基于深度交互的文本匹配模型研究  

Research on Text Matching Model Based on Deep Interaction

作　　者：余传明[1] 薛浩东 江一帆 Yu Chuanming;Xue Haodong;Jiang Yifan(School of Information and Safety Engineering,Zhongnan University of Economics and Law,Wuhan 430073)

机构地区：[1]中南财经政法大学信息与安全工程学院,武汉430073

出　　处：《情报学报》2021年第10期1015-1026,共12页Journal of the China Society for Scientific and Technical Information

基　　金：国家自然科学基金重大课题“国家安全大数据综合信息集成与分析方法”(71790612);国家自然科学基金面上项目“面向跨语言观点摘要的领域知识表示与融合模型研究”(71974202);中南财经政法大学中央高校基本科研业务费专项资金资助项目(202012020,2722021AJ011)。

摘　　要：针对文本匹配在信息检索、文本挖掘等领域的广泛应用,本文提出一种具有良好泛化能力的深度交互文本匹配(deepinteractiontextmatching,DITM)模型。基于匹配-聚合框架,DITM模型以编码层、共注意力层和融合层为交互模块,多次循环交互模块获取深层次的交互信息,经过多角度池化提取信息以预测文本对之间的关系。相比于基线方法,针对观点检索、答案挑选、释义识别和自然语言推理四个文本匹配任务,DITM模型在相应的数据集上均取得了最好的效果。本研究结果对于促进文本匹配模型在情报领域的实践具有重要意义。For the application of text matching in information retrieval,text mining,and other research fields,a deep inter‐active text matching(DITM)model with good generalization ability is proposed.Based on the matching-aggregation framework,the encoder layer,co-attention layer,and fusion layer are used as an interaction module.The interaction pro‐cess is iterated multiple times to obtain the in-depth interaction information.Finally,information is extracted through multi perspective pooling to predict the relationship between text pairs.Compared with baseline methods,the proposed approach has achieved best results on four text matching tasks,namely opinion retrieval,answer selection,paraphrase identification,and natural language inference.The experimental results are of great significance to promote the practice of text matching models in the field of information.

关 键 词：文本匹配 深度学习 深度交互 注意力机制 神经网络 

分 类 号：TP391.1[自动化与计算机技术—计算机应用技术]