Time Dependent Reputation Algorithm Based On Rating Systems

With the rapid development of information sharing platform,online rating systems become very popular in the field of e-commerce domain.In the rating applications,two kinds of entities exist ubiquitously: user and object.Users can rate for the objects,then system will calculate the final scores for the corresponding objects based on the user ratings,and the scores will decide object's ranking.Therefore,how to improve the accuracy of online rating system is a hot research topic in recent years.In order to improve the ranking accuracy of online rating systems,this thesis studies the reputation mechanism under the e-commerce platform,and proposes a time dependent reputation algorithm based on rating datasets.The main structure of this thesis is organized as follows:(1)We review the existing reputation models.The computing methods of the reputation,the way to evaluate trust relationship,the various factors affecting reputation estimation are discussed.We also introduce some attack and defense techniques for online reputation systems.In this part,the research direction of the online reputation systems is also pointed out.(2)By researching the factors that affect the user reputation and object quality from network perspective,we propose a time-based reputation evaluation algorithm to detect user reputation and object quality.In this algorithm,the user behavior weight factor,the reputation accumulation process and the object quality accumulation process are introduced to refine the user reputation and object quality.Furthermore,we try to understand the reputation model for the influence ability of the user reputation,including combine with up to date reputation models.Finally,we propose a new reputation model which is more applicable to the online rating systems.(3)In order to verify feasibility and effectiveness of our algorithm,by using three real datasets(Amazon,Netflix and Movie Lens),common-used metrics(such as accuracy,recall and AUC,F value)are presented in the thesis.Our results show that our algorithm is superior to the credibility and stability.What's more,quality and the ranking results have greatly improved.