摘要 |
To recommend an item which is highly unexpected to a user because its similarity to user preferences is low and which is useful to the user. A rule that modifies a set of keywords for recommending an item is randomly applied, and a keyword which a user does not prefer is added and a keyword which a user prefers is removed, and then this recommendation result is mixed with a recommendation result of a set of keywords before the above modification, and the mixed result is presented to a user and at the same time the application probability of a rule is learned on the basis of the user's evaluation to a recommended item.
|