An Algorithmic Framework for Adaptive Collaborative Filtering

摘要

In this paper, adaptive collaborative filtering algorithm to capture the errors in ratings and to utilize the knowledge in recommendation is proposed. The contributions of this thesis are summarized as follows. 1) A method of measuring the error level of each rating is proposed. By the proposed method, the adaptive collaborative filtering algorithm can recommend items based on more informative ratings while filtering out the noises from the erroneous ratings. 2) Presented the concept of adaptive weight and the method of computing it Adaptive weight is the relative informational value of a rating. Ratings of lower adaptive weight contain more error than the ratings of higher adaptive weight 3) The performance of the recommendation is improved. While traditional collaborative filtering algorithms treat all the ratings equally, the adaptive collaborative filtering algorithm discriminates the ratings based on the level of contained error. Consequently, the recommendations are more close to users' current preferences. 4) The adaptive collaborative filtering system is implemented. The system supports all the learning curves suggested in this study.