Traditionally, recommender systems have focused on similarities between items or users. While these systems provide good recommendations, they are by definition quite similar and tightly related. While mechanism to mitigate this exist (e.g., factoring in diversity), we researched the serendipitous recommendations in newspaper articles.
Serendipity, i.e., the discovery of unexpected, yet interesting items, surprise users and lead to non-obvious, yet well-perceived recommendations. Thus, we developed a serendipity-based news article recommendation based on the New York Times Corpus and compared it with traditional similarity-based recommendation methods in a user study. The paper is currently under review. On this website, you can find the articles used in the user study.
