{"id":"https://openalex.org/W4414360521","doi":"https://doi.org/10.24963/ijcai.2025/53","title":"Towards Safer Pretraining: Analyzing and Filtering Harmful Content in Webscale Datasets for Responsible LLMs","display_name":"Towards Safer Pretraining: Analyzing and Filtering Harmful Content in Webscale Datasets for Responsible LLMs","publication_year":2025,"publication_date":"2025-09-01","ids":{"openalex":"https://openalex.org/W4414360521","doi":"https://doi.org/10.24963/ijcai.2025/53"},"language":"en","primary_location":{"id":"doi:10.24963/ijcai.2025/53","is_oa":true,"landing_page_url":"https://doi.org/10.24963/ijcai.2025/53","pdf_url":"https://www.ijcai.org/proceedings/2025/0053.pdf","source":null,"license":null,"license_id":null,"version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"Proceedings of the Thirty-Fourth International Joint Conference on Artificial Intelligence","raw_type":"proceedings-article"},"type":"conference-paper","indexed_in":["crossref"],"open_access":{"is_oa":true,"oa_status":"gold","oa_url":"https://www.ijcai.org/proceedings/2025/0053.pdf","any_repository_has_fulltext":null},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5119679452","display_name":"Sai Krishna Mendu","orcid":null},"institutions":[{"id":"https://openalex.org/I4210164937","display_name":"Microsoft Research (United Kingdom)","ror":"https://ror.org/05k87vq12","country_code":"GB","type":"company","lineage":["https://openalex.org/I1290206253","https://openalex.org/I4210164937"]}],"countries":["GB"],"is_corresponding":false,"raw_author_name":"Sai Krishna Mendu","raw_affiliation_strings":["Microsoft"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Microsoft","institution_ids":["https://openalex.org/I4210164937"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5081949937","display_name":"Harish Yenala","orcid":null},"institutions":[{"id":"https://openalex.org/I4210164937","display_name":"Microsoft Research (United Kingdom)","ror":"https://ror.org/05k87vq12","country_code":"GB","type":"company","lineage":["https://openalex.org/I1290206253","https://openalex.org/I4210164937"]}],"countries":["GB"],"is_corresponding":false,"raw_author_name":"Harish Yenala","raw_affiliation_strings":["Microsoft"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Microsoft","institution_ids":["https://openalex.org/I4210164937"]}]},{"author_position":"middle","author":{"id":null,"display_name":"Aditi Gulati","orcid":null},"institutions":[{"id":"https://openalex.org/I4210164937","display_name":"Microsoft Research (United Kingdom)","ror":"https://ror.org/05k87vq12","country_code":"GB","type":"company","lineage":["https://openalex.org/I1290206253","https://openalex.org/I4210164937"]}],"countries":["GB"],"is_corresponding":false,"raw_author_name":"Aditi Gulati","raw_affiliation_strings":["Microsoft"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Microsoft","institution_ids":["https://openalex.org/I4210164937"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5110598239","display_name":"Shanu Kumar","orcid":null},"institutions":[{"id":"https://openalex.org/I4210164937","display_name":"Microsoft Research (United Kingdom)","ror":"https://ror.org/05k87vq12","country_code":"GB","type":"company","lineage":["https://openalex.org/I1290206253","https://openalex.org/I4210164937"]}],"countries":["GB"],"is_corresponding":false,"raw_author_name":"Shanu Kumar","raw_affiliation_strings":["Microsoft"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Microsoft","institution_ids":["https://openalex.org/I4210164937"]}]},{"author_position":"last","author":{"id":"https://openalex.org/A5101413392","display_name":"Parag Agrawal","orcid":"https://orcid.org/0000-0001-9183-8420"},"institutions":[{"id":"https://openalex.org/I4210164937","display_name":"Microsoft Research (United Kingdom)","ror":"https://ror.org/05k87vq12","country_code":"GB","type":"company","lineage":["https://openalex.org/I1290206253","https://openalex.org/I4210164937"]}],"countries":["GB"],"is_corresponding":false,"raw_author_name":"Parag Agrawal","raw_affiliation_strings":["Microsoft"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Microsoft","institution_ids":["https://openalex.org/I4210164937"]}]}],"institutions":[],"countries_distinct_count":1,"institutions_distinct_count":1,"corresponding_author_ids":[],"corresponding_institution_ids":["https://openalex.org/I4210164937"],"apc_list":null,"apc_paid":null,"fwci":1.9052,"has_fulltext":true,"cited_by_count":1,"citation_normalized_percentile":{"value":0.86650248,"is_in_top_1_percent":false,"is_in_top_10_percent":false},"cited_by_percentile_year":{"min":94,"max":97},"biblio":{"volume":null,"issue":null,"first_page":"466","last_page":"474"},"is_retracted":false,"is_paratext":false,"is_xpac":false,"primary_topic":{"id":"https://openalex.org/T12034","display_name":"Digital and Cyber Forensics","score":0.9135000109672546,"subfield":{"id":"https://openalex.org/subfields/1710","display_name":"Information Systems"},"field":{"id":"https://openalex.org/fields/17","display_name":"Computer Science"},"domain":{"id":"https://openalex.org/domains/3","display_name":"Physical Sciences"}},"topics":[{"id":"https://openalex.org/T12034","display_name":"Digital and Cyber Forensics","score":0.9135000109672546,"subfield":{"id":"https://openalex.org/subfields/1710","display_name":"Information Systems"},"field":{"id":"https://openalex.org/fields/17","display_name":"Computer Science"},"domain":{"id":"https://openalex.org/domains/3","display_name":"Physical Sciences"}}],"keywords":[{"id":"https://openalex.org/keywords/safer","display_name":"SAFER","score":0.7172999978065491},{"id":"https://openalex.org/keywords/offensive","display_name":"Offensive","score":0.6549000144004822},{"id":"https://openalex.org/keywords/adversarial-system","display_name":"Adversarial system","score":0.5238000154495239},{"id":"https://openalex.org/keywords/work","display_name":"Work (physics)","score":0.3490999937057495},{"id":"https://openalex.org/keywords/resource","display_name":"Resource (disambiguation)","score":0.3366999924182892},{"id":"https://openalex.org/keywords/risk-assessment","display_name":"Risk assessment","score":0.33629998564720154},{"id":"https://openalex.org/keywords/content-analysis","display_name":"Content analysis","score":0.31520000100135803}],"concepts":[{"id":"https://openalex.org/C2776654903","wikidata":"https://www.wikidata.org/wiki/Q2601463","display_name":"SAFER","level":2,"score":0.7172999978065491},{"id":"https://openalex.org/C176856949","wikidata":"https://www.wikidata.org/wiki/Q2001676","display_name":"Offensive","level":2,"score":0.6549000144004822},{"id":"https://openalex.org/C37736160","wikidata":"https://www.wikidata.org/wiki/Q1801315","display_name":"Adversarial system","level":2,"score":0.5238000154495239},{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.5045999884605408},{"id":"https://openalex.org/C112930515","wikidata":"https://www.wikidata.org/wiki/Q4389547","display_name":"Risk analysis (engineering)","level":1,"score":0.3677000105381012},{"id":"https://openalex.org/C2522767166","wikidata":"https://www.wikidata.org/wiki/Q2374463","display_name":"Data science","level":1,"score":0.35749998688697815},{"id":"https://openalex.org/C18762648","wikidata":"https://www.wikidata.org/wiki/Q42213","display_name":"Work (physics)","level":2,"score":0.3490999937057495},{"id":"https://openalex.org/C206345919","wikidata":"https://www.wikidata.org/wiki/Q20380951","display_name":"Resource (disambiguation)","level":2,"score":0.3366999924182892},{"id":"https://openalex.org/C12174686","wikidata":"https://www.wikidata.org/wiki/Q1058438","display_name":"Risk assessment","level":2,"score":0.33629998564720154},{"id":"https://openalex.org/C162446236","wikidata":"https://www.wikidata.org/wiki/Q653137","display_name":"Content analysis","level":2,"score":0.31520000100135803},{"id":"https://openalex.org/C58642233","wikidata":"https://www.wikidata.org/wiki/Q8269924","display_name":"Taxonomy (biology)","level":2,"score":0.31290000677108765},{"id":"https://openalex.org/C108827166","wikidata":"https://www.wikidata.org/wiki/Q175975","display_name":"Internet privacy","level":1,"score":0.3116999864578247},{"id":"https://openalex.org/C2777363581","wikidata":"https://www.wikidata.org/wiki/Q15098235","display_name":"Harm","level":2,"score":0.2964000105857849},{"id":"https://openalex.org/C31170391","wikidata":"https://www.wikidata.org/wiki/Q188619","display_name":"Hierarchy","level":2,"score":0.2865999937057495},{"id":"https://openalex.org/C38652104","wikidata":"https://www.wikidata.org/wiki/Q3510521","display_name":"Computer security","level":1,"score":0.28450000286102295},{"id":"https://openalex.org/C2777834853","wikidata":"https://www.wikidata.org/wiki/Q96776939","display_name":"Liability","level":2,"score":0.26600000262260437},{"id":"https://openalex.org/C32896092","wikidata":"https://www.wikidata.org/wiki/Q189447","display_name":"Risk management","level":2,"score":0.26179999113082886},{"id":"https://openalex.org/C95713431","wikidata":"https://www.wikidata.org/wiki/Q631425","display_name":"Vulnerability (computing)","level":2,"score":0.2540000081062317},{"id":"https://openalex.org/C26517878","wikidata":"https://www.wikidata.org/wiki/Q228039","display_name":"Key (lock)","level":2,"score":0.25209999084472656},{"id":"https://openalex.org/C195324797","wikidata":"https://www.wikidata.org/wiki/Q33742","display_name":"Natural language","level":2,"score":0.25119999051094055}],"mesh":[],"locations_count":1,"locations":[{"id":"doi:10.24963/ijcai.2025/53","is_oa":true,"landing_page_url":"https://doi.org/10.24963/ijcai.2025/53","pdf_url":"https://www.ijcai.org/proceedings/2025/0053.pdf","source":null,"license":null,"license_id":null,"version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"Proceedings of the Thirty-Fourth International Joint Conference on Artificial Intelligence","raw_type":"proceedings-article"}],"best_oa_location":{"id":"doi:10.24963/ijcai.2025/53","is_oa":true,"landing_page_url":"https://doi.org/10.24963/ijcai.2025/53","pdf_url":"https://www.ijcai.org/proceedings/2025/0053.pdf","source":null,"license":null,"license_id":null,"version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"Proceedings of the Thirty-Fourth International Joint Conference on Artificial Intelligence","raw_type":"proceedings-article"},"sustainable_development_goals":[],"awards":[],"funders":[],"has_content":{"pdf":true,"grobid_xml":true},"content_urls":{"pdf":"https://content.openalex.org/works/W4414360521.pdf","grobid_xml":"https://content.openalex.org/works/W4414360521.grobid-xml"},"referenced_works_count":0,"referenced_works":[],"related_works":[],"abstract_inverted_index":{"Large":[0],"language":[1,33],"models":[2,145],"(LLMs)":[3],"have":[4],"become":[5],"integral":[6],"to":[7,147,178],"various":[8],"real-world":[9],"applications,":[10],"leveraging":[11],"massive,":[12],"web-sourced":[13],"datasets":[14,25],"like":[15],"Common":[16],"Crawl,":[17],"C4,":[18],"and":[19,45,60,71,100,116,120,139,160],"FineWeb":[20],"for":[21,30,125,165],"pretraining.":[22],"While":[23],"these":[24,88],"provide":[26,140],"linguistic":[27],"data":[28,53],"essential":[29],"high-quality":[31],"natural":[32],"generation,":[34],"they":[35],"often":[36],"contain":[37],"harmful":[38,96,126],"content,":[39],"such":[40,51],"as":[41,162],"hate":[42],"speech,":[43],"misinformation,":[44,59],"biased":[46],"narratives.":[47],"Training":[48],"LLMs":[49],"on":[50,103],"unfiltered":[52],"risks":[54],"perpetuating":[55],"toxic":[56,149],"behaviors,":[57],"spreading":[58],"amplifying":[61],"societal":[62],"biases":[63],"which":[64],"can":[65],"undermine":[66],"trust":[67],"in":[68],"LLM-driven":[69],"applications":[70],"raise":[72],"ethical":[73],"concerns":[74],"about":[75],"their":[76,104],"use.":[77],"This":[78,171],"paper":[79,172],"presents":[80],"a":[81,91,109,113,121,132,163],"large-scale":[82],"analysis":[83],"of":[84,181],"inappropriate":[85],"content":[86,127,176],"across":[87],"datasets,":[89],"offering":[90],"comprehensive":[92],"taxonomy":[93],"that":[94],"categorizes":[95],"webpages":[97],"into":[98,143,155],"Topical":[99,115],"Toxic":[101,117],"based":[102],"intent.":[105],"We":[106],"also":[107],"introduce":[108],"prompt":[110],"evaluation":[111],"dataset,":[112],"high-accuracy":[114],"Prompt":[118],"(TTP),":[119],"transformer-based":[122],"model":[123],"(HarmFormer)":[124],"filtering.":[128],"Additionally,":[129],"we":[130],"create":[131],"new":[133],"multi-harm":[134],"open-ended":[135],"toxicity":[136],"benchmark":[137],"(HAVOC)":[138],"crucial":[141],"insights":[142,154],"how":[144],"respond":[146],"adversarial":[148],"inputs.":[150],"Our":[151],"work":[152],"offers":[153],"ensuring":[156],"safer":[157],"LLM":[158],"pretraining":[159],"serves":[161],"resource":[164],"Responsible":[166],"AI":[167],"(RAI)":[168],"compliance.":[169],"Disclaimer:":[170],"includes":[173],"potentially":[174],"offensive":[175],"due":[177],"the":[179,182],"nature":[180],"research.":[183]},"counts_by_year":[{"year":2026,"cited_by_count":1}],"updated_date":"2026-07-31T08:31:51.225901","created_date":"2025-10-10T00:00:00"}