USER CURIOSITY-BASED BAYESIAN PERSONALIZED RANKING RECOMMENDATION METHOD

摘要

A user curiosity-based Bayesian personalized ranking recommendation method, comprising the following steps: using a data set with a friend relationship, and traversing historical data of each user in the data set; for each user u in the data set, constructing a positive user-item set P_u of the user u, and a curiosity user-item set C_u of the user u and a negative user-item set N_u of the user u; providing an optimized ranking criterion, and acquiring a user matrix P and an item matrix Q; for each user u, randomly selecting items from P_u, C_u, and N_u to form item pairs (positive, negative) and (curiosity, negative) for training, and continuously updating the user matrix P and the item matrix Q during each training iteration; and performing predictive scoring on all the items having no feedback from the user u, and selecting top N items having the highest scores for recommendation.