A Probabilistic Model for Diversifying Recommendation Lists

摘要: We propose a probabilistic method to diversify the results of collaborative filtering. Recommendation diversity is being studied by many researchers as critical factor that significantly influences user satisfaction. Unlike conventional approaches recommendation diversification, we theoretically derived diversification method. Specifically, our naturally diversifies list maximizing probability selects at most one item from list. For enhanced practicality, formulate model for proposed on three policies — robust estimation, use only purchase history, and elimination any hyperparameters controlling diversity. In this paper, formally demonstrate practically superior methods, experimentally show competitive with methods in terms accuracy