Accuracy Is Not Enough: Serendipity Should Be Considered More

In recent years, information overload has led to the development of a wide variety of different types of recommender systems (RSs), which aim at harnessing the problem in different domains. Most RSs spare no effort to improve the accuracy to help users find items matching their latest preferences. However, RSs with high accuracy not always capture user satisfaction due to that users might get bored with the items which are similar to what they liked in the past. Furthermore, they narrow the horizon of users and limiting the dynamic nature of user interests. Given this, serendipity draws more attention when evaluating RSs. In this paper, we review the studies on serendipity. Working toward this direction, we propose a new metric aiming at evaluating serendipity of RSs and a new strategy to balance the tension between accuracy and serendipity in movie recommender domain.