Research On Emotional Mining Based On User's Views In E-commerce

With the rapid development of Internet technology and e-commerce,human society is rapidly entering the era of “all-people online shopping”.The consumer's comment on the product contains the experience of the product,and also provides important information resources for feedback from other consumers and enterprise products.How to efficiently mine the opinions of consumers on products and related aspects in user comments has become a hot issue in the field of emotional mining analysis.However,due to the diversity and complexity of Chinese natural language expressions,emotional analysis and research of user comments has become more challenging.At present,the research on text sentiment analysis mainly focuses on fine-grained emotion mining.It can penetrate into many aspects of user evaluation,extract elements such as evaluation objects and evaluation words involved in the evaluation information,and analyze the sentiment orientation to accurately reflect the user's evaluation intention.However,the existing fine-grained text emotion mining still has some problems that have not been solved well yet.For example,the evaluation object and the evaluation word are incomplete,and the evaluation word's emotional tendency is ambiguous.How to effectively solve these problems and achieve fine-grained text sentiment analysis is receiving wide attention from the academic community.This dissertation first analyzes the development status and existing problems of text sentiment analysis at home and abroad,and then uses the syntactic analysis and semantic analysis of text emotion mining as the main line to study the perspective of user comments in e-commerce environment.To sum up,the main research work of this dissertation is as follows:(1)Product review data acquisition and pre-processing of the e-commerce platform.First,use Python crawler to get Jingdong Mall's mobile reviews,camera reviews and computer reviews.Then,the obtained comment data is cleaned.Finally,the data set is preprocessed by clause,word segmentation,syntactic analysis,semantic analysis,etc.,and data preparation for subsequent research.(2)A semantic-based evaluation information extraction method is proposed.Applying the conditional random field(CRFs)model,it is proposed to introduce syntactic analysis and semantic analysis in the conditional random field model to solve the extraction of evaluation objects and evaluation words.First,the user's attention analysis is performed on the pre-processed comment content,and the data set is semiautomatically labeled.Then,an evaluation information extraction model suitable for the e-commerce platform data is constructed.Finally,an experimental analysis was performed.The results show that the F-values of the evaluation-based evaluation information extraction methods and evaluation words reach 92.31% and 89.90%,respectively,and verify that the syntactic relationship and semantic relationship are correct and effective for the evaluation information extraction.(3)A method for extracting evaluation units based on semantic relations is proposed.Aiming at the problem of incomplete evaluation objects and evaluation words,this dissertation proposes a Syntactic Semantic and Multi-grained Conditional Random Fields(SSMCRFs)method combining syntax and semantic relations to solve the extraction of evaluation units.problem.The effectiveness of the SSMCRFs method and the effect of syntactic and semantic characteristics on the evaluation unit identification are proved by three experiments.(4)Product evaluation sentiment orientation analysis.Aiming at the ambiguity problem of evaluation words,the dependence relationship between the evaluation object and the evaluation word is considered.The SSMCRFs plus support vector machine is used to analyze the sentiment orientation of the evaluation unit.First,the manually labeled evaluation unit data set is <evaluation object,evaluation word,polarity> triplet.Then,an emotional sentiment analysis was performed on the obtained comment information.Finally,visual display is based on the results of the sentiment orientation analysis and the frequency of the evaluation unit.