Abstract—How can recommendation accuracy be improved in a personalized news recommender system? In this paper, we utilize new attributes in addition to standard recency and popularity such as Reading Rate and Hotness. These attributes are defined in the user profile and news metadata and are used in neighborhood-based collaborative filtering in news recommendation. We analyze the proposed attributes in the user profile construction and the news metadata enrichment by exploring similar users’ interests in news reading. This is carried out via experiments using k-means. We then compare the precision, recall and F1-score in a series of experiments to evaluate the news recommendation with these attributes. The experimental results show that the proposed attributes improved the accuracy in news recommendation with higher precision and F1-score. We conclude that Reading Rate and Hotness in news have a significant impact on personalized news recommendation systems.

Index Terms—news recommendation, collaborative filtering, news hotness, user profile, news metadata

Cite: Asghar Darvishy, Hamidah Ibrahim, Aida Mustapha, and Fatimah Sidi, "New Attributes for Neighborhood-based Collaborative Filtering in News Recommendation," Vol. 7, No. 1, pp. 13-19, November, 2015. doi: 10.12720/jetwi.7.1.13-19