A Multi-Armed Bandit Recommender Algorithm Based on Conversation and KNN

With the wide application of recommendation systems in various fields, in order to effectively solve the cold-start problem in recommendation systems, contextual bandit algorithm uses user feedback to update user preferences online, converting the cold-start problem of recommendation systems into an exploration and exploitation problem. However, traditional contextual bandit algorithm is slow to learn due to the extensive exploration required. With the development of conversational recommendation, conversational contextual bandit algorithm learns the user's preference for key-term through conversation thus accelerating the learning speed. However, it only considers user feedback on key-term and ignores the relevance of key-term to each other. To solve the problem, a multi-armed bandit based on conversation and KNN (K-Nearest Neighbors) algorithm is proposed by introducing a more refined collaboration (KNNConUCB). Experiments on Synthetic data, as well as real datasets from Movielens and Last.FM, demonstrate the efficacy of the KNNConUCB algorithm.