Prediction of user appreciation of items and corresponding recommendation method

摘要

Method for predicting a first user's appreciation of a first item, comprising the following steps:
a) learning (10) at least a first appreciation of the first user for one second item;
b) learning (10) other appreciations of other users of the plurality of items;
c) clustering (14) the users into a finite number of user clusters;
d) clustering (14) the plurality of items into a finite number of item clusters;
e) establishing (16) a weight function that expresses statistics of a user giving a appreciation r for an item i conditioned on the user being in a user cluster t and the item being in an item cluster f;
f) calculating (18,20) for each user his user profile and for each item its item profile by exchanging messages between all user-item pairs for which the appreciations are known; and
g) predicting (22) the first user's appreciation of the first item.