Review Credibility Model Construction And Multidimensional Visual Analysis Of Topic Sentiment

With the popularity of consumer e-commerce,online reviews,as a new form of brand word-of-mouth publicity,play an important role in consumer decision-making.How to mine the valuable information behind the review data and analyze the emotional changes of consumers,so as to provide effective and objective measurement standards for review data is still an urgent problem to be solved.Traditional review analysis methods cannot intuitively express the potential information behind the text,which requires the combination of data analysis and visualization techniques.Aiming at the above problems,this paper proposes an interactive visual analysis method integrating credibility model from the perspective of individual differences,topics and sentiments of review users by using opinion mining,visualization and human-computer interaction technologies.Make the best choice for consumers,service innovation for enterprises,and provide reference opinions for the daily maintenance of review websites.Firstly,in order to solve the limitations of Latent Dirichlet Allocation(LDA)topic model in semantic relationship and context correlation,a review text topic clustering method based on pre-trained language model is proposed,which mainly includes four modules:semantic feature extraction based on Transformers,dimension reduction,clustering,and improved Term Frequency-Inverse Document Frequency(TD-IDF)topic creation.The effectiveness of the proposed method is verified by comparing with typical topic models.Then,according to the consistency curve of topic semantics,the optimal number of topics is selected as 6,and the topic meaning is obtained to pave the way for subsequent work.Secondly,the influence indicators of the credibility model are selected and extracted,including the characteristics of the review itself,sentiment characteristics and commenter characteristics.The sentiment feature is divided into topic mining and sentiment analysis from the semantic level.Sentic Net sentiment dictionary is used to construct the sentiment vector of the text topic,and the sentiment value of each topic is calculated.A review credibility model method based on emotional features and commenter information is proposed.In order to further improve the performance of the model,the credibility discriminant formula of reviews is defined.Four different experimental models is constructed to evaluate the performance of the model on six machine learning algorithms,which proved the effectiveness of the credibility model.Finally,the data dimensions such as topic,sentiment vector,credibility and commenter information obtained in the previous stage are displayed and analyzed from multiple levels such as time,user difference,sentiment evolution,topic proportion and credibility.A multidimensional visual analysis method of topic sentiment based on credible model is proposed.From the aspects of visualization design,layout method and interaction correlation,some novel visualization views such as fireworks diagram and galaxy diagram are designed.Finally,the effectiveness and innovation of the visual analysis method are proved through the comparison of visualization effects,case analysis and user experiments by using multidimensional correlation information exploration.