基于增强CNN获取丰富匹配特征的文本匹配方法  

RICH MATCHING FEATURES BASED ON ENHANCED CNN FOR TEXT MATCHING

作　　者：汤福平 陈红英[1] Tang Fuping;Chen Hongying(School of Computer,South China Normal University,Guangzhou 510631,Guangdong,China)

机构地区：[1]华南师范大学计算机学院,广东广州510631

出　　处：《计算机应用与软件》2025年第2期210-215,共6页Computer Applications and Software

基　　金：广东省自然科学基金项目(2020A161610445)。

摘　　要：在文本匹配问题中,为了获取两个文本表达中的匹配特征,提出一种基于增强的卷积神经网络来获取丰富匹配特征的文本匹配方法。通过多层对齐机制捕获句子交互特征,同时采用增强CNN获取的关键词特征进一步捕捉句子关键特征信息;接着通过一种基于门控机制的融合方法,融合句子内的关键特征和句子间的匹配特征,由门控网络选择加强关键匹配信息和弱化与匹配文本无关的局部信息。实验结果验证了该模型的有效性。To obtain matching features of two text expressions in the task of text matching,a text matching method based on enhanced convolutional neural network to obtain rich matching features is proposed.We captured sentence interactive features through the multi-layer alignment mechanism.The key characteristics obtained by enhanced CNN were used to capture the key features of sentences.Through a fusion method based on gating mechanism,the key features within sentences and the matching features between sentences were fused,and the gating network chose to strengthen the key matching information and weaken the local information unrelated to the matching text.Experimental results show that the model is effective.

关 键 词：自然语言处理 文本匹配 注意力机制 门控网络 深度学习 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]