Systems and methods for a cross media joint friend and item recommendation framework

Patent metadata

Field	Value
Publication number	US-2021272217-A1
Application number	US-201916525148-A
Country	US
Kind code	A1
Filing date	Jul 29, 2019
Priority date	Aug 2, 2018
Publication date	Sep 2, 2021
Grant date	—

Abstract

Various embodiments of systems and methods for cross media joint friend and item recommendations are disclosed herein.

First claim

What is claimed is:

1 . A method, comprising:
linking items across a plurality of social network sites using a sparse transfer learning method, wherein the sparse transfer learning method uses a shared dictionary and a set of sparse item representations between a source social network site and a target social network site to obtain a set of source features and a set of target features; obtaining a set of latent user features and a set of latent item features, wherein the set of latent item features is obtained using a projection matrix on the set of sparse item representations and wherein the set of latent user features is obtained by mapping a set of user-user link matrices to a shared space; and solving an objective function, wherein the objective function comprises a plurality of variables and wherein the plurality of variables comprises the shared dictionary, the set of sparse item representations, the latent user features, the projection matrix, and a shared interaction matrix; wherein the objective function optimizes the plurality of variables.

2 . The method of claim 1 , wherein the solved objective function produces a resultant latent matrix representation of users and a resultant latent matrix representation of items and wherein the resultant latent matrix representation of users and the resultant latent matrix representation of items are respectively used to perform friend recommendation tasks and item recommendation tasks.

3 . The method of claim 1 , wherein the set of source features, the shared dictionary, the set of sparse representations, a graph Laplacian matrix, and a maximum mean discrepancy matrix are used to develop a first term of the objective function using the sparse transfer learning method.

4 . The method of claim 3 , further comprising:
reconstructing a first matrix representative of the set of source features using the shared dictionary and a first matrix of the set of sparse item representations, wherein the first matrix of the set of sparse item representations belongings to the source social network site; and reconstructing a second matrix representative of the set of target features using the shared dictionary and a second matrix of the set of sparse item representations, wherein the second matrix of the set of sparse item representations belongings to the target social network site.

5 . The method of claim 1 , wherein the set of latent user features, a set of rating matrices, and the set of latent item features are used to create a second term of the objective function by modeling cross-site rating transfer between the source social network site and the target social network site, wherein cross-site rating transfer projects the set of sparse item representations to the set of latent item features using the projection matrix.

6 . The method of claim 1 , wherein the set of user-user link matrices, the shared interaction matrix, and the set of latent user features are used to create a third term of the objective function by modeling user relations transfer learning between the source social network site and the target social network site, wherein user relations transfer learning decomposes and maps the set of user-user link matrices to the shared space to obtain the set of latent user features, wherein the set of latent user features are modeled between the source social network site and the target social network site.

7 . The method of claim 1 , wherein the set of sparse item representations comprise sparse representations of item features inherent to the source social network site or the target social network site.

8 . The method of claim 1 , wherein the set of user-user link matrices are representative of user-user adjacency on the source social network site or the target social network site.

9 . The method of claim 1 , wherein the plurality of variables is iteratively updated until the objective function converges.

10 . The method of claim 3 , further comprising:
computing the graph Laplacian matrix and the maximum mean discrepancy matrix, wherein the graph Laplacian matrix and the maximum mean discrepancy matrix are used to ensure that the shared dictionary fits an intrinsic geometry of the source social network site or the target social network site.

11 . The method of claim 1 , wherein an alternating least square method is used to iteratively optimize each of the plurality of variables of the objective function if the objective function is not convex.

12 . A method, comprising:
linking items across a plurality of social network sites using a sparse transfer learning method, the method comprising:
reconstructing a first matrix representative of a set of source features using a shared dictionary and a first sparse item representation matrix, wherein the first sparse item representation matrix is representative of a first set of sparse item representations belonging to a source social network site;
reconstructing a second matrix representative of a set of target features using the shared dictionary and a second sparse item representation matrix, wherein the second sparse item representation matrix is representative of a second set of sparse item representations belonging to a target social network site; and
ensuring that the shared dictionary fits an intrinsic geometric structure of the set of source features or the set of target features using a graph regularized sparse coding algorithm;
projecting the first and second sets of sparse item representations to a set of latent item feature representations using a projection matrix; and obtaining a first and second set of latent user features by decomposing a first and second user-user link matrix and incorporating a shared interaction matrix, wherein the first user-user link matrix is representative of user-user adjacency on the source social network site, and wherein the second user-user link matrix is representative of user-user adjacency on the target social network site; wherein an objective function is developed, wherein the objective function comprises the shared dictionary, the pair of sparse item representation matrices, the latent user features, the projection matrix, and the shared interaction matrix.

13 . The method of claim 12 , further comprising:
solving the objective function to produce a resultant latent matrix representation of users and a resultant latent matrix representation of items.

14 . The method of claim 12 , further comprising:
developing a first term of the objective function using the set of source features, the shared dictionary, the set of sparse representations, a graph Laplacian matrix, and a maximum mean discrepancy matrix, wherein the set of source features, the shared dictionary and the set of sparse representations are obtained using the sparse transfer learning method.

15 . The method of claim 12 , further comprising:
developing a second term of the objective function using the set of latent user features, a set of rating matrices, and the set of latent item feature representations.

16 . The method of claim 12 , further comprising:
developing a third term of the objective function using the set of user-user link matrices, the shared interaction matrix, and the set of latent user features.

17 . A method, comprising:
developing an objective function, wherein the objective function comprises a shared dictionary, a set of sparse item representations, a set of latent user features, a projection matrix, and an interaction matrix; executing a dictionary learning method, wherein the dictionary learning method updates

Assignees

• Shu Kai
• Wang Suhang
• Tang Jiliang
• Wang Yilin
• Liu Huan
• Univ Arizona State

Inventors

• Shu Kai
• Wang Suhang
• Tang Jiliang
• Wang Yilin
• Liu Huan

Classifications

• G06Q10/40

  Business processes related to social networking or social networking services · CPC title

• G06Q30/0282 Primary

  Rating or review of business operators or products · CPC title

• G06F18/2136

  based on sparsity criteria, e.g. with an overcomplete basis · CPC title

• G06N5/022

  Knowledge engineering; Knowledge acquisition · CPC title

• G06N5/04

  Inference or reasoning models · CPC title