Research On Malicious User Detection Algorithm Based On Missing Rating Value Imputation

With the rapid development of e-commerce,e-commerce platforms have become one of the main channels for people to shop.Trust in goods and services plays a crucial role in the successful operation of e-commerce platforms,and product evaluation has become a key reference factor affecting consumer purchasing decisions.Therefore,internet merchants will use various means to increase their sales and reputation,and hiring spammers is one of the commonly used methods.Merchants hire these spammers to publish false evaluations to interfere with users’ judgments of the true quality of products,thereby affecting consumers’ purchasing decisions and achieving the goal of increasing product sales.This not only reduces consumers’ shopping experience,but also disrupts fair market order.In recent years,the presence of spammers has posed an increasingly serious threat to the reliability of commodity evaluation systems.In order to solve this problem,researchers have proposed various methods to detect spammers,and a common type of method is to analyze user rating patterns.Scholars use user rating information to construct rating networks,and then calculate the reputation values of each user based on specific assumptions,treating users with lower reputation values as malicious users.However,the accuracy of such methods in sparse datasets needs to be improved,so it is necessary to improve the accuracy of spammers detection in sparse data.In order to improve the performance of spam detection in sparse data,this thesis proposes a universal framework for identifying malicious users,which involves filling in the missing ratings of low-degree users to increase their contribution to the product evaluation system,thereby improving the accuracy of current spam detection algorithms.The general process of using this framework to detect spammers is to first predict the missing scores of low-degree users,then extend them to existing datasets,and then apply the expanded data set to existing spam detection algorithms to identify malicious users.As a method for predicting the score values of low-degree users in the framework,this thesis predicts the missing score based on the similarity of user behavior characteristics in the experiment in Chapter 3.Subsequently,in the experiment in Chapter 4,the Restricted Boltzmann Machines and Deep Belief Networks are used to make more accurate predictions.After predicting the missing scores of each low-end user,the scores predicted by each method are extended to the existing data set and combined with the spam detection algorithm to identify malicious users.The experimental results show that filling in the missing ratings of low-degree users can effectively improve the accuracy and robustness of the spam detection algorithm based on reputation value to identify malicious users.The main contribution of this thesis is to propose a general framework for identifying malicious users in sparse data by filling in low-degree user rating information to alleviate the sparsity of the data set.The method proposed in this thesis can effectively reduce the false detection rate and demonstrate good robustness in different types of social networks.In addition,after using the framework proposed in this article to alleviate data sparsity,it can be combined with other reputation based spam detection algorithms used in the article to identify malicious users,which has high universality.