Content boosted hybrid filtering for enhanced personalization in recommendation systems

Abstract

Recommendation systems play a vital role in easing the process of newlinedecision-making in our daily activities. Recommended items may be movies, newlinepieces of music, news articles, scientific articles, products, websites, books, elearning newlinecourses, games, jokes, and so on. Content-based filtering and newlineCollaborative filtering are the two popular methods used to provide newlinerecommendations. Content-based filtering (CBF) method considers the users newlinepreferences alone and recommends items similar to the user s interest. User newlinepreferences are identified from their activities, and user-profiles are constructed newlinefor each user. Collaborative filtering (CF) method takes into account the interests newlineof similar users in the recommendation. Nearest neighbours are drawn up using newlinethe similarity in their style of ratings. Ratings for the unseen items are predicted newlinebased on the similarity score and the neighbour s ratings. newlineContent-based methods are good at providing user independence newlineand transparency but suffer from over-specialization and limited content newlineanalysis. A pure content-based filtering method suggests nothing unexpected newlineor surprising. This limited degree of novelty is known as serendipity problem. newlineCollaborative approaches do not require domain knowledge and can discover newlinenew interests with the help of neighbours. However, they suffer from coldstart, newlinesparsity, and scalability issues. The drawbacks of one and the other can newlinebe rectified by combining these approaches into a hybrid system. While newlinecollaborative filtering techniques provide better performance than most of the newlinecontent-based filtering methods, the personalization in the latter should not be newlineneglected. newline