{"id":"https://openalex.org/W4416035395","doi":"https://doi.org/10.18653/v1/2025.emnlp-main.1636","title":"Text Detoxification: Data Efficiency, Semantic Preservation and Model Generalization","display_name":"Text Detoxification: Data Efficiency, Semantic Preservation and Model Generalization","publication_year":2025,"publication_date":"2025-01-01","ids":{"openalex":"https://openalex.org/W4416035395","doi":"https://doi.org/10.18653/v1/2025.emnlp-main.1636"},"language":null,"primary_location":{"id":"doi:10.18653/v1/2025.emnlp-main.1636","is_oa":true,"landing_page_url":"https://doi.org/10.18653/v1/2025.emnlp-main.1636","pdf_url":"https://aclanthology.org/2025.emnlp-main.1636.pdf","source":null,"license":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"Proceedings of the 2025 Conference on Empirical Methods in Natural Language Processing","raw_type":"proceedings-article"},"type":"article","indexed_in":["crossref"],"open_access":{"is_oa":true,"oa_status":"gold","oa_url":"https://aclanthology.org/2025.emnlp-main.1636.pdf","any_repository_has_fulltext":null},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5035578031","display_name":"Jing Yu","orcid":"https://orcid.org/0000-0002-3770-3186"},"institutions":[],"countries":[],"is_corresponding":true,"raw_author_name":"Jing Yu","raw_affiliation_strings":[],"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5009816039","display_name":"Yibo Zhao","orcid":"https://orcid.org/0009-0001-1239-1809"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Yibo Zhao","raw_affiliation_strings":[],"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5045628618","display_name":"Jiapeng Zhu","orcid":"https://orcid.org/0000-0001-9198-0304"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Jiapeng Zhu","raw_affiliation_strings":[],"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5018200521","display_name":"Wenming Shao","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Wenming Shao","raw_affiliation_strings":[],"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5101799604","display_name":"Bo Pang","orcid":"https://orcid.org/0000-0002-8947-141X"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Bo Pang","raw_affiliation_strings":[],"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5103205468","display_name":"Zhao Zhang","orcid":"https://orcid.org/0000-0002-1521-8163"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Zhao Zhang","raw_affiliation_strings":[],"affiliations":[]},{"author_position":"last","author":{"id":"https://openalex.org/A5100331136","display_name":"Xiang Li","orcid":"https://orcid.org/0009-0003-5859-2344"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Xiang Li","raw_affiliation_strings":[],"affiliations":[]}],"institutions":[],"countries_distinct_count":0,"institutions_distinct_count":7,"corresponding_author_ids":["https://openalex.org/A5035578031"],"corresponding_institution_ids":[],"apc_list":null,"apc_paid":null,"fwci":0.0,"has_fulltext":true,"cited_by_count":0,"citation_normalized_percentile":{"value":0.18386373,"is_in_top_1_percent":false,"is_in_top_10_percent":false},"cited_by_percentile_year":null,"biblio":{"volume":null,"issue":null,"first_page":"32160","last_page":"32174"},"is_retracted":false,"is_paratext":false,"is_xpac":false,"primary_topic":{"id":"https://openalex.org/T10181","display_name":"Natural Language Processing Techniques","score":0.23520000278949738,"subfield":{"id":"https://openalex.org/subfields/1702","display_name":"Artificial Intelligence"},"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/T10181","display_name":"Natural Language Processing Techniques","score":0.23520000278949738,"subfield":{"id":"https://openalex.org/subfields/1702","display_name":"Artificial Intelligence"},"field":{"id":"https://openalex.org/fields/17","display_name":"Computer Science"},"domain":{"id":"https://openalex.org/domains/3","display_name":"Physical Sciences"}},{"id":"https://openalex.org/T10028","display_name":"Topic Modeling","score":0.22450000047683716,"subfield":{"id":"https://openalex.org/subfields/1702","display_name":"Artificial Intelligence"},"field":{"id":"https://openalex.org/fields/17","display_name":"Computer Science"},"domain":{"id":"https://openalex.org/domains/3","display_name":"Physical Sciences"}},{"id":"https://openalex.org/T11550","display_name":"Text and Document Classification Technologies","score":0.04619999974966049,"subfield":{"id":"https://openalex.org/subfields/1702","display_name":"Artificial Intelligence"},"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/generalization","display_name":"Generalization","score":0.5514000058174133},{"id":"https://openalex.org/keywords/semantics","display_name":"Semantics (computer science)","score":0.35499998927116394},{"id":"https://openalex.org/keywords/feature","display_name":"Feature (linguistics)","score":0.3334999978542328},{"id":"https://openalex.org/keywords/set","display_name":"Set (abstract data type)","score":0.3165000081062317},{"id":"https://openalex.org/keywords/term","display_name":"Term (time)","score":0.24650000035762787}],"concepts":[{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.6394000053405762},{"id":"https://openalex.org/C177148314","wikidata":"https://www.wikidata.org/wiki/Q170084","display_name":"Generalization","level":2,"score":0.5514000058174133},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.5479000210762024},{"id":"https://openalex.org/C204321447","wikidata":"https://www.wikidata.org/wiki/Q30642","display_name":"Natural language processing","level":1,"score":0.4392000138759613},{"id":"https://openalex.org/C184337299","wikidata":"https://www.wikidata.org/wiki/Q1437428","display_name":"Semantics (computer science)","level":2,"score":0.35499998927116394},{"id":"https://openalex.org/C2776401178","wikidata":"https://www.wikidata.org/wiki/Q12050496","display_name":"Feature (linguistics)","level":2,"score":0.3334999978542328},{"id":"https://openalex.org/C177264268","wikidata":"https://www.wikidata.org/wiki/Q1514741","display_name":"Set (abstract data type)","level":2,"score":0.3165000081062317},{"id":"https://openalex.org/C61797465","wikidata":"https://www.wikidata.org/wiki/Q1188986","display_name":"Term (time)","level":2,"score":0.24650000035762787},{"id":"https://openalex.org/C67186912","wikidata":"https://www.wikidata.org/wiki/Q367664","display_name":"Data modeling","level":2,"score":0.2386000007390976},{"id":"https://openalex.org/C58489278","wikidata":"https://www.wikidata.org/wiki/Q1172284","display_name":"Data set","level":2,"score":0.23720000684261322}],"mesh":[],"locations_count":1,"locations":[{"id":"doi:10.18653/v1/2025.emnlp-main.1636","is_oa":true,"landing_page_url":"https://doi.org/10.18653/v1/2025.emnlp-main.1636","pdf_url":"https://aclanthology.org/2025.emnlp-main.1636.pdf","source":null,"license":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"Proceedings of the 2025 Conference on Empirical Methods in Natural Language Processing","raw_type":"proceedings-article"}],"best_oa_location":{"id":"doi:10.18653/v1/2025.emnlp-main.1636","is_oa":true,"landing_page_url":"https://doi.org/10.18653/v1/2025.emnlp-main.1636","pdf_url":"https://aclanthology.org/2025.emnlp-main.1636.pdf","source":null,"license":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"Proceedings of the 2025 Conference on Empirical Methods in Natural Language Processing","raw_type":"proceedings-article"},"sustainable_development_goals":[],"awards":[],"funders":[],"has_content":{"grobid_xml":true,"pdf":true},"content_urls":{"pdf":"https://content.openalex.org/works/W4416035395.pdf","grobid_xml":"https://content.openalex.org/works/W4416035395.grobid-xml"},"referenced_works_count":0,"referenced_works":[],"related_works":[],"abstract_inverted_index":{"The":[0],"widespread":[1],"dissemination":[2],"of":[3,95],"toxic":[4,108],"content":[5],"on":[6,56,91,147],"social":[7],"media":[8],"poses":[9],"a":[10,73,92,102,111],"serious":[11],"threat":[12],"to":[13,40,50,100,115],"both":[14],"online":[15],"environments":[16],"and":[17,48,84,110,143],"public":[18],"discourse,":[19],"highlighting":[20],"the":[21,33,117,131],"urgent":[22],"need":[23],"for":[24],"detoxification":[25,44],"methods":[26],"that":[27,77,126],"effectively":[28,129],"remove":[29],"toxicity":[30],"while":[31,62],"preserving":[32],"original":[34],"semantics.However,":[35],"existing":[36],"approaches":[37],"often":[38],"struggle":[39],"simultaneously":[41],"achieve":[42],"strong":[43,103],"performance,":[45],"semantic":[46],"preservation,":[47],"robustness":[49],"out-of-distribution":[51],"data.Moreover,":[52],"they":[53],"typically":[54],"rely":[55],"costly,":[57],"manually":[58],"annotated":[59,148],"parallel":[60,98],"corpora":[61],"showing":[63],"poor":[64],"data":[65,99],"efficiency.To":[66],"address":[67],"these":[68],"challenges,":[69],"we":[70,105],"propose":[71],"GEM,":[72],"two-stage":[74],"training":[75],"framework":[76],"jointly":[78],"optimizes":[79],"Model":[80],"Generalization,":[81],"Data":[82],"Efficiency,":[83],"Semantic":[85],"Preservation.We":[86],"first":[87],"perform":[88],"supervised":[89],"fine-tuning":[90],"small":[93],"set":[94],"high-quality,":[96],"filtered":[97],"establish":[101],"initialization.Then,":[104],"leverage":[106],"unlabeled":[107],"inputs":[109],"custom-designed":[112],"reward":[113],"model":[114],"train":[116],"LLM":[118],"using":[119],"Group":[120],"Relative":[121],"Policy":[122],"Optimization.Experimental":[123],"results":[124],"demonstrate":[125],"our":[127],"method":[128],"mitigates":[130],"trade-offs":[132],"faced":[133],"by":[134],"previous":[135],"work,":[136],"achieving":[137],"state-of-the-art":[138],"performance":[139],"with":[140],"improved":[141],"generalization":[142],"significantly":[144],"reduced":[145],"dependence":[146],"data.":[149]},"counts_by_year":[],"updated_date":"2026-03-10T14:07:55.174380","created_date":"2025-11-08T00:00:00"}