METHOD OF CREATING CONTENT RECOMMENDATIONS BASED ON USER RATINGS OF CONTENT WITH IMPROVED USER PRIVACY

• US 20160071170A1
• Filed: 02/08/2013
• Published: 03/10/2016
• Est. Priority Date: 02/08/2013
• Status: Abandoned Application

• Current Assignee: Thomson Licensing (Vantiva SA)
• Original Assignee: Thomson Licensing (Vantiva SA)
• Inventors: HEGDE, Nidhi, MASSOULIE, Laurent, BANERJEE, Siddhartha
• Application Number:
  US14/377,666
• Publication Number:
  US 20160071170A1
• Time in Patent Office: Days
• US Class Current: 1/1
• CPC Class Codes: G06Q 30/0282 Rating or review of business operators or products
• CPC Class Codes: G06Q 30/0631 Item recommendations

Abstract

A recommendation system provides improved privacy under the local model, in which users store their data locally, and differential privacy is ensured through randomization of ratings under the control of the user before submitting data to the recommender system. The recommender system performs a clustering on the data collected from a plurality of users and returns the results to the users. The users'"'"' devices fine-tune the recommendation based on locally stored, non-randomized ratings. The recommendation system provides a user-adjustable degree of privacy while still allowing for generating recommendations with a decent accuracy.

Claims

1-27. -27. (canceled)

28. A method of providing data to a user device for enabling the user device to recommend an item from a set of items to a user, the method including the steps of:

a recommending engine receiving, from the user device of the first user and from user devices of the second users, privatized ratings in accordance with an N-dimensional binary sensing vector, the sensing vector corresponding to a representation of selected items from the set of items to be rated, wherein a rating corresponds to a rating value selected from an interval lying between an upper and a lower rating value, wherein items are rated in accordance with the sensing vector, the sensing vectors received from different users not necessarily being identical, wherein each sensing vector has a similitude indicator assigned to it, the similitude indicator indicating if the rating values of at least two items represented by the sensing vector are above a predetermined value, wherein a trust value is assigned to each similitude indicator, the trust value indicating the probability of the similitude indicator being exact and reliable, wherein the N-dimensional binary sensing vector is received by the recommending engine from each user device along with the rating;

the recommending engine collecting rating information from a predetermined minimum number of users and determining, for each item, the sum of the rating values, in accordance with respective sensing vectors and the similitude indicator;

the recommending engine performing a clustering on the collected information, wherein the clustering includes clustering for similar content and/or similar rating; and

the recommending engine providing a clustered representation of selected items from the set of items to the first user'"'"'s device.

29. The method of claim 28, wherein the spectral clustering takes into account the trust value associated with the items.

30. The method of claim 28, wherein the clustering takes into account properties of the items not related to the rating, the properties including one or more from a non-exclusive list including genre, artist, director, distributor, year of release.

31. A recommending engine for a recommender system including a program memory, a processor, a data storage and a network interface, wherein the recommending engine is adapted to perform a method in accordance with claim 28.