基于知识迁移的情感—原因对抽取  

Emotion-Cause Pair Extraction Based on Knowledge-Transfer

作　　者：赵凤园 刘德喜[1,2] 万齐智 刘喜平[1,2] 廖国琼[1,2] 万常选[1,2] ZHAO Fengyuan;LIU Dexi;WAN Qizhi;LIU Xiping;LIAO Guoqiong;WAN Changxuan(School of Computing and Artificial Intelligence,Jiangxi University of Finance and Economics,Nanchang,Jiangxi 330013,China;Key Laboratory of Data and Knowledge Engineering,Jiangxi University of Finance and Economics,Nanchang,Jiangxi 330013,China)

机构地区：[1]江西财经大学计算机与人工智能学院,江西南昌330013 [2]江西财经大学数据与知识工程江西省高校重点实验室,江西南昌330013

出　　处：《中文信息学报》2025年第1期121-132,共12页Journal of Chinese Information Processing

基　　金：国家自然科学基金(62272206,62272205,62462034);江西省主要学科学术和技术带头人培养计划领军人才项目(20213BCJL22041);江西省自然科学基金(20212ACB202002,20242BAB25119);江西省教育厅科学技术研究项目(GJJ2200501)。

摘　　要：现有的情感—原因对抽取模型均没有通过加入外部知识来提升情感—原因对的抽取效果。该文提出基于知识迁移的情感—原因对抽取模型(ECPE-KT),采用知识库获取文本的显性知识编码;随后引入外部情感分类语料库迁移得到子句的隐性知识编码;最后拼接两个知识编码,加入情感(原因)子句预测概率及相对位置,搭配Transformer机制融合上下文,并采用窗口机制优化计算压力,实现情感—原因对抽取。在ECPE数据集上的实验结果表明,该文提出的方法超过当前最先进的模型ECPE-2D。The existing emotion-cause pair extraction models fail in incorporating external knowledge.In this paper,we propose an emotion-cause pair extraction model based on knowledge transfer(ECPE-KT),in which the knowledge base is utilized to obtain explicit knowledge encoding of text.The implicit knowledge encoding of clauses is obtained by the transfer of external sentiment classification corpus.The final ECPE-KT is achieved via the concatenation of two knowledge encodings,the relative position and prediction probability for the clause-effect cause,a window mechanism and the Transformer.Experimental results show that the proposed method outperforms the state-of-the-art ECPE-2D method.

关 键 词：情感—原因对抽取 知识辅助 相对位置 预测概率 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]