Enhance gated graph neural network with syntactic for sentiment analysis

Abstract

Sentiment Analysis is of great application value to business, politics and research. Recent years, Graphic Neural Network (GNN) has attracted wide attention in the field of Natural Language Processing (NLP). Constructing textual representations of graph structure is an important step for sentiment analysis using GNN. Existing construction methods mainly include the method based on word co-occurrence and the method based on syntactic structure. The method based on word co-occurrence fully considers the global co-occurrence information of words, the syntactic-based approach considers the syntactic information. Combining the above two methods, in order to extract more information contained in the text, this paper proposes a syntactic edge-enhanced graph neural network model. this model fully considers the global word Co-occurrence information and overall sentence structure information by uses the word co-occurrence method to construct the text graph structure, combines the syntactic structure information. Then extract text representations through gated graph neural network, and learn key text information in combination with attention mechanism. Experimental results on multiple data sets show that the proposed model can achieve best performance in sentiment classification compared with the existing models.