面向社交媒体的中文文本毒性检测研究综述

打开文本图片集

中图分类号：TP391.1 文献标志码：A 文章编号：1001-3695（2026）01-002-0011-12

doi：10.19734/j. issn.1001-3695.2025.05.0173

Survey of Chinese toxic text detection in social media contexts

Sun Lianyi，Xu Jingwen†（ScholofburitfodellcialtigatceUsfn）

Abstract：Withtherapiddevelopmentof social media，alarge volumeof toxiccontentcontainingdiscrimination，abuse，and hate speechhasemergedonline，posingseriousthreats totheinformationenvironmentandsocialstabilityThissurveyinvestigatedthecurrentstatusandtechnicaladvancesin Chinesetoxictextdetection.Itprovidedacomprehensivereviewof task definitions，evaluationmetrics，datasetresources，modeling approaches，typicalapplications，andmajorchallenges.Thestudyfocused onrule-based methods，featureengineering，deep learning，and pretrained language models，andfurtherdiscuedadvanced directionssuchasmulti-task learning，cros-lingual transfer，andlargemodeldistilation.Bycomparingtheperformanceandaplicabilityofdiferenttechnques，itsummarizedkeyostaclesinatasarityimplicitexpressions，modelbias， andmultimodalmodeling，andoutlinedpotentialresearchdirections.Thisreviewoferedtheoreticalinsightsandtechnicalreferences for the standardization and practical implementation of toxic text detection in Chinese contexts.

Key Words：toxicity detection；natural language processing；deep learning；data augmentation

0 引言

互联网高速发展，微博、推特、论坛等社交媒体平台中每天会产生大量的用户生成内容，如用户发布的微博评论、推特评论、论坛评论等。（剩余40820字）