Content-based and time-evolving social network analysis
View Patent ↗System and method for modeling a content-based network. The method includes finding single mode clusters from among network (sender and recipient) and content dimensions represented as a tensor data structure. The method allows for derivation of useful cross-mode clusters (interpretable patterns) that reveal key relationships among user communities and keyword concepts for presentation to users in a meaningful and intuitive way. Additionally, the derivation of useful cross-mode clusters is facilitated by constructing a reduced low-dimensional representation of the content-based network. Moreover, the invention may be enhanced for modeling and analyzing the time evolution of social communication networks and the content related to such networks. To this end, a set of non-overlapping or possibly overlapping time-based windows is constructed and the analysis performed at each successive time interval.
1. A computer-implemented system for modeling and analyzing the time evolution of social communication networks and a content related to such networks, said system comprising:
an input device configured for receiving data representing a social communication network;
a processing unit coupled to said input device for:
(a) constructing a multi-mode tensor having one or a plurality of content dimensions and one or a plurality of network dimensions, said constructing being performed at each of a plurality of successive time intervals;
(b) decomposing the multi-mode tensor into a core tensor and a set of factor matrices, one factor matrix for each mode;
(c) modeling each mode of the tensor, wherein the modeling comprises treating each corresponding factor matrix as a low-dimensional representation of that mode, and grouping rows of each factor matrix into clusters based on a distance measure;
(d) modeling a cross-mode correlation, wherein the cross-mode correlation comprises clustering each modeled mode of the multi-mode tensor, enumerating all combinations of clusters across different modes and identifying the cross-mode clusters that induce dense sub-tensors; and
(e) generating a visualization of the modeled cross-mode correlation; and
an output device configured for visualizing the modeled cross-mode correlation.
2. The computer-implemented system as claimed in claim 1 , further comprising receiving a content dimension representing a body or a topic of a message communicated from a respective sender to a recipient over said network.
3. The computer-implemented system as claimed in claim 1 , further comprising performing a tensor decomposition to reduce dimensionality of the social communication networks.
4. A method implemented in a computer system for modeling and analyzing social communication networks and a content related to such networks, wherein the computer system includes at least one processor unit, the method comprising:
constructing a data representation of a social communication network as a multi-mode tensor having one or a plurality of content dimensions and one or a plurality of network dimensions;
decomposing the multi-mode tensor into a core tensor and a set of factor matrices, one factor matrix for each mode;
modeling each mode of the tensor, wherein the modeling comprises treating each corresponding factor matrix as a low-dimensional representation of that mode, and grouping rows of each factor matrix into clusters based on a distance measure;
modeling a cross-mode correlation, wherein the cross-mode correlation comprises clustering each modeled mode of the multi-mode tensor, enumerating all combinations of clusters across different modes and identifying the cross-mode clusters that induce dense sub-tensors; and
generating a visualization of the modeled cross-mode correlation;
wherein a program using the at least one processor unit executes one or more of said constructing, decomposing, modeling each mode, modeling a cross-mode correlation, and generating steps.
5. The method as claimed in claim 4 , further comprising receiving a content dimension representing a body or a topic of a message communicated from a respective sender to a recipient over said network.
6. The method as claimed in claim 4 , further comprising performing a tensor decomposition to reduce dimensionality of the social communication networks.
7. The method as claimed in claim 6 , further comprising performing a similarity search on the reduced dimensional presentation.
8. The method as claimed in claim 6 , further comprising clustering dimensions based on the reduced dimensional presentation.
9. The method as claimed in claim 4 , wherein said data representation of a social communication network further includes a time dimension.
10. The method as claimed in claim 4 , further comprising:
constructing a corresponding data representation of said social communication network at each of successive corresponding time intervals.
11. The method as claimed in claim 10 , wherein each successive corresponding time interval overlaps in time with a previous time interval.
12. A method implemented in a computer system for modeling and analyzing the time evolution of social communication networks and a content related to such networks, wherein the computer system includes at least one processor unit, the method comprising the steps of:
constructing a data representation of a social communication network as a multi-mode tensor having one or a plurality of content dimensions and one or a plurality of network dimensions, said constructing being performed at each of a plurality of successive time intervals;
decomposing the multi-mode tensor into a core tensor and a set of factor matrices, one factor matrix for each mode;
modeling each mode of the tensor, wherein the modeling comprises treating each corresponding factor matrix as a representation of that mode, and grouping rows of each factor matrix into clusters based on a distance measure;
modeling a cross-mode correlation, wherein the cross-mode correlation comprises clustering each modeled mode of the multi-mode tensor, enumerating all combinations of clusters across different modes and identifying the cross-mode clusters that induce dense sub-tensors; and
generating a visualization of the modeled cross-mode correlation;
wherein a program using the at least one processor unit executes one or more of said constructing, decomposing, modeling each mode, modeling a cross-mode correlation, and generating steps.
13. The method as claimed in claim 12 , wherein, for each time interval, said constructing said data representation includes receiving a content dimension representing a body or a topic of a message communicated from a respective sender to a recipient over said network.
14. The method as claimed in claim 12 , further comprising: receiving specification of one or more of a starting time instant and a time duration for said time intervals.
15. The method as claimed in claim 12 , further comprising performing a tensor decomposition to reduce dimensionality of the social communication network for each time interval.
16. The method as claimed in claim 12 , further comprising performing a similarity search on the reduced dimensional presentation at each time interval.
17. The method as claimed in claim 16 , further comprising clustering dimensions based on the reduced dimensional presentation at each time interval.
18. The method as claimed in claim 12 , wherein each successive corresponding time interval overlaps in time with a previous time interval.
19. The method as claimed in claim 12 , further comprising: utilizing a processing result from a prior time interval to speed up approximations for reducing dimensionality of the data representation provided in a next successive time interval.
20. A non-transitory computer usable medium, comprising:
a computer readable program to execute a plurality of instructions in a processing unit for modeling and analyzing the time evolution of social communication networks and a content related to such networks, the computer readable program, when executed by the processing unit, causes the processing unit to perform the steps comprising:
constructing a data representation of a social communication network as a multi-mode tensor having one or a plurality of content dimensions and one or a plurality of network dimensions, said constructing being performed at each of a plurality of successive time intervals;
decomposing the multi-mode tensor into a core tensor and a set of factor matrices, one factor matrix for each mode;
modeling each mode of the tensor, wherein the modeling comprises treating each corresponding factor matrix as a low-dimensional representation of that mode, and grouping rows of each factor matrix into clusters based on a distance measure;
modeling a cross-mode correlation, wherein the cross-mode correlation comprises clustering each modeled mode of the multi-mode tensor, enumerating all combinations of clusters across different modes and identifying the cross-mode clusters that induce dense sub-tensors; and
generating a visualization of the modeled cross-mode correlation.
21. The computer usable medium as claimed in claim 20 , further comprising performing a tensor decomposition on said data representation to reduce dimensionality of the social communication networks.