A slope one algorithm based on item activeness and uncertain neighbors
Limei Sun, Yue Li, Ejike Ifeanyi Michael · 2017
The Slope One algorithm is one of the most widely used algorithms in personalized recommendation systems. It is efficient and easy to maintain, but it doesn't take the item similarity into consideration when it works. For instance, if the item similarity is low, there will be a deviation in the recommendation. So the ratings should be denoised before prediction. This paper analyses the item similarity and the item activeness in the nearest-neighborhood. Based on Slope One algorithm, a new collaborative filtering model which combines the item activeness in uncertain neighbors is presented. Firstly, the neighbors of each item are dynamically selected according to the item similarity and activeness in uncertain k-nearest neighbors matrix. Secondly, the average rating deviation between items is only generated from the selected neighboring items. Lastly, the target ratings are predicted by linear regression model. Experiments on the MovieLens dataset show that the proposed algorithm improves recommendation quality and gets better prediction accuracy than some other collaborative filtering algorithms.