Exploring the Power of Weak Ties on Serendipity in Recommender Systems

With our increasingly refined online browsing habits, the demand for high-grade recommendation systems has never been greater. Improvements constantly target general performance, evaluation, security, and explainability, but optimizing for serendipitous experiences is imperative since a serendipity-optimized recommender helps users discover unforeseen relevant content. Given that serendipity is a form of genuine unexpected experiences and recommenders are facilitators of user experiences, we aim at leveraging weak ties to explore their impact on serendipity. Weak links refer to social connections between individuals or groups that are not closely related or connected but can still provide valuable information and opportunities. On the other hand, the underlying social structure of recommender datasets can be misleading, rendering traditional network-based approaches ineffective. For that, we developed a network-inspired clustering mechanism to overcome this obstacle. This method elevates the system’s performance by optimizing models for unexpected content. By leveraging group weak ties, we aim to provide a novel perspective on the subject and suggest avenues for future research. Our study can also have practical implications for designing online platforms that enhance user experience by promoting unexpected discoveries.