基于网络表示的半监督问答文本情感分类方法

针对新颖的问答形式的文本展开研究，提出了基于网络表示的半监督问答文本情感分类方法，通过构建异构网络的联合学习提升半监督问答文本的情感分类性能。首先，通过分析标注和未标注样本构建一个异构网络，具体包括词-词网络、问题和答案文本-词网络、情感标签-词网络；其次，利用该异构网络学习获得词向量；最后，将学习到的词向量应用于目前性能最优的分层匹配情感分类模型（hierarchical matching network,HMN）中。实验结果表明，提出的方法在处理问答文本情感分类任务上具有一定优势。

A semi-supervised sentiment classification method towards question-answering text based on network representation was proposed. And the performance of semi-supervised sentiment classification of question-answering text was improved by constructing joint learning of heterogeneous network. A heterogeneous network was firstly constructed by analyzing labeled and unlabeled samples, which was composed of word-word network, question and answering document-word network, and sentiment-word network. Secondly, the heterogeneous network was used to learn word embedding. Finally, the word embedding was applied to the currently best-performing hierarchical matching network. Empirical results showed that the proposed method had certain advantages in processing the sentiment classification task on question-answering text.