Context-Aware Bandits. (arXiv:1510.03164v1 [cs.LG])

In this paper, we present the CAB (Context- Aware Bandits). With CAB we attempt to craft a bandit algorithm that can exploit collaborative effects and that can be deployed in a practical recommendation system setting, where the multi-armed bandits have been shown to perform well in particular with respect to the cold start problem. CAB exploits, a context-aware clustering technique augmenting exploration-exploitation strategies in a contextual multi-armed bandit settings. CAB dynamically clusters the users based on the content universe under consideration. We demonstrate the efficacy of our approach on extensive real-world datasets, showing the scalability, and more importantly, the significant increased prediction performance compared to related state-of-the-art methods.

from cs.AI updates on arXiv.org http://ift.tt/1ZwXM9W
via IFTTT