Useful acquiring ratings for collaborative filtering

Abstract

For any product recommendation systems, the most important thing is to improve the accuracy of prediction of customer preferences on products. If there is not enough information of a product, especially when a new product is introduced into the system, it is difficult to recommend the product to other customers. If we can select few customers to rate this product we may predict more accurate. We term this additional information as useful acquiring ratings. In this paper, we propose a useful acquiring rating sampling algorithm to select these potential customers. Using the Netflix Prize dataset, we experimented with our proposed method, uniform random sampling method, degree-based sampling method and the active learning sampling methods. The results showed that our method generally outperformed other three methods.