Aspect-level sentiment analysis based on attention-directed graph convolutional networks
Shaoguo Cui, Yaohuan Luo, Guangping Hu, Aodi Wang · 2022
The aspect level emotion analysis task is mainly used to judge the emotional tendency of aspect words in the text, which belongs to the fine-grained emotion analysis task of the emotion analysis task. The existing sentiment analysis method is to classify emotions through a graph neural network. Because the syntax dependency tree generated by the use of tools have some errors, which are propagated to the construction of graph neural network, so the performance of the algorithm is affected. In this paper an attention-guided graph convolutional network model is proposed to solve the problem of errors in syntactic dependency tree information. Firstly, BiLSTM is used to obtain deep sentence information between aspect words and contextual words. Secondly, the semantic information with syntactic dependency tree structure is captured by the graph convolutional network, and the global semantic information of sentences is captured by the graph convolutional network with an attention mechanism. Finally, the two semantic information is combined. Global semantic information is used to mitigate errors in semantic information with syntactic dependency tree structure. The model was tested on Rest14, Rest16, and Laptopl4 data sets. The experimental results showed that the accuracy and F1 value increased by 1.12% and 1.39% on average compared with the baseline model.