A Study Of User Credibility Evaluation Byintegrating Behavioral Analysis And Reviewcredibility Determination

The development of social media has brought various conveniences to our lives,and also promoted the upgrading and transformation of e-commerce in the information age.In mainstream e-commerce platforms,product reviews provide consumers with a way to understand the products' attributes.Due to commercial interests,some malicious users act as malicious review users deliberately devalue the value of competitors' products.It will also deliberately publish fake or malicious reviews to mislead consumers,it leads to the proliferation of a large number of low-quality reviews,and at the same time might lead to the loss of consumer property.User credibility,as a comprehensive measure of user influence and credibility in social networks,it has great importance in the entire evaluation system.Therefore,it has been the focus of e-commerce platforms to distinguish users with low credibility to differentiate users to optimize the online shopping environment.Although there are many evaluation methods for user credibility,most of them focus on the user-generated text features.Some methods do not distinguish between the quality of text and only limit the research to the scope of fake review,unable to consider the impact of low-quality text on the evaluation results.At the same time,there are many studies using link-based research methods to distinguish users,but these methods have the problems of reconstruction when new nodes join the network or user cold start problem,which leads to poor generalization ability of the method and low accuracy.Therefore,in terms of users,e-commerce providers not only need to distinguish between real users and users who make fake reviews,but also need to find a stable method to distinguish between highly credibility users and ordinary users in the data.Base on the current research problems,this paper proposes a user credibility evaluation model that combines user behavior analysis and review credibility judgment.This is a user credibility evaluation framework based on user features and user text.In order to solve the problem of low evaluation accuracy,this paper obtains the credibility of each user by deeply digging the user's social attributes,instead of simply using user attributes or constructing user relationship networks.In addition,in response to the problem of insufficient text processing methods,this paper designed two text quality evaluation methods based on user sentiment analysis to ensure that the text quality evaluation method has an efficient generalization ability.In addition,for user text credibility.This paper designs two methods to evaluate user text's quality through related research work on user sentiment analysis.It ensures that the method in this paper has efficient generalization ability.Finally,this paper describes the specific details of the model.In the user credibility evaluation model in this paper,the three aspects feature of user reliability,user activity and user text quality are extracted as the measurement basis of user credibility,and then a supervised learning model is constructed to predict user credibility.At the same time,in order to ensure the accuracy of the text quality judgment and prevent the cold start problem,this paper proposes two text quality evaluation methods,which are the text quality evaluation method based on the topic model and the text quality evaluation method based on deep learning.In order to solve the problem of imbalance between high-quality and low-quality reviews in real data,this paper also proposes an automatic selection sampling method to deal with the data.Finally,this paper uses the Amazon review data set to verify two text quality evaluation methods.In this paper,we uses the Dianping review data set and the YELP review data set to verify the user credibility evaluation model.The experimental results prove the effectiveness and advanced.