To solve the problem that the existing aspect-level sentiment analysis methods lack syntactic constraints and lexical information,the syntactic dependency tree and knowledge graph were innovatively integrated to encode sentences,and a graph neural network model combining lexical syntax was proposed.The syntactic information in the syntactic dependency tree and the lexical information in the knowledge graph were extracted using the graph neural network,and the words with higher importance were captured using the position encoding module and the mask weighting module.The two features were combined to obtain a fusion syntactic lexical text representation of information for sentiment classification.Experimental results on three public data-sets validate the effectiveness of the model.
维普
万方数据