An improved slope one algorithm for collaborative filtering

Abstract

The slope one scheme is a rating-based recommendation algorithm, and it is simple, efficient, easy to implement. However, the slope one scheme suffers from both data sparsity and new item problems, which seriously affect the performance of recommender systems. To solve these problems, we propose an improved slope one algorithm for collaborative filtering. In our algorithm, the new item problem is dealed with by introducing content similarity computation into the slope one scheme. According to the idea of item-based collaborative filtering algorithms, the target user's rating to the target item can be predicted based on the ratings that the target user has rated and the content similarities of items. And clustering algorithm is adopted to tackle the problem of data sparsity. By merging the set of items into several clusters based on the item rating data, the target user's rating to the target item can be predicted based on which cluster the target item belongs to. The final rating of the target user to the target item is the linear combination of the above two ratings. Experiments on the Movielens dataset show that our approach outperforms other three slope one algorithms and two traditional collaborative filtering algorithms on the prediction performance.