Large-scale data clustering with dynamic social context
View Patent ↗A system and method for dynamic, semi-supervised clustering comprises receiving data attributes, generating a set of ensemble partitions using the data attributes, forming a convex hull using the set of ensemble partitions, generating a simplex vector by performing ensemble clustering on the convex hull, receiving dynamic links, deriving an optimal simplex vector using the simplex vector and the dynamic links, computing a current optimal clustering result using the optimal simplex vector, and outputting the current optimal clustering result.
1. A computer-implemented method for dynamic, semi-supervised clustering, comprising:
receiving data attributes;
generating a set of ensemble partitions using the data attributes;
forming a convex hull using the set of ensemble partitions;
generating a simplex vector by performing ensemble clustering on the convex hull, the simplex vector being a vector whose elements are non-negative and the summation of all of the elements equals one;
receiving one or more pairwise constraints;
deriving an optimal simplex vector using the simplex vector and the one or more pairwise constraints;
computing a current optimal clustering result using the optimal simplex vector; and
outputting the current optimal clustering result,
wherein a hardware processor device is configured to perform said attribute receiving, generating partitions, forming the convex hull, generating the simplex vector, dynamic link receiving, deriving, computing and outputting.
2. The method according to claim 1 , further comprising repeating said receiving links, deriving an optimal simplex vector, and computing a current best clustering result when additional one or more pairwise constraints are received.
3. The method according to claim 1 , wherein the attributes are received from one of: handwritten images as vectors, and a profile obtained from a social networking platform.
4. The method according to claim 1 , wherein the ensemble partitions are generated using a fast spectral clustering with subspace sampling.
5. The method according to claim 1 , wherein the ensemble partitions are further refined using the fast spectral clustering with a Hungarian algorithm.
6. The method according to claim 1 , wherein the one or more pairwise constraints comprise one or more must-link links and one or more cannot-link links.
7. A system for dynamic, semi-supervised clustering, comprising:
a memory device;
a display device; and
a hardware processor coupled to the memory device, the processor configured to:
receive data attributes;
generate a set of ensemble partitions using the data attributes;
form a convex hull using the set of ensemble partitions;
generate a simplex vector by performing ensemble clustering on the convex hull, the simplex vector being a vector whose elements are non-negative and the summation of all of the elements equals one;
receive one or more pairwise constraints;
derive an optimal simplex vector using the simplex vector and the one or more pairwise constraints;
compute a current optimal clustering result using the optimal simplex vector; and
output the current optimal clustering result to the display device.
8. The system according to claim 7 , further comprising repeating receiving links, deriving an optimal simplex vector, and computing a current best clustering result when additional one or more pairwise constraints are received.
9. The system according to claim 7 , wherein the attributes are received from one of: handwritten images as vectors, and a profile obtained from a social networking platform.
10. The system according to claim 7 , wherein the ensemble partitions are generated using a fast spectral clustering with subspace sampling.
11. The system according to claim 10 , wherein the ensemble partitions are further refined using the fast spectral clustering with a Hungarian algorithm.
12. The system according to claim 7 , wherein the one or more pairwise constraints comprise one or more must-link links and one or more cannot-link links.
13. A computer program product for using a multiuser system, the computer program product comprising a non-transitory storage medium readable by a processing circuit and storing instructions run by the processing circuit for performing method steps for dynamic, semi-supervised clustering, comprises:
receiving data attributes;
generating a set of ensemble partitions using the data attributes;
forming a convex hull using the set of ensemble partitions;
generating a simplex vector by performing ensemble clustering on the convex hull, the simplex vector being a vector whose elements are non-negative and the summation of all of the elements equals one;
receiving one or more pairwise constraints;
deriving an optimal simplex vector using the simplex vector and the one or more pairwise constraints;
computing a current optimal clustering result using the optimal simplex vector; and
outputting the current optimal clustering result.
14. The computer program product according to claim 13 , further comprising receiving links, deriving an optimal simplex vector, and computing a current best clustering result when additional one or more pairwise constraints are received.
15. The computer program product according to claim 13 , wherein the attributes are received from one of: handwritten images as vectors, and a profile obtained from a social networking platform.
16. The computer program product according to claim 13 , wherein the ensemble partitions are generated using a fast spectral clustering with subspace sampling.
17. The computer program product according to claim 16 , wherein the ensemble partitions are further refined using the fast spectral clustering with a Hungarian algorithm.
18. The computer program product according to claim 13 , wherein the one or more pairwise constraints comprise one or more must-link links and one or more cannot-link links.