Bisociative Serendipity Music Recommendation

With the traditional similarity-based approaches to recommender systems, it is unlikely to discover truly novel things. Users no longer find the outputs interesting or surprising since the outputs are locked into clusters of similarities on single domain. To make recommendations more attractive to the users, the system must provide a serendipitous recommendation list which is new, exciting and unexpectable. Serendipitous recommendation increases the chance of discovering music that is truly novel and unexpectedly useful leading to better performance and higher efficiency. In this paper, we propose a bisociative based approach to automatically generate a serendipitous recommendation list for particular users. The unexpected interesting links crossing different context domains are found by applying bisociative knowledge discovery concept and inducting the rules for generating the serendipitous list using probabilistic logic framework. For the music recommendation on subset of Amazon product dataset including users’ music and movie preferences, we are able to achieve a recommendation list with 60% accuracy. The list includes recommendations which are not found in the single-domain based systems.