Collaborative filtering: user similarity in Slope One algorithm
Ruisheng Zhang, Qidong Liu, Rongjing Hu, Huiyi Ma, Yongna Yuan · Lanzhou University Institutional Repository · 2014
Slope One algorithm was kind of item-based collaborative filtering. As a recently proposed algorithm, Slope One has several desirable properties such as simple, being updatable on the fly, efficient to compute. Even though the Slope One algorithm was widely used on large data sets, it performs not so well in terms of accuracy. The reason is that the Slope One algorithm ignored an important factor: similarity between users. Using MovieLens data, we find that the Slope One algorithm is only applicable to similar user groups, the noise which the users have the opposite profile leads to the low accuracy. Based on our findings, we proposed an improved weighted Slope One algorithm which is more accurate. In the end, the results of the experiments on MovieLens data sets confirmed the effectiveness of our methods.