基于CNN并融合注意力机制的充盈态三分句关系识别方法

Recognition Method of Fullness Three-clause Relationship Based on CNN and Fusion Attention Mechanism

汉语复句的关系类型识别是对分句间语义关系的判断,是分析复句语义的关键。充盈态汉语复句虽然可以通过句中提取关系词并根据关系词的搭配规则来识别语义关系,但是需要足够的语言学相关知识并且制定大量的规则才能将其划分到正确的种类。为了避免耗费大量精力创建规则,提出了一种基于注意力机制的卷积神经网络,使用卷积神经网络自动分析三个分句之间的内在联系,进而识别出复句的关系类别。使用所提方法在充盈态有标三分复句语料库上对复句关系类型识别的准确率为94%,通过实验证明了该方法的有效性。

Compound sentences relation recognition is to identify for the semantic relation of clauses,and it is the key to ana⁃lyze the semantics of the whole compound sentences.Although saturated Chinese compound sentences with three clauses can be identified for the semantic relation of clauses through extracted relative words and collocation rules of the relative words,it requires sufficient linguistic related knowledge and a large number of rules to recognize.In order to avoid spending a lot of effort to establish rules,this paper proposes an attention based convolutional neural network,which automatically analyzes the inner relationship be⁃tween the three clauses through the convolutional neural network,thereby recognizing compound sentences relation.Using the meth⁃od proposed,the accuracy of compound sentences relation recognition is 94%,and the effectiveness of the method is proved through experiments on the CCSCS.