Abstract
To address the information cocoon phenomenon in short video recommender systems, unexpectedness in recommendations is increasingly recognized as a critical factor, extending beyond the traditional focus on personalization. However, limited research has explored how unexpectedness influences the longitudinal performance of these systems. Thus, we use agent-based simulations to systematically examine the optimal timing for introducing unexpectedness to maximize recommender system performance. We find that as users consume more recommendations within a session, the positive impact of personalization diminishes, while the positive effect of unexpectedness increases. Consequently, recommender systems should initially prioritize personalization and gradually introduce unexpected recommendations during longer sessions, preserving the early benefits of personalization while leveraging the advantages of unexpectedness later. Additionally, user attributes such as boredom proneness, curiosity, memory capacity, and preference diversity significantly shape the optimal timing for emphasizing unexpectedness. These insights offer valuable guidance for designing nuanced, user-tailored recommender systems on short video platforms.
Recommended Citation
Yang, Guorui; Wang, Xueqing; Fan, Longwei; Liu, Junming; and Xu, David Jingjun, "When Does Unexpectedness Matter in Short Video Recommender
Systems? An Agent-Based Simulation" (2025). ACIS 2025 Proceedings. 55.
https://aisel.aisnet.org/acis2025/55