Experience in item based recommender system

Abstract

Recommender system (RS) was a topic of discussion since from first paper on collaborative filtering in mid-1990s describing how social information can be extracted to make good decisions on a behalf of user. According to [1] there are two main strategies in recommendation Content based (CB) and Collaborative filtering (CF) we discuss only CF in this paper. CF try to suggest item rated by other user who were similar to the targeted user, but it have drawbacks like Data Sparsity and Cold Start despite approach specific problems some were common in both type of RS like a flexibility[4], Multidimensionality [3] of recommendation and lack of non-intrusive feedback mechanism, means RS often needs significant level of user involvement for example, before recommending a website link to user RS needs Ratings and Reviews from previous user experience to came up with good recommendation, more Rated and Reviewed item will always get higher ranking in its recommendation pool. This way system get experienced from user input but most of the time users were not interested in rating the items or there would be a scenario of biasing for certain types of item that would make RS prone to attack and easy to manipulated by ad-hoc users in this paper we talk about how to develop a structural user rank metric based on their interest without using explicit feedback like users rating and reviews.