Contribution-based User Reputation Modeling in Collaborative Recommender Systems

User reputation is an important factor in collaborative filtering approaches, in which every user may be another's nearest neighbor and may provide recommendations. In order to generate accurate results, the recommender system assigns different weights to users according to their reputations. However, existing methods for evaluating user reputation consider only the number of feedback ratings and cannot fully reflect user experience and credibility. In this paper, we propose a method of assigning reputations to nearest neighbors on the basis of their contributions in service recommendation. The contribution is evaluated by two factors: rating accuracy and importance of influence. Rating accuracy represents the consistency in perception between a neighbor and a consumer, and importance of influence determines the weight of a neighbor's rating. Using actual contributions as bases for modeling reputations prevents a neighbor from being penalized or awarded for another's poor or excellent recommendation behaviors. Experiment results show that the proposed approach can fairly assign reputation to each neighbor, thereby enhancing the credibility of a recommender system.