ENRS: An Effective Recommender System Using Bayesian Model

Abstract

Traditional content-based news recommender systems strive to use a bag of words or a topic distribution to capture readers’ reading preference. However, they didn’t take advantage of the named entities extracted from news articles and the relations among different named entities to model readers’ reading preference. Named entities contain much more semantic information and relations than a bag of words or a topic distribution. In this paper, we design and implement a prototype system named ENRS, which combines the named entity with the naïve Bayesian algorithm, to recommend readers news articles. The key technical merit of our work is that we built a probabilistic entity graph to capture the relations among different named entities, based on which ENRS can increase the diversity of recommendation significantly. The architecture of ENRS and the recommendation algorithm are discussed and a demonstration of ENRS is also presented.