Aspect-based Sentiment Analysis via Dual-channel Graph Convolutional Network
At present,most aspect-level sentiment analysis aggregates the emotional information between aspects and context from syntactic dependency trees by constructing graph neural networks on syntactic dependency trees,and lacks the mining of semantic correlation of text and emotional knowledge in text.To this end,we propose a dual-channel graph convolutional network(DC-GCN)model that integrates semantic information and text self-emotional information,which is built by the SrG(Sentiment-relationship Graph)and the SeG(Sentiment Graph)with contextual semantic phase information.Specifically:firstly,the attention mechanism is used to capture the semantic relevance of the text and integrate it into the syntactic dependency,and the SrG with contextual semantic information is designed;secondly,the emotional information of words in SenticNet dictionary is integrated to construct the SeG;finally,a single-channel graph convolutional neural network is constructed based on SrG and SeG,and then the two single-channel graph convolutional networks are fused into a two-channel graph convolutional neural network to fuse the syntactic features of the text with semantic information and the emotional characteristics of the text itself.The four public datasets of Twitter,Rest14,Lap14 and Rest16 a-chieved better results than that of the comparison model,and the Marco-F1 values of the proposed model were0.95%,1.24%,0.62%and 2.75%higher than that of the comparison model,respectively.
万方数据
维普
