An Approach to Content Based Recommender Systems Using Decision List Based Classification with k-DNF Rule Set

Abstract

Recommender systems are the software or technical tools that help user to find out items/things according to his/her preferences from a wide range of items/things. For example, selecting a movie from a large database of movies from on-line or selecting a song of his/her own kind from a large number of songs available in the internet and much more. In order to generate recommendations for the users the system has to first learn the user preferences from the user's past behaviours so that it can predict new items/things that are suitable for the respective user. These systems generally learn user's preferences from user's past experiences, using any machine learning algorithm and predict new items/things for the user using the learned preferences. In this paper we introduce a different approach to recommender system which will learn rules for user preferences using classification based on Decision Lists. We have followed two Decision List based classification algorithms like Repeated Incremental Pruning to Produce Error Reduction and Predictive Rule Mining, for learning rules for users past behaviours. We also list out our proposed recommendation algorithm and discuss the advantages as well as disadvantages of our approach to recommender system with the traditional approaches. We have validated our recommender system with the movie lens data set that contains hundred thousand movie ratings from different users, which is the bench mark dataset for recommender system testing.