyse and visualize data\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003en\u0026thinsp;=\u0026thinsp;20\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003en\u0026thinsp;=\u0026thinsp;20\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eOffer expertise\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003en\u0026thinsp;=\u0026thinsp;15\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003en\u0026thinsp;=\u0026thinsp;15\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003ePromote subject area\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003en\u0026thinsp;=\u0026thinsp;19\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003en\u0026thinsp;=\u0026thinsp;19\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eprovide communication assistance\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003en\u0026thinsp;=\u0026thinsp;8\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003en\u0026thinsp;=\u0026thinsp;8\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003cp\u003eFive themes presented in Table\u0026nbsp;\u003cspan refid=\"Tab2\" class=\"InternalRef\"\u003e2\u003c/span\u003e explain how LLMs influence doctoral students\u0026rsquo; academic research (RO1). Table\u0026nbsp;\u003cspan refid=\"Tab3\" class=\"InternalRef\"\u003e3\u003c/span\u003e presents the frequency of mention during the interviews and the number of participants who highlighted specific aspects. The results show LLMs contribute to academic research in five key areas: literature review, research design, academic writing, knowledge construction, and admin assistance. These aspects are structured into a five-dimensional model for academic research (RO 2).\u003c/p\u003e \u003cdiv id=\"Sec8\" class=\"Section2\"\u003e \u003ch2\u003e3.1 Literature reading\u003c/h2\u003e \u003cp\u003eLLMs assist literature reading from 3 aspects: extract main content, provide research topics, and make reading convenient. Firstly, LLMs can extract the main content of the literature immediately. Just like Interviewee 2 said, \u003cem\u003eChatGPT can summarize the main idea of the literature immediately\u003c/em\u003e. What\u0026rsquo;s more, Interviewee 6 mentioned \u003cem\u003eChatGPT provides an easy way to make academic papers understood\u003c/em\u003e. Secondly, LLMS can provide research topics for researchers. Just like interviewee 11 and 15 mentioned based on the paper or topics found by ChatGPT, they could have ideas about their own research topics. Third, LLMs can make reading convenient. According to Interviewee 7, \u003cem\u003eChatGPT provides the function of reading and comparing two papers at the same time\u003c/em\u003e. In addition, Interviewee 17 mentioned \u003cem\u003emultiple rounds of interaction can be conducted on ChatGPT\u003c/em\u003e. This greatly improves the efficiency of reading.\u003c/p\u003e \u003cp\u003eThese findings are aligned with some previous research and enrich them in some degree. LLMs have become valuable tools for academic reading, providing comprehensive support throughout the research process. Antu and Richards (2024) found that ChatGPT, for example, can effectively organize literature, presenting a structured overview that expedites the reading process. LLMs excel at drawing out essential information from extensive sources, skilfully analyzing and summarising large volumes of text to extract key concepts, arguments and findings according to the needs of users. LLMs condense lengthy passages into succinct summaries, saving time and effort in reading. Eysenbach\u0026rsquo;s (\u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e2023\u003c/span\u003e) study highlights their capability to generate medical research summaries, aiding medical students\u0026rsquo; comprehension of key findings and implications. This feature is particularly useful for academics who need to sift through a large amount of literature to inform their research or stay up-to-date with the latest developments in their field.\u003c/p\u003e \u003cp\u003eFurthermore, LLMs are essential for proposing potential research directions by reviewing the existing literature. They analyze academic articles to identify gaps, controversies, or emerging themes within a particular field, providing valuable insights that inspire researchers to explore new avenues of inquiry or develop research questions to address unmet needs or unresolved issues. LLMs serve as a source of inspiration and guidance, allowing researchers to undertake innovative and meaningful research that contributes to the advancement of knowledge. This result supports the finding in Antu et al.(2023)\u0026rsquo; research that ChatGPT may help researchers identify trends and gaps in current research.\u003c/p\u003e \u003cp\u003eIn a word, LLMs enhance the accessibility and convenience of the research reading process by offering a range of features and capabilities. This enables researchers to interact with literature more effectively and efficiently, accommodating their diverse requirements and inclinations in academic reading.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec9\" class=\"Section2\"\u003e \u003ch2\u003e3.2 Research design\u003c/h2\u003e \u003cp\u003eLLMs are also helpful in research design. According to Interviewee 3, \u003cem\u003eChatGPT can be used to generate experimental designs or research protocols\u003c/em\u003e. According to Interviewee 13, \u003cem\u003eChatGPT and Bing Chat can generate interview questions, questionnaire questions and examination papers\u003c/em\u003e. Besides these, Interviewee 50 mentioned, \u003cem\u003ethrough the conversation with ChatGPT, the rationality of the research design can be justified.\u003c/em\u003e\u003c/p\u003e \u003cp\u003eLLMs can significantly support the research design process by assisting researchers at every stage. They can help to explore and refine research inquiries, identify relevant variables and factors, and select appropriate methodologies and approaches. Similarly, De Kok (\u003cspan citationid=\"CR4\" class=\"CitationRef\"\u003e2023\u003c/span\u003e) mentioned GPT-4 and ChatGPT were used to select an approach first and then choose an appropriate model so as to eliminate discrepancies in the research approach. Additionally, LLMs can simplify literature reviews by summarizing existing research, identifying areas lacking coverage, and suggesting potential paths for further exploration. Leveraging synthesized knowledge and insights from extensive textual data, LLMs can also assist in formulating hypotheses and defining research objectives.\u003c/p\u003e \u003cp\u003eThis finding can be interpreted by shedding light on ChatGPT is efficient in generating text, especially creative writing, and can produce human-like academic and professional tasks (Rasul et al., \u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e2023\u003c/span\u003e; Suaverdez \u0026amp; Suaverdez, \u003cspan citationid=\"CR27\" class=\"CitationRef\"\u003e2023\u003c/span\u003e). The transformative impact of AI on the research landscape is highlighted by the discovery that LLMs contribute to the design of research and make it possible to justify the rationality of research. By using LLMs, researchers can improve the accuracy, efficiency, and credibility of their research, leading to innovation and progress in various fields.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec10\" class=\"Section2\"\u003e \u003ch2\u003e3.3 Academic writing\u003c/h2\u003e \u003cp\u003eLLMs can optimize academic writing, including generate outlines, polish writing as well as analyze and visualize data. For example, \u003cem\u003eGoogle Bard can provide some initial ideas, arguments, paper structure\u003c/em\u003e (Interviewee 10). \u003cem\u003eChatGPT can be used to refine the details, such as adding specific information, data or examples to the article to support the argument and enrich the content\u003c/em\u003e (Interviewee 18). What\u0026rsquo;s more, for EFL (English as foreign) learners, \u003cem\u003eChatGPT can be used to replace those unprofessional or too colloquial phrases and words, so as to improve the professionalism and formality of writing\u003c/em\u003e (Interviewee 5). In addition, for data analysis, \u003cem\u003eChatGPT-4 can generate charts and make content visual by adding charts and visual elements\u003c/em\u003e (Interviewee 4). \u003cem\u003eIt can also receive various types of data and conduct data analysis through natural language\u003c/em\u003e (Interviewee 17).\u003c/p\u003e \u003cp\u003eLLMs are adept at generating outlines for academic writing. They automatically arrange and organize complex information into coherent structures. Researchers can input their research objectives, main points, and supporting evidence into LLMs, and then generate well-structured outlines outlining the paper\u0026rsquo;s logical progression. This not only saves researchers\u0026rsquo; time during the initial planning stages of writing but also ensures that their papers adhere to a clear and organized format. Generated outlines by LLMs can act as a guide for writing, steering researchers through the writing process and ensuring comprehensive coverage of essential aspects in their paper. This finding supports the idea that ChatGPT can assist students in generating new ideas for their writing tasks by suggesting topics, themes, and perspectives they may not have otherwise considered (Kasne et al., 2023).\u003c/p\u003e \u003cp\u003eLLMs are proficient in providing recommendations to improve the clarity, coherence, and conciseness of academic writing. It is crucial to maintain objectivity and impartiality by avoiding biased or emotional language and clearly indicating subjective assessments. LLMs can refine language, strengthen arguments, and enhance the overall readability of academic texts. They can detect grammatical errors, improve sentence structure, suggest synonyms for repetitive terms, and propose stylistic enhancements to improve the quality of academic writing. It is imperative to ensure a clear and coherent structure, use precise language, and adhere to formal conventions. The creation of ChatGPT increases the potential for producing accurate and consistent content (Imran \u0026amp; Almusharraf, \u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e2023\u003c/span\u003e). LLMs can be used by researchers to refine their writing and generate polished and professional manuscripts that captivate and resonate with readers. The substantial utility of ChatGPT in effectively addressing various writing tasks has been underscored by global academic and scientific discourse (Sallam, 2023).\u003c/p\u003e \u003cp\u003eLLMs are valuable tools for data-driven research writing as they can analyze and visualize data. Researchers can input raw data or descriptive statistics into LLMs. LLMs can then generate insightful analyses, interpret findings, and create visualizations such as charts, graphs, and tables to illustrate key trends and patterns. LLMs can assist researchers in discovering concealed insights in their data, identifying correlations and relationships, and communicating complex findings clearly and compellingly. For example, Fink\u0026rsquo;s group (2023) found that the LLMs\u0026mdash;ChatGPT and GPT-4 perform well in data mining and labeling oncologic phenotypes. Furthermore, LLMs can aid in producing data-driven narratives that contextualize research findings within the wider academic discourse, thereby enhancing the credibility and impact of research publications.\u003c/p\u003e \u003cp\u003eThrough the use of chatbots, researchers can optimize the writing process, increase the quality and effectiveness of their publications, and contribute to the advancement of scientific knowledge across disciplines. Consistent with this, Imran and Almusharraf (\u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e2023\u003c/span\u003e) also found that chatbots are useful tools for facilitating and supporting the academic process.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec11\" class=\"Section2\"\u003e \u003ch2\u003e3.4 Knowledge Construction\u003c/h2\u003e \u003cp\u003eLLMs can assist in knowledge construction. In the process of research, it is very common to encounter unfamiliar or difficult professional questions. Asking others takes a long time and is not conducive to cultivating the exploration spirit of doctoral students. Searching on the Internet is the choice of many doctoral students, but the information on the Internet is too complex, and opinions on the same question or concept are varied, so it is difficult to find a definite answer in a short time. LLMs can help to solve this problem. Just like Interviewee 19 said \u003cem\u003eIf I input the question into ChatGPT, I can get one answer quickly, greatly improving the learning efficiency. On the other hand, LLMS can be used to promote science.\u003c/em\u003e According to Interviewee 1, \u003cem\u003eIf you can\u0026rsquo;t explain your research project clearly to the children, it means that your understanding is not deep enough. Fortunately, ChatGPT helps us understand scientific concepts and projects from the perspective of the general public.\u003c/em\u003e\u003c/p\u003e \u003cp\u003eLLMs are trained on large amounts of text data from diverse sources, allowing them to acquire comprehensive knowledge across various disciplines, such as science, humanities, social sciences, technology, and medicine. LLMs possess a broad range of knowledge, making them a valuable resource for researchers, educators, students, and professionals seeking subject matter expertise. Imran and Almusharraf\u0026rsquo;s (\u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e2023\u003c/span\u003e) findings support the notion that ChatGPT efficiently comprehends and interprets user input commands to generate appropriate answers, including examples related to any subject. Namely, LLMs can generate relevant and informative responses based on their understanding of the subject matter. This instantaneous access to information enables users to obtain insights, explanations, and perspectives on complex topics in real time, facilitating research, decision-making, and problem-solving processes.\u003c/p\u003e \u003cp\u003eLLMs possess a vast knowledge base and demonstrate a deep understanding of specific domains. Their exposure to specialized literature and texts within particular fields enables them to develop detailed insights, understand terminological nuances, and comprehend contexts deeply within those subject areas. This profound understanding equips LLMs to provide detailed explanations, analyses, and interpretations on complex subjects, similar to experts in those respective domains. Yang et al.(2023) also agreed that LLM could provide a detailed and understandable explanation.\u003c/p\u003e \u003cp\u003eIn conclusion, the proficiency of LLMs in knowledge construction marks a substantial transformation in how information is accessed, synthesized, and shared in the digital era. Leveraging the capabilities of LLMs enables individuals and institutions to tap into extensive expertise spanning various fields, expedite research and creativity, and tackle intricate societal issues.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec12\" class=\"Section2\"\u003e \u003ch2\u003e3.5 Admin assistance\u003c/h2\u003e \u003cp\u003eLLMs can serve as an Admin assistant for every doctoral student. For example, it can be used to write E-mails. For students whose native language is not English, sending letters to foreign supervisors has always been a major problem. \u003cem\u003eWith the help of ChatGPT, a well-structured letter can be generated by typing a single sentence\u003c/em\u003e (Interviewee 2). In addition, \u003cem\u003eLLMs can generate well-formulated emails that help authors ask about the review progress of publishing\u003c/em\u003e (Interviewee 9).\u003c/p\u003e \u003cp\u003eLLMs can be of great assistance to doctoral students in managing their academic communication more effectively. They can help with organizing emails, scheduling meetings, and prioritizing tasks related to research, collaborations, and academic commitments. LLMs can save time and reduce cognitive load by providing students with concise overviews of important information and action items. According to Cotton et al. (\u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e2023\u003c/span\u003e), LLMs have the potential to enhance work pace and efficiency in daily routines.\u003c/p\u003e \u003cp\u003eLLMs can assist doctoral students in preparing for academic conferences and presentations by generating presentation slides, summarising key points, and providing speaking prompts. They can also help organize conference schedules, coordinate travel arrangements, and facilitate networking with other researchers and scholars in the field. In addition, LLMs can generate answers to frequently asked questions and anticipate potential questions during Q\u0026amp;A sessions.\u003c/p\u003e \u003cp\u003eIn conclusion, the invention of ChatGPT has the potential to streamline academic research processes by significantly reducing time and effort. Its capacity to generate coherent and well-structured text across various subjects can alleviate the workload for students and educators, allowing them to allocate more time to other tasks (Lund et al., \u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e2023\u003c/span\u003e; Yan, \u003cspan citationid=\"CR29\" class=\"CitationRef\"\u003e2023\u003c/span\u003e). The idea of employing LLMs as administrative aids for doctoral students in academic communication shows promise in enhancing efficiency, productivity, and collaboration throughout the research journey. With the help of LLMs, doctoral candidates can benefit from personalized assistance, access up-to-date information, and utilize advanced communication tools to navigate academic pursuits more efficiently.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec13\" class=\"Section2\"\u003e \u003ch2\u003e3.6 Five-dimension framework of AI-assisted Academic Research (AIAAR)\u003c/h2\u003e \u003cp\u003eBased on the above findings, a five-dimension framework of AIAAR (shown in Fig.\u0026nbsp;1) is established, which explains LLMs\u0026rsquo; application in academic research to offer guidance for future research work (RO2). The five dimensions are literature reading, research design, academic writing, knowledge construction and admin assistance.\u003c/p\u003e"},{"header":"4. Limitations","content":"\u003cp\u003eThe limitations of this study are that the results drawn from this study should be considered preliminary and subject to further investigation. These conclusions are based on the subjective views of 50 PhD students who provided their perspectives on the use of LLMs in academic research, each potentially influenced by their individual backgrounds, motivations and expectations of ChatGPT in its early stages of development. Students were not involved in the data collection as ChatGPT is still in its early stages and students need time to become familiar with the technology.\u003c/p\u003e"},{"header":"5. Implication and Conclusion","content":"\u003cp\u003eThis study conducted semi-structured interviews with 50 PhD students, using thematic analysis to explore how they use LLMs in academic research. The results show that LLMs can assist literature reading, generate research design, optimize academic writing, improve knowledge construction and offer admin assistance. Based on these, a five-dimension framework of AIAAR is established to explain LLMs\u0026rsquo; application in academic research to offer guidance for future research work.\u003c/p\u003e \u003cp\u003eIn general, it is of great significance and value to explore how doctoral students use LLMs for academic research, which can help doctoral students improve research efficiency, broaden research horizons, and enhance research quality. By using LLMs, doctoral students can more quickly access relevant literature, summarise research results, write papers, etc., thus improving research efficiency. Through LLMs, doctoral students can easily access research results and viewpoints from different disciplines, broaden their research horizons and promote interdisciplinary integration. By using LLMs, doctoral students can express their research ideas more precisely, avoid vagueness and imprecision in language, and thus improve the quality of research.\u003c/p\u003e \u003cp\u003eThe findings of this study can contribute to the understanding of the use of LLMs in academic research and to the future development of LLMs in academic research. First of all, it should be made clear that LLMs are designed to facilitate research, not to replace researchers, especially in paper writing. Over-reliance on AI for writing will undermine researchers\u0026rsquo; creativity and hinder their learning process. Second, completely blocking the application of LLMs in academic research is not wise. There are still many scholars who hope that ChatGPT can develop into an academic research tool. Therefore, it is necessary to use LLMs rationally to improve the efficiency of research and expand the depth and breadth of academic research. What\u0026rsquo;s more, relevant departments should set up relevant policies and regulations to standardize the use of LLMs.\u003c/p\u003e \u003cp\u003eThe insights garnered from this research not only contribute to a deeper understanding of LLMs\u0026rsquo; utilization in doctoral academic research but also provide a structured framework for optimizing their application. This framework is invaluable for researchers, educators, and policymakers seeking to harness the potential of AI-driven technologies to advance scholarly productivity, innovation, and knowledge dissemination in academia.\u003c/p\u003e"},{"header":"Declarations","content":"\u003ch2\u003eAuthor Contribution\u003c/h2\u003e\u003cp\u003eYihuan Yuan wrote the main manuscript text,Jamalludin Harun revised the manuscript.\u003c/p\u003e\u003ch2\u003eData Availability\u003c/h2\u003e\u003cp\u003eThis study involved conducting interviews with individuals and recording the interviews. To protect the privacy of the interviewees, the interview recordings are not publicly available. However, researchers may provide interview summaries or excerpts upon specific request for academic or research purposes. Please contact Yihuan Yuan ([email protected]) for more information and requests for access to interview data.\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\u003cli\u003e\u003cspan\u003eAntu, S. A., Chen, H., \u0026amp; Richards, C. K. (2023). Using LLMS (Large Language Model) to Improve Efficiency in Literature Review for Undergraduate Research.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eBin-Hady, W. R. A., Al-Kadi, A., Hazaea, A., \u0026amp; Ali, J. K. M. (2023). Exploring the dimensions of ChatGPT in English language learning: A global perspective. \u003cem\u003eLibrary Hi Tech\u003c/em\u003e, (ahead-of-print).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eCotton, D. R., Cotton, P. A., \u0026amp; Shipway, J. R. (2023). Chatting and cheating: Ensuring academic integrity in the era of ChatGPT. \u003cem\u003eInnovations in Education and Teaching International\u003c/em\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eDe Kok, T. (2023). Generative LLMs and Textual Analysis in Accounting:(Chat) GPT as Research Assistant?. \u003cem\u003eAvailable at SSRN\u003c/em\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eDempere, J., Modugu, K., Hesham, A., \u0026amp; Ramasamy, L. K. (2023, September). The impact of ChatGPT on higher education. Frontiers in Education, 8,1206936.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eDevlin, J., Chang, M.-W., Lee, K., \u0026amp; Toutanova, K. (2018). BERT: Pre-training of deep bidirectional transformers for language understanding. arXiv. preprint arXiv: 1810.04805.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eDijkstra, R., Gen\u0026ccedil;, Z., Kayal, S., \u0026amp; Kamps, J. (2022). Reading comprehension quiz generation using generative pre-trained transformers. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://e.humanities.uva.nl/publications/2022/dijk_read22.pdf\u003c/span\u003e\u003cspan address=\"https://e.humanities.uva.nl/publications/2022/dijk_read22.pdf\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eEysenbach, G. (2023). The role of ChatGPT, generative language models, and artificial intelligence in medical education: a conversation with ChatGPT and a call for papers. JMIR Medical Education, \u003cem\u003e9\u003c/em\u003e(1), e46885.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eFink, M. A., Bischoff, A., Fink, C. A., Moll, M., Kroschke, J., Dulz, L., \u0026hellip; Weber, T. F. (2023). Potential of ChatGPT and GPT-4 for data mining of free-text CT reports on lung cancer. Radiology, \u003cem\u003e308\u003c/em\u003e(3), e231362.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eFloridi, L., \u0026amp; Chiriatti, M. (2020). GPT-3: Its nature, scope, limits, and consequences. Minds and Machines, \u003cem\u003e30\u003c/em\u003e(4), 681\u0026ndash;694.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eGabajiwala, E., Mehta, P., Singh, R., \u0026amp; Koshy, R. (2022). Quiz maker: Automatic quiz generation from text using NLP. \u003cem\u003eIn Futuristic trends in networks and computing technologies\u003c/em\u003e, 523\u0026ndash;533.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eHeidt, A. (2023). \u0026lsquo;Arms race with automation\u0026rsquo;: Professors fret about AI-generated coursework. Nature.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eImran, M., \u0026amp; Almusharraf, N. (2023). Analyzing the role of ChatGPT as a writing assistant at higher education level: A systematic review of the literature. Contemporary Educational Technology, \u003cem\u003e15\u003c/em\u003e(4), ep464.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eJin, H., Zhang, Y., Meng, D., Wang, J., \u0026amp; Tan, J. (2024). A Comprehensive Survey on Process-Oriented Automatic Text Summarization with Exploration of LLM-Based Methods. arXiv preprint arXiv:2403.02901.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eKandpal, N., Deng, H., Roberts, A., Wallace, E., \u0026amp; Raffel, C. (2023). Large language models struggle to learn long-tail knowledge. In \u003cem\u003eInternational Conference on Machine Learning\u003c/em\u003e, 15696\u0026ndash;15707). PMLR.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eKasneci, E., Se\u0026szlig;ler, K., K\u0026uuml;chemann, S., Bannert, M., Dementieva, D., Fischer, F., \u0026hellip; Kasneci, G. (2023). ChatGPT for good? On opportunities and challenges of large language models for education. Learning and individual differences, \u003cem\u003e103\u003c/em\u003e, 102274.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eKelly, D., Chen, Y., Cornwell, S. E., Delellis, N. S., Mayhew, A., Onaolapo, S., \u0026amp; Rubin, V. L. (2023). Bing Chat: The Future of Search Engines?. \u003cem\u003eProceedings of the Association for Information Science and Technology\u003c/em\u003e, \u003cem\u003e60\u003c/em\u003e(1), 1007\u0026ndash;1009.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLametti, D. (2022). AI could be great for college essays. slate.com. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://slate.com/technology/2022/12/chatgpt-college-essay-plagiarism.html\u003c/span\u003e\u003cspan address=\"https://slate.com/technology/2022/12/chatgpt-college-essay-plagiarism.html\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLiu, G., \u0026amp; Ma, C. (2023). Measuring EFL learners\u0026rsquo; use of ChatGPT in informal digital learning of English based on the technology acceptance model. Innovation in Language Learning and Teaching, 1\u0026ndash;14.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLiu, Y., Ott, M., Goyal, N., Du, J., Joshi, M., Chen, D., Levy, O., Lewis, M., Zettlemoyer, L., \u0026amp; Stoyanov, V. (2019). Roberta: A robustly optimized bert pretraining approach. \u003cem\u003earXiv. preprint arXiv\u003c/em\u003e:1907.11692.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLund, B. D., Wang, T., Mannuru, N. R., Nie, B., Shimray, S., \u0026amp; Wang, Z. (2023). ChatGPT and a new academic reality: Artificial Intelligence-written research papers and the ethics of the large language models in scholarly publishing. Journal of the Association for Information Science and Technology, \u003cem\u003e74\u003c/em\u003e(5), 570\u0026ndash;581.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eRadford, A., Narasimhan, K., Salimans, T., Sutskever, I., et al. (2018). Improving language understanding by generative pre-training Accessed: 2023-01-22.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eRaffel, C., Shazeer, N., Roberts, A., Lee, K., Narang, S., Matena, M., Zhou, Y., Li, W., Liu, P. J., et al. (2020). Exploring the limits of transfer learning with a unified textto-text transformer. Journal of Machine Learning Research, \u003cem\u003e21\u003c/em\u003e(140), 1\u0026ndash;67.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eRasul, T., Nair, S., Kalendra, D., Robin, M., de Oliveira Santini, F., Ladeira, W. J., Sun, M., Day, I., Rather, R. A., \u0026amp; Heathcote, L. (2023). The role of ChatGPT in higher education: Benefits, challenges, and future research directions. Journal of Applied Learning and Teaching, \u003cem\u003e6\u003c/em\u003e(1)\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eRosenfeld, A., \u0026amp; Lazebnik, T. (2024). Whose LLMS is it Anyway? Linguistic Comparison and LLMS Attribution for GPT-3.5, GPT-4 and Bard. \u003cem\u003earXiv preprint arXiv:2402.14533\u003c/em\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eScao, T. L., Fan, A., Akiki, C., Pavlick, E., Ili\u0026acute;c, S., Hesslow, D., Castagn\u0026acute;e, R., Luccioni, A. S., Yvon, F., Gall\u0026acute;e, M., et al. (2022). BLOOM: A 176B-parameter openaccess multilingual language model. arXiv. preprint arXiv:2211.05100.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSuaverdez, J. B., \u0026amp; Suaverdez, U. V. (2023). Chatbots impact on academic writing. \u003cem\u003eGlobal Journal of Business and Integral Security\u003c/em\u003e, (2).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eUrman, A., \u0026amp; Makhortykh, M. (2023). The Silence of the LLMs: Cross-Lingual Analysis of Political Bias and False Information Prevalence in ChatGPT, Google Bard, and Bing Chat, \u003cem\u003eOSF Preprints\u003c/em\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eYan, D. (2023). Impact of ChatGPT on learners in a L2 writing practicum: An exploratory investigation. \u003cem\u003eEducation and Information Technologies\u003c/em\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eYang, J., Li, H. B., \u0026amp; Wei, D. (2023). The impact of ChatGPT and LLMs on medical imaging stakeholders: perspectives and use cases. Meta-Radiology, 100007.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eYang, Y., Tang, Y., \u0026amp; Tam, K. Y. (2023). Investlm: A large language model for investment using financial domain instruction tuning. \u003cem\u003earXiv preprint arXiv:2309.13064\u003c/em\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eYang, Z., Dai, Z., Yang, Y., Carbonell, J., Salakhutdinov, R., \u0026amp; Le, Q. V. (2019). XLNet: Generalized Autoregressive Pretraining for Language Understanding. Advances in neural information processing systems, 32. preprint \u003cem\u003earXiv\u003c/em\u003e:1810.04805.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eZhang, X., Yu, B., Yu, H., Lv, Y., Liu, T., Huang, F., \u0026hellip; Li, Y. (2023). Wider and deeper llm networks are fairer llm evaluators. arXiv preprint arXiv:2308.01862.\u003c/span\u003e\u003c/li\u003e\u003c/ol\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":false,"hideJournal":false,"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":"scientific-reports","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"scirep","sideBox":"Learn more about [Scientific Reports](http://www.nature.com/srep/)","snPcode":"","submissionUrl":"","title":"Scientific Reports","twitterHandle":"","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"stoa","reportingPortfolio":"Scientific Reports","inReviewEnabled":true,"inReviewRevisionsEnabled":true},"keywords":"Large language model, artificial intelligence, academic research, doctoral students, AIAAR","lastPublishedDoi":"10.21203/rs.3.rs-4337026/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-4337026/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003cp\u003eThe ever-expanding volume and complexity of academic research pose significant challenges for researchers, particularly doctoral students. In response to these challenges, utilizing Large Language Models (LLMs) has emerged as a promising alternative solution. Such LLMs as ChatGPT, Bing Chat and Google Bard are applied in academic research. This study conducted semi-structured interviews with 50 PhD students and used thematic analysis to explore the application of LLMs in academic research. The results indicate that LLMs assist literature reading by extracting main content, providing research topics, and making reading convenient; assist research design by generating research design ideas; assist academic writing by generating writing ideas, polishing writing, analyzing and visualizing data; assist knowledge construction by offering subject matter expertise and promoting science; assist admin works by writing admin emails. Based on these, a five-dimensional framework of AI-assisted academic research (AIAAR) has been established to explain the assistance of LLMs in academic research. This research not only sheds light on the practical benefits of integrating LLMs in academic research but also provides insights into optimizing their usage for enhanced scholarly productivity and knowledge advancement.\u003c/p\u003e","manuscriptTitle":"Empowering Doctoral Academic Research: Artificial Intelligence-driven Insights from Large Language Models","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2024-05-10 06:56:05","doi":"10.21203/rs.3.rs-4337026/v1","editorialEvents":[{"type":"communityComments","content":0},{"type":"editorInvitedReview","content":"","date":"2024-07-11T07:43:35+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"291936482207962297649655324838905732025","date":"2024-07-03T00:22:02+00:00","index":"hide","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2024-06-08T01:08:55+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"131428518116753078425391722545763797747","date":"2024-05-29T00:15:49+00:00","index":"hide","fulltext":""},{"type":"reviewersInvited","content":"","date":"2024-05-26T23:27:26+00:00","index":"","fulltext":""},{"type":"editorAssigned","content":"","date":"2024-05-26T23:22:44+00:00","index":"","fulltext":""},{"type":"editorInvited","content":"","date":"2024-05-07T18:47:09+00:00","index":"","fulltext":""},{"type":"checksComplete","content":"","date":"2024-05-06T04:19:34+00:00","index":"","fulltext":""},{"type":"submitted","content":"Scientific Reports","date":"2024-04-28T08:38:42+00:00","index":"","fulltext":""}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"scientific-reports","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"scirep","sideBox":"Learn more about [Scientific Reports](http://www.nature.com/srep/)","snPcode":"","submissionUrl":"","title":"Scientific Reports","twitterHandle":"","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"stoa","reportingPortfolio":"Scientific Reports","inReviewEnabled":true,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"c2a0654a-9162-4833-9f81-3d911dcd4ffa","owner":[],"postedDate":"May 10th, 2024","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"under-review","subjectAreas":[{"id":31642504,"name":"Physical sciences/Mathematics and computing/Computational science"},{"id":31642505,"name":"Physical sciences/Mathematics and computing/Software"}],"tags":[],"updatedAt":"2024-05-10T06:56:06+00:00","versionOfRecord":[],"versionCreatedAt":"2024-05-10 06:56:05","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-4337026","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-4337026","identity":"rs-4337026","version":["v1"]},"buildId":"8U1c8b4HqxoKbykW_rLl7","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}