Research And Implementation Of Aspect-Level Sentiment Analysis System Based On Knowledge Supplement

Research and implementation of aspect-level sentiment analysis with the development of Internet platforms based on user content creation,the amount of text content on the Internet is rapidly increasing every day.When people output what they see and hear on the Internet,they also express their emotional tendencies.It is very meaningful to effectively analyze the text with a subjective sentiment on the Internet.In recent years,fine-grained sentiment analysis has received a lot of attention.Aspect-level sentiment analysis aims to predict the sentiment polarity of sentences given an certain aspect.Aspect-level sentiment analysis based on external knowledge has always been a concern of industry and academia.By adding external knowledge,a supervisory signal can be added to the model.How to effectively utilize external knowledge is a problem to be solved in aspect-level sentiment analysis.This paper mainly studies how to perform aspect category sentiment analysis based on external knowledge.The main research points are the following three aspects:First,an aspect category sentiment analysis model SDAN based on dependency parsing is proposed.First,use the auxiliary task aspect category detection to obtain the indicator words,prune and reconstruct the original dependency tree of the sentence according to the indicator words,and form the dependency tree for aspect category sentiment analysis.Finally,SDAN uses the syntax graph attention network G-GAT to get the final sentiment polarity.Experimental results applying SDAN to three public datasets demonstrate its effectiveness.Second,we propose a model EKAN for aspect category sentiment analysis using knowledge graphs.The first is to construct an external knowledge supplementary graph.the original sentence is parsed into the form of a graph through syntactic dependency analysis,and then the external extended knowledge is added to the graph according to certain rules to complete the expansion of the original sentence.Afterward,the additional attention mechanism is used to measure the relative importance of the extended knowledge and the similarity of the extended knowledge to the original sentence,and to a certain extent,disambiguate and reduce the noise introduced from the knowledge graph.Secondly,find the path from the aspect category words in the knowledge graph to the words in the sentence,and solve the problem of implicit aspect words to a certain extent.Finally,the sentiment polarity in terms of aspect is obtained by fusing the original sentence vector representation and the supplementary graph node representation.Thirdly,this paper designs and develops an aspect category sentiment analysis system to give play to the value of research in the application.The main functions of the system include data management,text preprocessing,aspect category detection,aspect category sentiment analysis.