Effect of User’s Judging Power on the Recommendation Performance

In most B2C E-commerce websites, recommender systems make recommendations for each individual user based on his/her historical rating behaviors. Previous literatures focus on the overall performance of recommender system, while the performance of individual level receives little attention. In this paper, we discover that recommendation algorithms perform better on users who have strong judging power, and vice versa. We test our conclusion on three benchmark data sets, namely MovieLens, Netflix, and Amazon, which further provide evidence of the validity of our finding. Moreover, our finding may provide some guidance for designing recommendation algorithms more efficiently by concerning users' different judging power.