Knowledge Graphs for News Recommendation in a Local News Organization

In this paper, we investigate the efficacy of using knowledge graphs for news recommendation. Small institutions usually do not have the required amount of data needed to build robust recommender systems. Knowledge graphs allow us to recommend news, based on their content, with small amount of data. The graph can be created using related data and calibrated to use domain-specific data. Despite gaining popularity recently, the problem of how to construct knowledge graphs has not been adequately addressed. In this work, we consider the effects of using sentences extracted from the titles and the bodies of news in different languages. Our test case is focused on news data provided by a local news organization in Japan. We develop an effective knowledge graph construction method for the available data that is used by a recommender systems. We evaluate the effectiveness of this system by predicting the clicks of users.