A Novel Neighborhood Calculation Method by Assessing Users' Varying Preferences in Collaborative Filtering

To recommend an item to a target user, Collaborative Filtering (CF) considers the preferences of other similar users or neighbors. The accuracy of the recommendation depends on the effectiveness of assessing the neighbors. But over the time, the mutual likings of two individuals change; hence, the neighbors of the target user also should change. However, this shifting of preferences is not considered by traditional methods of calculating neighborhood in CF. As a result, the calculated set of neighbors does not always reflect the optimal neighborhood at any given point of time. In this paper, we argue for considering the continuous change in likings of the previous similar users and calculating the neighborhood of a target user based on different time periods. We propose a method that assesses the similarity between users in the different time period by using K-means clustering. This approach significantly improves the accuracy in the personalized recommendation. The performance of the proposed algorithm is tested on the MovieLens datasets (ml-100k and ml-1m) using different performance metrics viz. MAE, RMSE, Precision, Recall, F-score, and accuracy. keywords: Recommendation systems, Collaborative Filtering, Top-N neighbor, K-means clustering, User similarity, Time variance, Personalized recommendation