Research On Rating Algorithm For The Reputation System

With the popularity and rapid development of Internet,great changes have been brought to people's daily life.Modern society is entering the era of Internet Plus.E-commerce plays an important role in the Internet Plus,and has gained great success in recent years.Trust problem has always been a great challenge to the Internet,and it is also the bottleneck for the further development of E-commerce.A quantifiable trust guarantee mechanism,the reputation system,is introduced to solve the above problems.Reputation systems collect evidence about the properties of products,analyze and aggregate the evidence,and disseminate the aggregated results as reputation scores.The introduction of reputation system promotes the development of E-commerce.However,the temptation to fake reputation ratings for benefits is also growing significantly,as the reputation system becomes more and more popular.Malicious users attempt to manipulate the reputation scores of products,by the means of giving a large number of positive evaluations to the products they can benefit from,or giving a large number of negative evaluations to their competitors.How to defend against attacks from malicious users,in the case of even if there are non-fair evaluations,reputation scores consistent with the true quality of products can also be calculated,and this is a big challenge to reputation system as well as the emphasis of this paper.Specifically,mainly working contents of this dissertation as follows:(1)Research on basic iterative algorithm.Most traditional rating algorithms are based on a detection mechanism for malicious users or non-fair ratings.These methods try to find out the malicious groups or regulation of non-fair and delete them after matching,through monitoring system status and user behavior persistently.Thus,their system costs are expensive.In this paper,we introduce an iteration-based rating algorithm,which calculates with both fair ratings and non-fair ratings and eliminates the impacts of non-fair ratings during the iteration steps.This method performs well and robust to malicious attacks.(2)Research on improvement of basic iterative rating algorithm.In view of the rating diversity and the user preference problem,we propose a quantiles-based iterative rating algorithm.To be specific,in the step of rating preprocessing,we eliminate the preference of user by means of statistical analysis of user ratings to find its distribution,standardize the distribution,and then transform the ratings according to corresponding quantile.Then combining with the bucket mechanism,we can also resolve the problem of rating diversity.For the trust degree of user during the iterative steps,we put forward an optional trust initialization strategy based on clustering method.Finally,aiming at the problem of user interest changes over time,namely the time-shift characteristics of ratings,we propose the thought of iteration based on time weight function.By several experiments on the real film rating datasets of MovieLens and the comparison with existing algorithm,we verify the effectiveness and superiority of our algorithm.