Employing F-MADM to derive user preference model from item features and rating information for personalized recommendation
Jing Zhang, Qinke Peng, Shiquan Sun, Tao Zhong · 2015
Considering the facts that people access to item information more easily than to user information given user privacy, and the features of items selected by the user always imply his/her preferences, we hope to utilize item features to mine user preferences besides ratings. What is more, ratings are often linguistic labels and fuzzy set is tailor-made to represent them. Therefore, we propose a novel recommendation method that firstly uses fuzzy sets to represent ratings; secondly applies fuzzy multiple attributes decision making (F-MADM) to build optimization models based on item features and ratings for determining user preference models; finally combines user preference models with collaborative filtering to make recommendations. This method not only makes good use of item features and uncertain rating information to mine user preferences, but also derives explicit model of user preferences from the optimization model constructed based on F-MADM. We compare our method with other widely used methods on MovieLens. Our method achieves the best accuracy while maintaining an acceptable level of diversity.