Multimodal Detection of Hateful Memes by Applying a Vision-Language Pre-Training Model

Multimodal Detection of Hateful Memes by Applying a Vision-Language Pre-Training Model | 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 Method Article Multimodal Detection of Hateful Memes by Applying a Vision-Language Pre-Training Model Yuyang Chen, Feng Pan This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-1414253/v2 This work is licensed under a CC BY 4.0 License Status: Posted Version 2 posted You are reading this latest preprint version Show more versions Abstract Detrimental to individuals and society, online hateful messages have recently become a major social issue. Among them, one new type of hateful message, named “hateful meme”, has emerged and brought difficulties in traditional deep learning-based detection. Because hateful memes were formatted with both text and image to express users’ intents, they cannot be accurately identified by singularly analyzing embedded text or images. In order to effectively detect a hateful meme, the algorithm must possess strong vision and language fusion capability. In this study, we move closer to this goal by feeding a triplet by stacking the visual features, object tags, and textual features of memes generated by the object detection model VinVL and the optical character recognition (OCR) technology into a Transformer-based Vision-Language Pre-Training model OSCAR+ to perform the cross-modal learning of memes. After fine-tuning and connecting to a random forest classifier (RF), our model (OSCAR+RF) achieved a 0.768 AUROC score on the hateful meme detection task in a public dataset, which was higher than the published baselines. In conclusion, this study has demonstrated that Vision-Language PTMs with the addition of anchor points can improve the performance of deep learning-based detection of hateful memes by involving a more substantial alignment between the textual and visual information. Artificial Intelligence and Machine Learning Artificial Intelligence Deep Learning Multimodal Hate Speech Self-attention Mechanism Full Text Cite Share Download PDF Status: Posted Version 2 posted You are reading this latest preprint version Show more versions 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-1414253","acceptedTermsAndConditions":true,"allowDirectSubmit":true,"archivedVersions":[],"articleType":"Method Article","associatedPublications":[],"authors":[{"id":88990257,"identity":"20fc9ba3-5378-4089-85a7-5dcfe1fa6e94","order_by":0,"name":"Yuyang Chen","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAA+ElEQVRIiWNgGAWjYFACHiCugPMkiNVyBkKRoIWxDa6FCGBw/OzBx4Xz7iTul25+uvFHjYWcOfvhAwwf99Ti1nImL9l45rZniT0yx8xu8xyTMLbsSUtgnPHsOG4tN3jMpHm3HU7skUgwu83YIJG44QaPATPPgWP4tJj/5p0D0pL+7ebPBon6DTf4PxDSYsbM2wDSkmN2g7dBIgEowgDUUoNTi+SZHGNpnmPPjHtu5JSB/GK44UyawcEZBw7g1MJ3/IzhZ56aO7LtM9K33fxRUydvcPzwwwcfDtTh1KIAMQ3NTCD3ME4t8g3YtAABbltGwSgYBaNgxAEAfkRcqpG8C8EAAAAASUVORK5CYII=","orcid":"","institution":"Putnam Science Academy, 18 Maple St, Putnam, CT, USA","correspondingAuthor":true,"submittingAuthor":false,"prefix":"","firstName":"Yuyang","middleName":"","lastName":"Chen","suffix":""},{"id":88990258,"identity":"dbd9e12c-156c-4841-9d54-790f7c25d9cd","order_by":1,"name":"Feng Pan","email":"","orcid":"https://orcid.org/0000-0003-2820-768X","institution":"Department of Radiology, Union Hospital, Tongji Medical College, Huazhong University of Science and Technology, Jiefang Ave 1277, Wuhan 430022, China","correspondingAuthor":false,"submittingAuthor":false,"prefix":"","firstName":"Feng","middleName":"","lastName":"Pan","suffix":""}],"badges":[],"createdAt":"2022-03-03 07:03:42","currentVersionCode":2,"declarations":"","doi":"10.21203/rs.3.rs-1414253/v2","doiUrl":"https://doi.org/10.21203/rs.3.rs-1414253/v2","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":20198880,"identity":"5f6c3fea-1d15-4413-bd81-47023562a352","added_by":"auto","created_at":"2022-04-11 15:01:48","extension":"pdf","order_by":1,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":677129,"visible":true,"origin":"","legend":"","description":"","filename":"MultimodalDetectionofHatefulMemesbyApplyingaVisionLanguagePreTrainingModel2022.4.10.pdf","url":"https://assets-eu.researchsquare.com/files/rs-1414253/v2_covered.pdf"}],"financialInterests":"","formattedTitle":"\u003cp\u003eMultimodal Detection of Hateful Memes by Applying a Vision-Language Pre-Training Model\u003c/p\u003e","fulltext":[{"header":"Full Text","content":"This preprint is available for \u003ca href='/article/rs-1414253/latest.pdf' target='_blank'\u003edownload as a PDF\u003c/a\u003e."}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":false,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":true,"hideJournal":true,"highlight":"","institution":"","isAcceptedByJournal":false,"isAuthorSuppliedPdf":true,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true},"keywords":"Artificial Intelligence, Deep Learning, Multimodal, Hate Speech, Self-attention Mechanism","lastPublishedDoi":"10.21203/rs.3.rs-1414253/v2","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-1414253/v2","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003cp\u003eDetrimental to individuals and society, online hateful messages have recently become a major social issue. Among them, one new type of hateful message, named “hateful meme”, has emerged and brought difficulties in traditional deep learning-based detection. Because hateful memes were formatted with both text and image to express users’ intents, they cannot be accurately identified by singularly analyzing embedded text or images. In order to effectively detect a hateful meme, the algorithm must possess strong vision and language fusion capability. In this study, we move closer to this goal by feeding a triplet by stacking the visual features, object tags, and textual features of memes generated by the object detection model VinVL and the optical character recognition (OCR) technology into a Transformer-based Vision-Language Pre-Training model OSCAR+ to perform the cross-modal learning of memes. After fine-tuning and connecting to a random forest classifier (RF), our model (OSCAR+RF) achieved a 0.768 AUROC score on the hateful meme detection task in a public dataset, which was higher than the published baselines. In conclusion, this study has demonstrated that Vision-Language PTMs with the addition of anchor points can improve the performance of deep learning-based detection of hateful memes by involving a more substantial alignment between the textual and visual information.\u003c/p\u003e","manuscriptTitle":"Multimodal Detection of Hateful Memes by Applying a Vision-Language Pre-Training Model","msid":"","msnumber":"","nonDraftVersions":[{"code":2,"date":"2022-04-11 15:01:41","doi":"10.21203/rs.3.rs-1414253/v2","editorialEvents":[{"type":"communityComments","content":0}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true}},{"code":1,"date":"2022-03-07 23:10:03","doi":"10.21203/rs.3.rs-1414253/v1","editorialEvents":[{"type":"communityComments","content":0}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"bb95e76d-523c-4e38-9886-b0d7d2c837b2","owner":[],"postedDate":"April 11th, 2022","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"posted","subjectAreas":[{"id":11743016,"name":"Artificial Intelligence and Machine Learning"}],"tags":[],"updatedAt":"2022-03-07T23:10:03+00:00","versionOfRecord":[],"versionCreatedAt":"2022-04-11 15:01:41","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v2","identity":"rs-1414253","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-1414253","identity":"rs-1414253","version":["v2"]},"buildId":"-HB7Z8yhvgn0wM9Nzuekk","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}