Secret sharing with a trusted initializer

1. A computer-implemented method, comprising:

identifying, by an item rating and recommendation platform, rating data comprising respective ratings of a plurality of items with respect to a plurality of users;

identifying, by the item rating and recommendation platform, user-feature data comprising a plurality of user features contributing to the respective ratings of the plurality of items with respect to the plurality of users, wherein the user-feature data is representable in a first matrix;

receiving, from a social network platform by the item rating and recommendation platform via a secret sharing scheme with a trusted initializer, manipulated social network data computed based on social network data and first input data from the trusted initializer, wherein the social network data is representable in a second matrix and indicates social relationships between any two of the plurality of users, wherein, in the secret sharing scheme with the trusted initializer, the social network platform shares with the item rating and recommendation platform the manipulated social network data without disclosing the social network data; and

updating, by the item rating and recommendation platform, the user-feature data based on the rating data and the manipulated social network data, wherein updating the user-feature data comprises computing a matrix product of the first matrix and the second matrix by computing a sum of manipulated user-feature data and the manipulated social network data.

2. The computer-implemented method of claim 1 , wherein, in the secret sharing scheme with the trusted initializer, the item rating and recommendation platform shares with the social network platform the manipulated user-feature data without disclosing the user-feature data, and wherein the manipulated user-feature data are computed based on the user-feature data and second input data from the trusted initializer.

3. The computer-implemented method of claim 2 , further comprising, in the secret sharing scheme with the trusted initializer:

obtaining, by the item rating and recommendation platform, the second input data from the trusted initializer;

generating, by the item rating and recommendation platform, first manipulated user-feature data based on the user-feature data and the second input data from the trusted initializer;

transmitting, by the item rating and recommendation platform to the social network platform, first manipulated user-feature data;

receiving, by the item rating and recommendation platform from the social network platform, first manipulated social network data computed based on the social network data and first input data from the trusted initializer;

generating, by the item rating and recommendation platform, a plurality of intermediate data based on two or more of the first manipulated user-feature data, the first manipulated social network data, or the second input data from the trusted initializer;

transmitting, by the item rating and recommendation platform to the social network platform, plurality of intermediate data;

receiving, by the item rating and recommendation platform from the social network platform, a plurality of intermediate social network data computed based on two or more of the first manipulated user-feature data, the first manipulated social network data, or the first input data from the trusted initializer; and

generating, by the item rating and recommendation platform, manipulated user-feature data based on two or more of the plurality of intermediate data, the plurality of intermediate social network data, or the second input data from the trusted initializer.

4. The computer-implemented method of claim 1 , wherein the manipulated user-feature data and the manipulated social network data are represented in a matrix form, respectively.

5. The computer-implemented method of claim 1 , further comprising:

identifying item-feature data comprising a plurality of item features contributing to the respective ratings of the plurality of items with respect to the plurality of users, wherein a product of the user-feature data and the item-feature data in an estimate of the rating data; and

updating the item-feature data based on the rating data and the user-feature data.

6. The computer-implemented method of claim 5 , further comprising:

generating predicted rating data of the plurality of items with respect to the plurality of users based on the product of the user-feature data and the item-feature data; and

generating a recommendation of a particular item for a particular user based on the predicted rating data.

7. The computer-implemented method of claim 6 , wherein updating the user-feature data comprises solving an optimization problem to minimize a weighted sum of a difference between the predicted rating data and rating data, the user-feature data weighted by the social network data, and an overfitting-preventing term.

8. The computer-implemented method of claim 7 , wherein solving an optimization problem comprises solving:

arg

⁢

⁢

min

U

i

,

V

j

⁢

⁢

L

=

∑

u

i

∈

u

,

v

j

∈

v

⁢

(

R

ij

-

U

i

T

⁢

V

j

)

2

+

λ

1

⁢

∑

u

i

,

u

k

∈

u

⁢

S

ik

⁡

(

U

i

-

U

k

)

2

+

λ

2

(



∑

u

i

∈

u

⁢

U

i



2

+



∑

v

j

∈

v

⁢

V

i



2

)

,

wherein:

U i represents a vector of the user-feature data corresponding to user i;

V j represents a vector of the item-feature data corresponding to item j;

R ij represents a rating of item j by user i;

S ik represents a social relationship between user i and user k;

λ 1 represents a predefined weight associated with the social network data; and

λ 2 represents a predefined weight to prevent overfitting.

9. A non-transitory, computer-readable medium storing one or more instructions executable by a computer system to perform operations comprising:

identifying, by an item rating and recommendation platform, rating data comprising respective ratings of a plurality of items with respect to a plurality of users;

identifying, by the item rating and recommendation platform, user-feature data comprising a plurality of user features contributing to the respective ratings of the plurality of items with respect to the plurality of users, wherein the user-feature data is representable in a first matrix;

receiving, from a social network platform by the item rating and recommendation platform via a secret sharing scheme with a trusted initializer, manipulated social network data computed based on social network data and first input data from the trusted initializer, wherein the social network data is representable in a second matrix and indicates social relationships between any two of the plurality of users, wherein, in the secret sharing scheme with the trusted initializer, the social network platform shares with the item rating and recommendation platform the manipulated social network data without disclosing the social network data; and

updating, by the item rating and recommendation platform, the user-feature data based on the rating data and the manipulated social network data, wherein updating the user-feature data comprises computing a matrix product of the first matrix and the second matrix by computing a sum of manipulated user-feature data and the manipulated social network data.

10. The non-transitory, computer-readable medium of claim 9 , wherein, in the secret sharing scheme with the trusted initializer, the item rating and recommendation platform shares with the social network platform the manipulated user-feature data without disclosing the user-feature data, and wherein the manipulated user-feature data are computed based on the user-feature data and second input data from the trusted initializer.

11. The non-transitory, computer-readable medium of claim 10 , the operations further comprising, in the secret sharing scheme with the trusted initializer:

obtaining, by the item rating and recommendation platform, the second input data from the trusted initializer;

generating, by the item rating and recommendation platform, first manipulated user-feature data based on the user-feature data and the second input data from the trusted initializer;

transmitting, by the item rating and recommendation platform to the social network platform, first manipulated user-feature data;

receiving, by the item rating and recommendation platform from the social network platform, first manipulated social network data computed based on the social network data and first input data from the trusted initializer;

generating, by the item rating and recommendation platform, a plurality of intermediate data based on two or more of the first manipulated user-feature data, the first manipulated social network data, or the second input data from the trusted initializer;

transmitting, by the item rating and recommendation platform to the social network platform, plurality of intermediate data;

receiving, by the item rating and recommendation platform from the social network platform, a plurality of intermediate social network data computed based on two or more of the first manipulated user-feature data, the first manipulated social network data, or the first input data from the trusted initializer; and

generating, by the item rating and recommendation platform, manipulated user-feature data based on two or more of the plurality of intermediate data, the plurality of intermediate social network data, or the second input data from the trusted initializer.

12. The non-transitory, computer-readable medium of claim 9 , wherein the manipulated user-feature data and the manipulated social network data are represented in a matrix form, respectively.

13. The non-transitory, computer-readable medium of claim 9 , the operations further comprising:

identifying item-feature data comprising a plurality of item features contributing to the respective ratings of the plurality of items with respect to the plurality of users, wherein a product of the user-feature data and the item-feature data results in an estimate of the rating data; and

updating the item-feature data based on the rating data and the user-feature data.

14. The non-transitory, computer-readable medium of claim 13 , further comprising:

generating predicted rating data of the plurality of items with respect to the plurality of users based on the product of the user-feature data and the item-feature data; and

generating a recommendation of a particular item for a particular customer based on the predicted rating data.

15. A computer-implemented system, comprising:

one or more computers; and

one or more computer memory devices interoperably coupled with the one or more computers and having tangible, non-transitory, machine-readable media storing instructions, that when executed by the one or more computers, perform operations comprising:

identifying, by an item rating and recommendation platform, rating data comprising respective ratings of a plurality of items with respect to a plurality of users;

identifying, by the item rating and recommendation platform, user-feature data comprising a plurality of user features contributing to the respective ratings of the plurality of items with respect to the plurality of users, wherein the user-feature data is representable in a first matrix;

receiving, from a social network platform by the item rating and recommendation platform via a secret sharing scheme with a trusted initializer, manipulated social network data computed based on social network data and first input data from the trusted initializer, wherein the social network data is representable in a second matrix and indicates social relationships between any two of the plurality of users, wherein, in the secret sharing scheme with the trusted initializer, the social network platform shares with the item rating and recommendation platform the manipulated social network data without disclosing the social network data; and

updating, by the item rating and recommendation platform, the user-feature data based on the rating data and the manipulated social network data, wherein updating the user-feature data comprises computing a matrix product of the first matrix and the second matrix by computing a sum of manipulated user-feature data and the manipulated social network data.

16. The computer-implemented system of claim 15 , wherein, in the secret sharing scheme with the trusted initializer, the item rating and recommendation platform shares with the social network platform the manipulated user-feature data without disclosing the user-feature data, and wherein the manipulated user-feature data are computed based on the user-feature data and second input data from the trusted initializer, wherein the manipulated user-feature data and the manipulated social network data.

17. The computer-implemented system of claim 15 , the operations further comprising:

identifying item-feature data comprising a plurality of item features contributing to the respective ratings of the plurality of items with respect to the plurality of users, wherein a product of the user-feature data and the item-feature data results in an estimate of the rating data; and

updating the item-feature data based on the rating data and the user-feature data.

18. The computer-implemented system of claim 17 , the operations further comprising:

generating predicted rating data of the plurality of items with respect to the plurality of users based on the product of the user-feature data and the item-feature data; and

generating a recommendation of a particular item for a particular customer based on the predicted rating data.

19. The computer-implemented system of claim 18 , wherein updating the user-feature data comprises solving an optimization problem to minimize a weighted sum of a difference between the predicted rating data and rating data, the user-feature data weighted by the social network data, and an overfitting-preventing term.

20. The computer-implemented system of claim 19 , wherein solving an optimization problem comprises solving:

arg

⁢

⁢

min

U

i

,

V

j

⁢

⁢

L

=

∑

u

i

∈

u

,

v

j

∈

v

⁢

(

R

ij

-

U

i

T

⁢

V

j

)

2

+

λ

1

⁢

∑

u

i

,

u

k

∈

u

⁢

S

ik

⁡

(

U

i

-

U

k

)

2

+

λ

2

(



∑

u

i

∈

u

⁢

U

i



2

+



∑

v

j

∈

v

⁢

V

i



2

)

,

wherein:

U i represents a vector of the user-feature data corresponding to user i;

V j represents a vector of the item-feature data corresponding to item j;

R ij represents a rating of item j by user i;

S ik represents a social relationship between user i and user k;

λ 1 represents a predefined weight associated with the social network data; and

λ 2 represents a predefined weight to prevent overfitting.