融合知识图谱的文本情感分析

目前研究文本情感分类往往只关注文档内容,对文本信息缺失和歧义等特点考虑不够,导致模型性能较低,为此提出一种融合知识图谱的用户和产品层次化注意力网络.首先通过双向长短期记忆网络获取词汇层隐藏表示,利用具有哨兵注意力机制将知识图谱中的知识与文本相结合获取词汇的知识感知状态向量;其次利用注意力机制结合用户和产品信息;最终利用归一化指标函数识别情感极性.结果表明,该方法在Yelp和IMDB数据集上的精确率和均方根误差优于基准方法,验证了模型的有效性.

Emotional classification of combining knowledge graph

The purpose of document-level sentiment classification is analyze the opinions and emotional polarity based on the text generated by people. Recently, most of the work often only focuses on the content of the document and the characteristics of the lack of text information and ambiguity are not considered enough. The same word may represent different semantic information in different contexts and suffered from lower model performance. Hence, This paper proposes a sentiment analysis method that combining the knowledge graph. The method first obtains the hidden layer representation of word through BiLSTM, and uses the sentimental attention mechanism to combine the knowledge in the knowledge graph with the text to obtain the vocabulary knowledge. Perceive the state vector, and then use the attention mechanism to combine user and product information, and finally use the softmax function to identify the emotional polarity. The experimental results show that our model obviously outperforms other state-of-the-art methods on IMDB and Yelp datasets.