A hybrid approach for personalized news recommendation with ordered clustering algorithm, rich user and news metadata

One of the most commonly used of online services is news reading. A key challenge is selecting news articles from millions of sources considering user behavior and actual nature of news articles to recommend accurately. A personalized news recommendation system provides a news set that is extracted from multiple news press releases in order to handle such a high number of news articles and releases these news items to the user. A news item has a specific nature and it is different from the other items to recommend. In recent years, there has been much focus on the design and development of personalized news recommendation systems that observe and learn user behavior and generate news set based on this behavior. Commonly, the current news recommendation systems employ the collaborative filtering-based (CF-based), Content-based filtering (Content-based) or hybrid methods. Scalability is one of the issues in news recommendation that requires effective algorithms to deal with large news corpus. One of the common strategies used for solving scalability is clustering. However, the existing clustering algorithms do not take into consideration the news nature in clustering the news items. The early recommender systems use popularity or recency, or both as properties to demonstrate the interestingness of the news items which do not reflect the actual nature of the news items. On the other hand, the existing personalized news recommendation systems do not make an attempt to filter the number of news items to recommend based on the reading rate behavior of a user. Moreover, early researches only consider explicit profile, short-term profile, and long-term profile of a user but none of them use all of the above user profiles in a single solution. News selection is another issue that requires new solution to effectively select news items to recommend. In this research work, we have proposed a personalized news recommendation framework named Hybrid Personalized NEws Recommendation (HYPNER) which aims to recommend a personalized news set to the users. HYPNER combines both the collaborative filtering-based and the Content-based filtering methods in its framework. To address the above issues, the following have been proposed and incorporated into HYPNER. We have proposed a clustering algorithm named Ordered Clustering with a specific characteristic that allows multiple membership in clusters which reflects the news nature and user behavior. Furthermore, new models for user profile and news metadata construction are proposed where new properties have been incorporated, namely: Reading Rate, Hotness Rate, and Hotness. A new model in news selection is proposed based on sub-modularity model. Our proposed framework has been validated through extensive experiments on real dataset. Results exhibit that HYPNER achieved 81.56% improvement in F1-score and 5.33% in diversity compared to the existing work, SCENE