Secure multi-party computation with no trusted initializer

A non-transitory, computer-readable storage medium storing one or more instructions executable by a computer system to perform operations for performing a secure collaborative computation, the operations comprising:

• obtaining, by a first party, a first matrix comprising first private data recorded by the first party, wherein the first party comprises a recommendation platform that provides a plurality of services to a plurality of users and the first private data comprises first social relationships between any two of the plurality of users of the recommendation platform, wherein the recommendation platform comprises first user data and item data, and wherein a first portion of the first private data comprises the first user data and a second portion of the first private data comprises the item data, the first user data comprising first user names, first user ages, first user genders, first user addresses, purchase histories, first browsing histories, or search histories and the item data comprising item names, categories, prices, keywords, or instructions;
• generating, by the first party, a first random matrix having a size identical to the first matrix, wherein the first random matrix ensures security of the first private data included in the first matrix;
• identifying, by the first party, a first sub-matrix of the first random matrix and a second sub-matrix of the first random matrix, wherein the first sub-matrix of the first random matrix comprises all even columns of the first random matrix, and the second sub-matrix of the first random matrix comprises all odd columns of the first random matrix;
• computing, by the first party, first scrambled private data of the first party based on the first matrix, the first random matrix, the first sub-matrix of the first random matrix, and the second sub-matrix of the first random matrix, wherein the first private data of the first party are not derivable solely from the first scrambled private data of the first party thereby preventing exposure of the first private data to a second party, wherein the first scrambled private data of the first party comprises a first sum of the first matrix and the first random matrix and a second sum of the first sub-matrix of the first random matrix and the second sub-matrix of the first random matrix;
• transmitting, by the first party directly to the second party without a trusted initializer, the first scrambled private data of the first party, wherein the second party comprises a social network platform that accumulates second private data from the plurality of users, wherein the social network platform comprises second user data comprising second user names, second user ages, second user genders, second user addresses, second browsing histories, employment information, or recreational information, at least a portion of the second user data corresponding to at least a portion of the first user data;
• receiving, by the first party from the second party, second scrambled private data of the second party, wherein second private data of the second party are not derivable solely from the second scrambled private data of the second party, thereby preventing exposure of the second private data to the first party, wherein the second private data comprises second social relationships between any two of the plurality of users;
• computing, by the first party, a first addend of a matrix product of the first matrix comprising the first private data of the first party and a second matrix comprising second private data of the second party;
• receiving, by the first party from the second party, a second addend of the matrix product of the first matrix comprising the first private data of the first party and the second matrix comprising the second private data of the second party;
• performing, by the first party, a local computation of the matrix product of the first matrix comprising the first private data of the first party and the second matrix comprising the second private data of the second party by solving an optimization problem using gradient descents of the first addend and the second addend, wherein the local computation generates updated user data and updated feature data, and wherein the local computation reduces network and processing resources and conserves the security of the first private data included in the first matrix;
• and
• generating, by the first party, a recommendation for a particular service of the plurality of services for one of the plurality of users of the recommendation platform, the recommendation being based on the updated user data and the updated feature data.

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 one or more instructions that, when executed by the one or more computers, perform operations for performing a secure collaborative computation, the operations comprising:
• obtaining, by a first party, a first matrix comprising first private data recorded by the first party, wherein the first party comprises a recommendation platform that provides a plurality of services to a plurality of users and the first private data comprises first social relationships between any two of the plurality of users of the recommendation platform, wherein the recommendation platform comprises first user data and item data, and wherein a first portion of the first private data comprises the first user data and a second portion of the first private data comprises the item data, the first user data comprising first user names, first user ages, first user genders, first user addresses, purchase histories, first browsing histories, or search histories and the item data comprising item names, categories, prices, keywords, or instructions;
• generating, by the first party, a first random matrix having a size identical to the first matrix, wherein the first random matrix ensures security of the first private data included in the first matrix;
• identifying, by the first party, a first sub-matrix of the first random matrix and a second sub-matrix of the first random matrix, wherein the first sub-matrix of the first random matrix comprises all even columns of the first random matrix, and the second sub-matrix of the first random matrix comprises all odd columns of the first random matrix;
• computing, by the first party, first scrambled private data of the first party based on the first matrix, the first random matrix, the first sub-matrix of the first random matrix, and the second sub-matrix of the first random matrix, wherein the first private data of the first party are not derivable solely from the first scrambled private data of the first party thereby preventing exposure of the first private data to a second party, wherein the first scrambled private data of the first party comprises a first sum of the first matrix and the first random matrix and a second sum of the first sub-matrix of the first random matrix and the second sub-matrix of the first random matrix;
• transmitting, by the first party directly to the second party without a trusted initializer, the first scrambled private data of the first party, wherein the second party comprises a social network platform that accumulates second private data from the plurality of users, wherein the social network platform comprises second user data comprising second user names, second user ages, second user genders, second user addresses, second browsing histories, employment information, or recreational information, at least a portion of the second user data corresponding to at least a portion of the first user data;
• receiving, by the first party from the second party, second scrambled private data of the second party, wherein second private data of the second party are not derivable solely from the second scrambled private data of the second party, thereby preventing exposure of the second private data to the first party, wherein the second private data comprises second social relationships between any two of the plurality of users;
• computing, by the first party, a first addend of a matrix product of the first matrix comprising the first private data of the first party and a second matrix comprising second private data of the second party;
• receiving, by the first party from the second party, a second addend of the matrix product of the first matrix comprising the first private data of the first party and the second matrix comprising the second private data of the second party;
• performing, by the first party, a local computation of the matrix product of the first matrix comprising the first private data of the first party and the second matrix comprising the second private data of the second party by solving an optimization problem using gradient descents of the first addend and the second addend, wherein the local computation generates updated user data and updated feature data, and wherein the local computation reduces network and processing resources and conserves the security of the first private data included in the first matrix;
• and
• generating, by the first party, a recommendation for a particular service of the plurality of services for one of the plurality of users of the recommendation platform, the recommendation being based on the updated user data and the updated feature data.