情报杂志2025，Vol.44Issue(5)：82-90,9.DOI:10.3969/j.issn.1002-1965.2025.05.011
基于BERT和提示学习的网络暴力言论识别研究
Network Violence Speech Detection Based on BERT and Prompt Learning
摘要
Abstract
[Research purpose]In order to solve the problem of neglecting deep semantic information of text and having a strong depend-ence on dataset size in the traditional online violent speech recognition,a network violent speech recognition model is constructed based on BERT and prompt learning to improve the accuracy of online violent speech recognition and maintain a healthy online order.[Research method]By utilizing BERT's context understanding and mask prediction mechanism,the accuracy of identifying online violent speech in complex contexts is improved.At the same time,discrete and continuous templates are constructed to guide continuous comment text vec-tors to strengthen specific context and prompt associations through examples.Combined with discrete comment text vectors,integrated prompts are formed,and weighted fusion is used to obtain classification predictions,which provides new research ideas and solutions for the field of online violent speech recognition.[Research result/conclusion]Using three public events in Weibo and Tiktok as data sources,we constructed a small-scale violent speech recognition dataset for empirical analysis.The experimental results show that the model based on BERT and prompt learning has an Fl value increased by 3.63％ compared with the BERT+P-tuning method with the best performance in the comparative experiment,which verifies the superiority of the model performance.关键词
网络暴力/暴力言论识别/BERT/提示学习/微博/抖音Key words
online violence/identification of violent speech/BERT/prompt learning/Microblog/Tik Tok分类
社会科学引用本文复制引用
曾江峰,高鹏钰,李玲,马霄..基于BERT和提示学习的网络暴力言论识别研究[J].情报杂志,2025,44(5):82-90,9.基金项目
国家自然科学基金青年项目"情感感知的可解释虚假新闻检测研究"(编号:62102159) （编号:62102159）
湖北省自然科学基金一般面上项目"基于多层语义融合的多模态社交媒体虚假信息检测研究"(编号:2023AFB1018) （编号:2023AFB1018）
中央高校基本科研业务费项目"基于多大语言模型协商和多特征语义融合的虚假信息检测研究"(编号:CCNU24ZZ148)研究成果. （编号:CCNU24ZZ148）
