Research Of Recommendation-based Anti-attack Reputation Mechanism On E-commerce

With the growing prosperity of electronic commerce, it is increasing that somemerchants sell defective or fake goods, and at the same time, the E-commercereputation evaluation system is not perfect is becoming an important factor restrictingthe development of electronic commerce. The current accumulation and average modelsare not well against attack on the reputation system. Some behaviors breaking the lawand rule, such as “brush credit” and malicious blackmail to merchants, continue despiterepeated prohibition. To solve the issues, the paper analyzed the present applicationsituation and development direction of the reputation system on E-commerce,referenced those grown-up P2P reputation models, researched the reputation mechanismunder the E-commerce environment, and proposed a recommendation-based anti-attackcentralized reputation model. The paper has done the work below:①Study the existing reputation model, analyze their computing methods of thereputation value, master the way of their trust evaluation, summarize the various factorsaffecting reputation value calculation and introduce briefly attack and defensetechniques for reputation systems. Lay the foundation of methods for that the paperproposed a new reputation model.②Analyze the present application situation and development direction of theonline reputation systems, which laid the foundation of directions for the new reputationmodel proposed by the paper.③Put forward a reputation model based on recommendation, introduce some newconcepts such as the time attenuation function based on the forgetting curve, consumerbehavior similarity and dynamic confidence for direct trust value, and so on, think aboutthose factors affecting trust evaluation containing the time factor, recommendation trustfactor, number of transactions, trading satisfaction and et al. Beyond that, the modelconsidered the initial trust constant between customers, which solved the problem ofcold start of the reputation system. The paper used real transaction data to do simulationexperiments. The results verified the effectiveness of the reputation model and defensecapacity against whitewashing, slandering and self-promoting attacks, and proved thatthe model had more advantages than the existing reputation models on E-commerce.④Improve further the proposed reputation model through adding the detectionand treatment mechanism to the inconformity feedback from normal buyers. The paper quantized the emotion of the qualitative text comment and calculated differencebetween the comment and grade. If the difference didn’t equal zero, the model wouldtreat the estimate as one inconformity feedback. Then the difference was used as offsetof quantitative score. The both constituted more accurate satisfaction value thatweakened effects of the inconformity feedback from normal buyers on reputation ofcommodities. The experiment proved the feasibility of the improvement direction andthe effectiveness of the improved model.In the last part of dissertation, on the basis of summing up the work of the whole,the further research direction is given.