SEREDIPITIOUS RECOMMENDATION FOR MOBILE APPLICATIONS USING ITEM - SIMILARITY GRAPH

Abstract

Abstract The domain of mobile applications(apps) has recently surfaced and generated lot of interest in academia and industry alike. With an App-store for every leading operating system - Apple, Android, Blackberry,Windows, an explosive growth of Mobile applications is not a surprise. The absolute number of apps currently in existence, as well as their rates of growth, are remarkable. This might be good news for the developers from the revenue perspective but for consumers it means the inherent task of ”App Discovery” being intensified. A reasonable solution to this problem are Recommender systems. They usually deal with indicators of user preferences(purchase history/ rating history) for suggesting/predicting items for a target user. An e↵ective way to cut-the-queue and straightaway hit the user’s interest in shortest possible time, RS are extremely popular with commercial systems today. To generate relevant recommendations for users, our system tries to leverage the interest patterns in the downloaded applications on mobile phones of users themselves by using item-item similarity graphs. This work essentially tries to overcome the inherent problem of over-specialization in content based recommender system by using graph approach. This thesis first presents the background literature for recommender systems and then proposes a graph based approach for recommending serendipitous recommendations to a user.