关键词: 方面级情感分析, 依赖关系, 词性, 语法依赖树, 图注意力网络

Abstract: In response to several issues related to the interference of multiple aspect words in the grammar dependency tree, redundant information caused by invalid words and punctuation marks, as well as weak correlations between aspect words and corresponding sentiment words, a fused aspect-level sentiment analysis model is proposed, which combines strong correlated dependencies and concise grammar. Firstly, the model constructs a sentiment part-of-speech (POS) list to enhance the relevance between aspect words and corresponding sentiments. Then, a combined list incorporating POS and dependency relationships is constructed to eliminate redundant information of invalid words and punctuation marks in the optimized dependency tree. Next, the model combines this information with a graph attention network to model and extract contextual features. Finally, it interacts with the feature information of dependency relationship types, learns and fuses the features, and enhances the feature representation, enabling the classifier to efficiently predict the sentiment polarity of each aspect word. The proposed model is experimented and analyzed on four public datasets. Compared to the classical model CDT, which incorporates dependency tree grammar, the proposed model achieves an average improvement of 3.88% in prediction accuracy and 4.82% in Macro-F1 score. Experimental results demonstrate that the proposed model effectively enhances the connection between aspect words and sentiment words, resulting in more accurate prediction of aspect word sentiment polarity.

Key words: aspect-level sentiment analysis, dependency relationship, part-of-speech, syntactic dependency tree, graph attention network