Research On Sentence Sentiment Analysis Using Graph Attention Network With Social Netowrk Information

Sentiment analysis is one important domain of Natural Language Processing.Traditional sentiment analysis did not take documents' relations into consideration.But in real life,there is relationships in two documents.In the condition of social network,we can get documents' relations according to users' follow information.There is a little work focus on using documents' relations,some methods use user similarity to measure documents similarity.However these methods can not differentiate the difference of documents which were post by same user.To solve the problem above,this paper proposes to create documents relations graph,and using Graph Attention Network extract the structure future by using Graph Attention Network can calculate the convolution of un Euclidean graph.The model consists of two modules,one of the module extract sentiment future by using Graph Attention Network and dependency analysis.The other module can extract documents relations by using Graph Attention Network.And by controlling the iterations of Graph Attention Network,we can control which order nodes' future will be aggregated.And because of the attention mechanism the model can allocation the weights of each neighbor node,thus we can solve the problem above.In order to verify the effectiveness of the model above,we compare the model with some other models in one dataset,The result shows that the model we proposed perform better than the other baseline models.