The Use of Bandit Algorithms in Intelligent Interactive Recommender Systems

This can be naturally modeled constantly explore innovative ways to provide optimal online as contextual bandit problems (e.g., LinUCB [18] and Thompson user experiences for gaining competitive advantages. The great sampling [7]), where each arm corresponds to an item, pulling an needs of developing intelligent interactive recommendation systems item indicates recommending an item, and the reward is the instant are indicated, which could sequentially suggest users the most feedback from a user after the recommendation. Contextual proper items by accurately predicting their preferences, while receiving bandit algorithms have been widely applied in various interactive the up-to-date feedback to refine the recommendation results, recommender systems by achieving an optimal tradeoff between continuosly. Multi-armed bandit algorithms, which have been exploration and exploitation. Based on the preliminary studies [15, widely applied into various online systems, are quite capable of 18, 1], several practical challenges are identified in modern recommender delivering such efficient recommendation services.