Private recommendation in a client-server environment
Methods and systems for recommending content to a client device operated by a user include receiving a set of ratings for each of a first set of content items by a user from a client device for use in a factor model. The set of ratings is not maintained in the server for longer than necessary to calculate a rating vector and/or to update a matrix factor defined by the rank of the factor model and a total number of content items eligible for ranking.
1. A method for recommending content to a client device operated by a user, the method comprising:
receiving, by at least one processor, a set of ratings for each of a first set of content items by a user from a client device;
estimating, by the at least one processor, a user weight vector in a factor model for recommendations;
generating, by the at least one processor based on the user weight vector and a factor model, a rating vector comprising a predicted rating for each of second content items;
sending the rating vector to the client device; and
deleting the set of ratings received from the user.
2. The method of claim 1 , wherein estimating the user weight vector comprising solving for a vector argmin of a difference function.
3. The method of claim 2 , wherein the difference function includes a summation of squares difference in a factor model.
4. The method of claim 3 , wherein the factor model is a non-negative factor model and the solving is by non-negative least squares.
5. The method of claim 1 , wherein generating the rating vector comprises a product of the vector argmin and a matrix B=r×I, wherein r is a rank of a factor model and I is a total number of content items.
6. The method of claim 1 , further comprising updating B using a stochastic approximation based on the set of ratings received from the client device.
7. The method of claim 6 , wherein the deleting the set of ratings is triggered by the updating B.
8. The method of claim 1 , wherein the deleting the set of ratings is triggered by the sending the rating vector to the client device.
9. The method of claim 1 , further comprising truncating the rating vector prior to sending to the client device.
10. An apparatus for recommending content to a client device operated by a user, the apparatus comprising at least one processor coupled to a memory, the memory holding program instructions that when executed by the at least one processor, cause the apparatus to perform:
receiving a set of ratings for each of a first set of content items by a user from a client device;
estimating a user weight vector in a factor model for recommendations by alternating minimization;
generating, based on the user weight vector and the factor model, a rating vector comprising a predicted rating for each of second content items;
sending the rating vector to the client device; and
deleting the set of ratings received from the user.
11. The apparatus of claim 10 , wherein the memory holds further instructions for estimating the user weight vector at least in part by solving for a vector argmin of a difference function.
12. The apparatus of claim 11 , wherein the memory holds further instructions for calculating the difference function comprising a summation of squares difference in a factor model.
13. The apparatus of claim 12 , wherein the memory holds further instructions for the factor model being a non-negative factor model and for the solving by non-negative least squares.
14. The apparatus of claim 10 , wherein the memory holds further instructions for generating the rating vector at least in part by a product of the vector argmin and a matrix B=r×I, wherein r is a rank of a factor model and I is a total number of content items.
15. The apparatus of claim 10 , wherein the memory holds further instructions for updating B using a stochastic approximation based on the set of ratings received from the client device.
16. The apparatus of claim 15 , wherein the memory holds further instructions for the deleting the set of ratings triggered by the updating B.
17. The apparatus of claim 15 , wherein the memory holds further instructions for the deleting the set of ratings triggered by the sending the rating vector to the client device.
18. The apparatus of claim 10 , wherein the memory holds further instructions for truncating the rating vector prior to sending to the client device.