Aspect-level sentiment analysis based on external knowledge two-channel graph convolutional networks

Aspect-Based Sentiment Analysis (ABSA) is a subtask of sentiment categorization that aims at mining the sentiment polarity of a given sentiment aspect in a sentence. Much of the recent research has focused on the use of the sentence itself, and the use of external knowledge is relatively lacking. Therefore, this paper proposes a dual-channel graph convolutional network (DKC-GCN) model that incorporates external knowledge. Firstly, using syntactic encoding as input, the probability matrix in all dependent arcs is used to encode syntactic information and capture rich structural information, and the graph convolution network is used to process the syntactic structural feature information of the sentence; then the sentiment dictionary is used to assign a sentiment score to each word in the sentence, highlighting the importance of the key sentiment words; after that, a lexical auxiliary matrix is designed to fully take into account the effect of the lexicality of the words on the classification of the sentiments; Finally, the accuracy of aspectual sentiment pairs is improved by fusing sentence information and external knowledge information through a cross bimodal attention mechanism. The study is evaluated on three public datasets with F1 value and accuracy, and the final experimental results have improved accuracy and F1 value, verifying that the model is effective.