Research On Aspect Level Emotion Analysis Method Based On Attention Mechanism

With the rapid development of the Internet,people like to express their opinions and comments on various events on the Internet,so there will be a large number of comment texts reflecting users’ emotional views,which contain abundant information.As a technology,Aspect Emotional Analysis can judge the emotional tendency of all aspects of the text,and it can provide comprehensive and scientific decision-making basis for the government,enterprises and consumers.Based on this background,this thesis studies aspect-level sentiment analysis and proposes two models.In the research of aspect-level sentiment analysis,it is a hot issue to use attention mechanism to model aspect words and context,but most researches calculate attention by calculating the average vector of aspect words and context,which will introduce unnecessary noise and affect the classification accuracy of the model.To solve this problem,this thesis proposes an aspect level affective analysis model based on iterative attention mechanism(IAM),IAM model uses the attention expression of aspect words to obtain the attention expression of context instead of the average vector of aspect words,which solves the influence of other aspect words or prepositions on the classification accuracy of the model to a certain extent.IAM model does not fully consider the positional relationship between context and aspect words when seeking attention expression of aspect words and context,which may lead to semantic errors and other problems.To solve this problem,this thesis introduces location prior knowledge into IAM model,which makes the contextual attention expression of specific aspect words pay more attention to the contextual words near the aspect words,and assigns more weight to the contextual words which are closer to the aspect words.Inspired by memory network,this thesis proposes an aspect level affective analysis model integrated memory network and attention mechanism(IAM-MEM),which integrates memory network and attention mechanism.Memory network can realize external storage,with multiple computing layers(hops).Compared with IAM model,IAM-MEM model assigns the weights of aspect words and context words by combining the semantics of context,which improves the weight assignment of traditional attention mechanism to some extent.The IAM model and IAM-MEM model proposed in this thesis are compared with other emotion analysis methods.Experiments show that the accuracy of the model proposed in this thesis has been improved compared with other models,and its effectiveness has been verified.