Abstract:

The classical Slope One algorithm employs the linear regression model to generate recommendation. However, the recommendation suffers from decreased precision due to the noise data emerged in item score deviation table construction. We propose a Slope One collaborative filtering recommendation algorithm to solve this problem based on local nearest neighbors. Neighbor users set dynamically change along with the change of the target item by calculating target users based on the neighbor users of different target items. The average deviation between items is further optimized and recommendation is generated according to the neighbor user data of different target items. Experimental results on MovieLens dataset show that the improved algorithm can promote the prediction accuracy of the recommendation.

Key words: collaborative filtering, recommendation system, local nearest neighbors, Slope One