基于知识图谱与双重图卷积的情感分类模型

Emotion Classification Model Based on Knowledge Graph and Dual Graph Convolution

摘要: 在情感分析的任务中，获取文本语义信息、句法信息和外部知识是至关重要的。但是目前的算法模型中，文本与外部知识的关联较少，导致语义表示不够全面。所以本文提出一种基于知识图谱与双重图卷积的情感分类模型。模型先利用LSTM算法提取文本的上下文信息，同时采用知识图谱卷积网络获取与文本相关的实体的外部知识；再通过实体对齐和对齐预测，进行动态融合文本嵌入与知识嵌入；最后将融合了文本和知识的信息作为输入，使用句法图卷积网络和语义图卷积网络获取句法和语义信息，实现情感分析。根据实验表明，该模型融合了文本与实体知识，增强了语义语法信息，有效提升了情感分类的准确性。

Abstract: In sentiment analysis tasks, acquiring textual semantic information, syntactic information, and external knowledge is crucial. However, current algorithmic models often lack sufficient association between text and external knowledge, resulting in incomplete semantic representation. Therefore, this paper proposes a sentiment classification model based on a knowledge graph and dual graph convolution. The model first employs the LSTM algorithm to extract contextual information from the text, while using a knowledge graph convolutional network to acquire external knowledge related to entities in the text. Then, through entity alignment and alignment prediction, it dynamically integrates text embeddings and knowledge embeddings. Finally, the fused information combining text and knowledge is used as input, and a syntactic graph convolutional network and a semantic graph convolutional network are applied to capture syntactic and semantic information, thereby achieving sentiment analysis. Experiments demonstrate that this model effectively integrates text and entity knowledge, enhances semantic and syntactic information, and significantly improves the accuracy of sentiment classification.