Aspect Emotion Classification Based On Semantic Attention Mechanism And Capsule Network

Emotion analysis plays an important role in the field of natural language processing.It can explore the emotional polarity of text,such as positive,negative or neutral.With the development of social networks,the role of Internet users is gradually changing.More and more people tend to express their opinions on the Internet.It is very useful for individuals or enterprises to refer to their opinions.Valuable information can be obtained from comments on various platforms,which is conducive to better improvement and service.These comments resulted in a complete text.These texts contain a wealth of usable information that is useful for understanding what users are thinking when evaluating social networks or products.In recent years,aspect-level emotion classification has become more and more popular and has been widely used in real life.In the existing studies,on the one hand,researchers mainly consider the extraction of context semantics,and tend to ignore some semantic information based on syntactic dependency.On the other hand,the lack of aspect-level marker data is also a challenge in aspect-level emotion classification.In this thesis,we propose a hybrid model based on semantically dependent attention mechanism and capsule network(SATTCap).We introduce additional emotional document knowledge for multitask learning,mining useful emotional information,and auxiliary aspect-level emotional classification.In addition,semantic reconstruction of the text is carried out,and local attention mechanism is introduced to extract the deep-level feature information based on syntax dependence,and then aspect routing method is adopted to encapsulate the deep-level aspect semantic representation into semantic capsules.On SemEval datasets experiments show the effectiveness of the method,and compared in this thesis,differences in the effect of the analysis model and the traditional classification model,the results show that compared with the base class model of emotion classification method,this thesis used based on the analysis of semantic focus mechanism and capsule network hybrid model can more accurately identify polarity text emotions.