BordaRank: A ranking aggregation based approach to collaborative filtering

Abstract

Recommender systems are widely used in today's online applications. Traditional rating-oriented methods predict user ratings on items, but they fail to capture user preference among different items. This paper regards recommendation problem as a ranking task and proposes a new ranking-oriented collaborative filtering framework based on ranking aggregation methods. In this framework, recommendation lists are generated according to item rankings given by users who are similar to the target user. Then, a two-step method called BordaRank is proposed to further explain the framework. The method first uses item collaborative filtering to predict unknown ratings and then uses Borda count method to aggregate item rankings of neighbors. Finally, BordaRank is modified as a pure ranking-oriented method, which could be directly applied on the sparse rating matrix without rating prediction as an intermediate step. The methods are evaluated on real world movie rating data. Experimental results show that BordaRank improves the precision and recall of original rating-oriented methods and modified BordaRank also outperforms traditional methods.