An Aspect-Level Sentiment Analysis Method Based on Grammatical Knowledge and Message Passing Neural Network
Zhaolei Kang, Changtao Wang · 2024
Aspect-level sentiment analysis is committed to predicting the sentiment polarity of aspect-level words in sentences, and aspect-level sentiment analysis using the dependency analysis of graph attention network has been proven to be an effective method to improve the accuracy of aspect-level sentiment analysis. However, most of the existing methods use the dependencies between words to construct graphs, without taking into account the influence of other relationship features, which may lead to underutilization and ambiguity of grammatical knowledge. To solve these problems, this paper proposes an aspect-level sentiment analysis model based on grammatical knowledge, which can capture more grammatical information between words. Firstly, the emotion dictionary was used to enhance the role of emotion words in sentences, the root features of each word were extracted, the part-of-speech information was used to remove the redundant dependencies in the sentence to obtain the pruned syntactic dependency tree, and the message passing mechanism was introduced to input the pruned syntactic dependency tree into the graph attention network, so as to make better use of the correlation between different features. On this basis, by masking the aspect words in the input text, the model can learn the representation and emotional tendency of the aspect words more intensively. In this paper, a comprehensive experiment is conducted on a publicly available dataset to demonstrate its effectiveness. Experimental results show that the proposed model is superior to the strong baseline model.