Profile based content retrieval for recommender systems
The present invention relates to an apparatus, a method and a computer program product for controlling a recommender system, wherein a user profile normally used by a recommender to predict user ratings is employed to generate a targeted query for the remote database yielding a set of results that can be scored by the recommender and provided as suggestions to the user.
1. An apparatus for controlling a recommender system, said apparatus comprising:
a feature-value extractor for analyzing a user profile and for extracting feature-value pairs which characterize a user of said recommender system;
a feature selector for selecting from said feature-value pairs extracted by said feature-value extractor those feature-value pairs which are most discriminative for said user profile,
a query composer for composing, based on feature-value pairs extracted by said feature-value extractor and selected by said feature selector, a query to be sent to at least one content source; and
a recommender engine for rating results obtained in response to said query so as to provide a recommendation for said user;
wherein said feature-value selector is adapted to use a Relief algorithm as a feature selection algorithm for selecting said most discriminative feature-value pairs.
2. The apparatus according to claim 1 , wherein said feature-value extractor is adapted to extract feature-value pairs based on positive user ratings.
3. The apparatus according to claim 1 , wherein said query composer is adapted to expand said query using a thesaurus or an ontology.
4. The apparatus according to claim 1 , wherein said query composer is adapted to extend said query to categories other than those of said user profile.
5. The apparatus according to claim 4 , wherein cross-domain recommendations are provided.
6. The apparatus according to claim 1 , wherein said user profile has been derived from said recommender system.
7. The apparatus according to claim 1 , wherein said query composer is configured count co-occurrences of feature-value pairs, and to combine only those feature-value pairs with a sufficient co-occurrence count in the query.
8. A method of controlling a recommender system, said method comprising:
analyzing a user profile and extracting feature-value pairs which characterize a user of said recommender system;
selecting from said extracted feature-value pairs those feature-value pairs which are most discriminative for said user profile;
composing based on extracted and selected feature-value pairs a query to be sent to at least one content source; and
rating results obtained in response to said query so as to provide a recommendation for said user; wherein a Relief algorithm is used as a feature selection algorithm for selecting said most discriminative feature-value pairs.
9. The method according to claim 8 , further comprising counting co-occurrences of feature-value pairs that appear in a same item and combining only those feature-value pairs with a sufficient co-occurrence count in said query.
10. The method according to claim 8 , wherein said composing comprises extending said query to categories other than those of said user profile.
11. The method according to claim 10 , further comprising providing cross-domain recommendations.
12. A non-transitory computer readable medium stored with code, which when executed by a processor of a computer device, causes the computer device to perform the method of claim 8 .