Fairness Identification of Large Language Models in Recommendation

Fairness Identification of Large Language Models in Recommendation | 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 Fairness Identification of Large Language Models in Recommendation Wei Liu, Baisong Liu, Jiangcheng Qin, Xueyuan Zhang, Weiming Huang, and 1 more This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-5228643/v1 This work is licensed under a CC BY 4.0 License Status: Published Journal Publication published 14 Feb, 2025 Read the published version in Scientific Reports → Version 1 posted 11 You are reading this latest preprint version Abstract Ensuring fairness in recommendation systems necessitates that models do not discriminate against users based on demographic information such as gender and age. Current fairness strategies often apply a unified fairness intervention, presuming that users' recommendation results are adversely influenced by sensitive attributes. This approach can sometimes diminish both the utility and fairness of recommendations for certain users. Drawing inspiration from the studies of human-like behavior in large language models(LLMs), we investigate whether LLMs can serve as fairness recognizers in recommendation systems. Specifically, we explore if the fairness awareness inherent in LLMs can be harnessed to construct fair recommendations. To this end, we generate recommendation results on MovieLens and LastFM datasets using the Variational Autoencoder(VAE) and VAE with integrated fairness strategies. Our findings reveal that LLMs can indeed recognize fair recommendations by evaluating the fairness of users' recommendation results. We then propose a method to design fair recommendations by incorporating LLMs: replacing the recommendation results generated by the VAE of users identified as unfair by LLMs with those generated by a fair VAE. Evaluating these reconstructed recommendations demonstrates that leveraging the fairness recognition capabilities of LLMs achieves a better balance between effectiveness and fairness. Recommendation fairness Large language models Fairness identification Sensitive attributes Full Text Additional Declarations No competing interests reported. Cite Share Download PDF Status: Published Journal Publication published 14 Feb, 2025 Read the published version in Scientific Reports → Version 1 posted Editorial decision: Revision requested 26 Nov, 2024 Reviews received at journal 23 Nov, 2024 Reviews received at journal 22 Nov, 2024 Reviewers agreed at journal 14 Nov, 2024 Reviewers agreed at journal 12 Nov, 2024 Reviewers agreed at journal 12 Nov, 2024 Reviewers invited by journal 10 Nov, 2024 Editor assigned by journal 10 Nov, 2024 Editor invited by journal 07 Nov, 2024 Submission checks completed at journal 04 Nov, 2024 First submitted to journal 08 Oct, 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-5228643","acceptedTermsAndConditions":true,"allowDirectSubmit":false,"archivedVersions":[],"articleType":"Article","associatedPublications":[],"authors":[{"id":378192008,"identity":"10d467ea-7cac-4508-aa16-b8947f1905a0","order_by":0,"name":"Wei Liu","email":"","orcid":"","institution":"Ningbo University","correspondingAuthor":false,"prefix":"","firstName":"Wei","middleName":"","lastName":"Liu","suffix":""},{"id":378192009,"identity":"2818db6e-b01e-445a-875f-53649b94bc8a","order_by":1,"name":"Baisong Liu","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAAoklEQVRIiWNgGAWjYBACPgkGxgeMDSBmApFa2CQYmA1I1sImQaIW6fZn1YU7DjPws+cYMPzcQYwWmTNmt2eeOcwg2fPGgLH3DFEOy2G7zdt2mMHgRo4BM2MbUVrSnxWDtNiToCXBjBlsiwTRWmTOGEvznknnkTjzrOBgLzFa+KXbH37m3WEtx9+evPHBT2K0wAAPiDhAgoZRMApGwSgYBfgAAFS/LruxbJIvAAAAAElFTkSuQmCC","orcid":"","institution":"Ningbo University","correspondingAuthor":true,"prefix":"","firstName":"Baisong","middleName":"","lastName":"Liu","suffix":""},{"id":378192010,"identity":"fa2c1439-6c2d-4aa7-8931-a5c42193cf76","order_by":2,"name":"Jiangcheng Qin","email":"","orcid":"","institution":"Ningbo University","correspondingAuthor":false,"prefix":"","firstName":"Jiangcheng","middleName":"","lastName":"Qin","suffix":""},{"id":378192011,"identity":"786e9636-40cd-45ff-bc61-653bea984f84","order_by":3,"name":"Xueyuan Zhang","email":"","orcid":"","institution":"Ningbo University","correspondingAuthor":false,"prefix":"","firstName":"Xueyuan","middleName":"","lastName":"Zhang","suffix":""},{"id":378192012,"identity":"5753ac11-3633-4db4-8acd-c819a0f14835","order_by":4,"name":"Weiming Huang","email":"","orcid":"","institution":"Ningbo University","correspondingAuthor":false,"prefix":"","firstName":"Weiming","middleName":"","lastName":"Huang","suffix":""},{"id":378192013,"identity":"a32518e5-f669-4d48-8e8f-14c6da252296","order_by":5,"name":"Yangyang Wang","email":"","orcid":"","institution":"Ningbo University","correspondingAuthor":false,"prefix":"","firstName":"Yangyang","middleName":"","lastName":"Wang","suffix":""}],"badges":[],"createdAt":"2024-10-09 02:38:15","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-5228643/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-5228643/v1","draftVersion":[],"editorialEvents":[{"content":"https://doi.org/10.1038/s41598-025-89965-3","type":"published","date":"2025-02-14T15:57:28+00:00"}],"editorialNote":"","failedWorkflow":false,"files":[{"id":76487730,"identity":"24be18f1-5bda-458b-bbc9-af246abf9601","added_by":"auto","created_at":"2025-02-17 16:11:56","extension":"pdf","order_by":1,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":366852,"visible":true,"origin":"","legend":"","description":"","filename":"FairnessIdentificationofLargeLanguageModelsinRecommendation.pdf","url":"https://assets-eu.researchsquare.com/files/rs-5228643/v1_covered_0291d552-2bde-4141-a581-fa84de425c83.pdf"}],"financialInterests":"No competing interests reported.","formattedTitle":"Fairness Identification of Large Language Models in Recommendation","fulltext":[],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":false,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":false,"hideJournal":false,"highlight":"","institution":"","isAcceptedByJournal":true,"isAuthorSuppliedPdf":true,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":true,"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":"Recommendation fairness, Large language models, Fairness identification, Sensitive attributes","lastPublishedDoi":"10.21203/rs.3.rs-5228643/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-5228643/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003cp\u003eEnsuring fairness in recommendation systems necessitates that models do not discriminate against users based on demographic information such as gender and age. Current fairness strategies often apply a unified fairness intervention, presuming that users' recommendation results are adversely influenced by sensitive attributes. This approach can sometimes diminish both the utility and fairness of recommendations for certain users. Drawing inspiration from the studies of human-like behavior in large language models(LLMs), we investigate whether LLMs can serve as \u003cb\u003efairness recognizers\u003c/b\u003e in recommendation systems. Specifically, we explore if the \u003cb\u003efairness awareness\u003c/b\u003e inherent in LLMs can be harnessed to construct fair recommendations. To this end, we generate recommendation results on MovieLens and LastFM datasets using the Variational Autoencoder(VAE) and VAE with integrated fairness strategies. Our findings reveal that LLMs can indeed recognize fair recommendations by evaluating the fairness of users' recommendation results. We then propose a method to design fair recommendations by incorporating LLMs: replacing the recommendation results generated by the VAE of users identified as unfair by LLMs with those generated by a fair VAE. Evaluating these reconstructed recommendations demonstrates that leveraging the fairness recognition capabilities of LLMs achieves a better balance between effectiveness and fairness.\u003c/p\u003e","manuscriptTitle":"Fairness Identification of Large Language Models in Recommendation","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2024-11-15 17:01:44","doi":"10.21203/rs.3.rs-5228643/v1","editorialEvents":[{"type":"communityComments","content":0},{"type":"decision","content":"Revision requested","date":"2024-11-26T06:10:30+00:00","index":"","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2024-11-23T12:48:41+00:00","index":"hide","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2024-11-22T09:46:24+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"198287783689157794396830090077918197212","date":"2024-11-14T14:31:25+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"20650814852462086147052065289103182499","date":"2024-11-12T14:02:05+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"164264406212783617231302887519788578506","date":"2024-11-12T06:20:40+00:00","index":"hide","fulltext":""},{"type":"reviewersInvited","content":"","date":"2024-11-11T04:11:31+00:00","index":"","fulltext":""},{"type":"editorAssigned","content":"","date":"2024-11-10T08:43:44+00:00","index":"","fulltext":""},{"type":"editorInvited","content":"","date":"2024-11-07T07:11:31+00:00","index":"","fulltext":""},{"type":"checksComplete","content":"","date":"2024-11-04T06:31:42+00:00","index":"","fulltext":""},{"type":"submitted","content":"Scientific Reports","date":"2024-10-09T02:32:43+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":"d05ec763-d335-4260-a491-7137a0349d07","owner":[],"postedDate":"November 15th, 2024","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"published-in-journal","subjectAreas":[],"tags":[],"updatedAt":"2025-02-17T16:05:50+00:00","versionOfRecord":{"articleIdentity":"rs-5228643","link":"https://doi.org/10.1038/s41598-025-89965-3","journal":{"identity":"scientific-reports","isVorOnly":false,"title":"Scientific Reports"},"publishedOn":"2025-02-14 15:57:28","publishedOnDateReadable":"February 14th, 2025"},"versionCreatedAt":"2024-11-15 17:01:44","video":"","vorDoi":"10.1038/s41598-025-89965-3","vorDoiUrl":"https://doi.org/10.1038/s41598-025-89965-3","workflowStages":[]},"version":"v1","identity":"rs-5228643","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-5228643","identity":"rs-5228643","version":["v1"]},"buildId":"qtupq5eGEP_6zYnWcrvyt","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}