Personalized Implicit Sentiment Analysis Fusing Multi-source Heterogeneous Commonsense And User Knowledge

Personalized implicit sentiment analysis is one of the more difficult research hot area in the field of text sentiment analysis.It has great academic research value and has great application prospects in the fields of enterprise reputation analysis,social opinion calculation and content-based recommendation.This paper focuses on the research of personalized implicit sentiment analysis from the aspects of characteristics and difficulties analysis,corpus construction,knowledge representation and fusion methods.The main research studies is as follows:(1)The construction of user personalized implicit sentiment analysis corpus.Detailed user information is the basis for personalized implicit sentiment analysis.Currently,the public implicit sentiment corpus is usually only used to label the implicit sentiment expression itself,lacking relevant user information.For this reason,this paper constructs a user personalized implicit sentiment analysis corpus,which includes implicit sentiment content information,user basic information and user social information.On this basis,a part of explicit sentiment sentences is reserved,and a personalized universal sentiment analysis corpus is built to verify the generalization ability of the model in the general sentiment analysis task.(2)Implicit sentiment analysis based on user multi-knowledge fusion.Traditional implicit sentiment analysis methods usually model the information of implicit sentiment text itself,without considering the subjective differences of implicit sentiment.In this regard,this paper proposes a personalized implicit sentiment analysis task,and proposes an implicit sentiment analysis model based on heterogeneous user knowledge fusion.Based on the dynamic pre-training model,it combines the sequence model and the graph neural network model to learn the representation of heterogeneous user knowledge respectively,and integrates with the text information to realize the subjective difference modeling of implicit sentiment.The experimental results show that user multi-knowledge has a good promotion effect on implicit affective analysis.(3)Personalized implicit sentiment analysis integrating heterogeneous common sense and user knowledge.On the one hand,in order to solve the problem of vague expression in implicit sentiment analysis,it is necessary to combine common sense knowledge to carry out deep semantic understanding.On the other hand,common sense knowledge can supplement missing information in user content.Therefore,this paper proposes a heterogeneous common sense and user knowledge fusion model.Based on the dynamic graph representation learning mechanism,it proposes a model that can capture high-level common sense knowledge.It enhances knowledge and expands semantics from the two dimensions of implicit sentiment expression and user internal knowledge to solve the problems of vague expression and user information missing in personalized implicit sentiment analysis.The experimental results show that the combination of commonsense knowledge and personalized implicit sentiment analysis can further improve the effect of implicit sentiment analysis.