A New Asymmetric User Similarity Model Based on Rational Inference for Collaborative Filtering to Alleviate Cold Start Problem

摘要

For user-based collaborative filtering, the similarity methods used to calculate the target user's neighbors are very important. More similar neighbors lead to better recommendations and more accurate results. There are a lot of similarity methods till now, but there is still a room for improvement, especially when the data is sparse. It is well known that sparse data can easily lead to cold start problems. The performances of most traditional methods are disappointing under cold start conditions. In order to get a better performance under the cold start conditions, we proposed a new similarity method based on the idea that users with similar interests in the past will show similar tastes in the future. While considering similarities between items and rational inferences, the proposed method focuses on how to utilize more ratings information. At the same time, in order to reduce the time spent on calculations and reduce the impact of excessive ratings information, we have limited the range of items neighbors through experiments. Besides, the proportion of co-rate items to personally rated items is different from each user, base on which the asymmetric factor is considered. Experiments on the dataset MovieLens prove that the proposed method outperforms state-of-the-art methods.