ADORES: A Diversity-oriented Online Recommender System

Abstract

Browsing a content platform usually does not require a user identification. In this context, personalized approaches can not be used since no information related to the user is available. In that case, it is important to consider the variety of potential interests of users when providing recommendations. In this paper, we propose a scalable recommendation diversity-oriented model which considers solely the current visited document, the available collection and the past clickthrough documents to produce a list of diversified recommendations. A learning phase is integrated to improve recommendation relevance. Our proposals are validated through several experiments.