Supervised change detection in graph streams

1. A method, comprising the steps of:

obtaining a graph stream;

obtaining historical data of one or more nodes associated with the graph stream, wherein each node has associated with it a time-dependent label representative of activity of the node at a given time;

extracting one or more features from the graph stream for one or more nodes; and

creating one or more alarm levels for the one or more nodes based on the one or more extracted features and the historical data, wherein the step of creating one or more alarm levels comprises generating a regression model based on the one or more extracted features and the historical data and further wherein each of the one or more alarm levels is indicative of a probability of a change in a label class of an associated node in the graph stream;

wherein the steps are performed by at least one processor device coupled to a memory.

2. The method of claim 1 , wherein the step of extracting one or more features from the graph stream comprises creating a horizon graph.

3. The method of claim 2 , wherein the step of extracting one or more features from the graph stream further comprises using a random walk method on the horizon graph to construct one or more neighborhoods.

4. The method of claim 1 , wherein the one or more features extracted from the graph stream comprises one or more neighborhood features.

5. The method of claim 4 , wherein the one or more neighborhood features comprises a neighborhood size.

6. The method of claim 4 , wherein the one or more neighborhood features comprises a neighborhood class concentration.

7. The method of claim 4 , wherein the one or more alarm levels are generated as a score for each of the one or more nodes at each time instant.

8. The method of claim 1 , further comprising reporting the one or more alarm levels as a continuous time series for one or more nodes.

9. The method of claim 1 , further comprising applying a threshold to the one or more alarm levels.

10. The method of claim 9 , wherein an alarm level associated with a node that exceeds the threshold is indicative of an anomaly.

11. The method of claim 9 , wherein the one or more alarm levels are displayed as a discrete output.

12. The method of claim 1 , wherein the graph stream is associated with at least one pair-wise activity between two parties.

13. An apparatus comprising:

a memory; and

a processor operatively coupled to the memory and configured to:

obtain a graph stream;

obtain historical data of one or more nodes associated with the graph stream, wherein each node has associated with it a time-dependent label representative of activity of the node at a given time;

extract one or more features from the graph stream for one or more nodes; and

create one or more alarm levels for the graph stream based on the one or more extracted features and the historical data, wherein the one or more alarm levels are created by generating a regression model based on the one or more extracted features and the historical data and further wherein each of the one or more alarm levels is indicative of a probability of a change in a label class of an associated node in the graph stream.

14. The apparatus of claim 13 , wherein extracting one or more features from the graph stream comprises creating a horizon graph.

15. The apparatus claim 13 , wherein extracting one or more features from the graph stream further comprises using a random walk method on the horizon graph to create one or more neighborhoods.

16. The apparatus of claim 13 , wherein the one or more extracted features comprises at least one of a neighborhood size and a neighborhood class concentration.

17. The apparatus of claim 13 , wherein the one or more alarm levels are generated as a score for each of the one or more nodes at each time instant.

18. The apparatus of claim 13 , wherein the alarm is reported as a continuous time series for one or more nodes.

19. An article of manufacture comprising a computer readable storage medium for storing computer readable program code, which, when executed, causes a computer to:

obtain a graph stream;

obtain historical data of one or more nodes associated with the graph stream, wherein each node has associated with it a time-dependent label representative of activity of the node at a given time;

extract one or more features from the graph stream for one or more nodes; and

create one or more alarm levels for the graph stream based on the one or more extracted features and the historical data wherein the one or more alarm levels are created by generating a regression model based the one ore extracted features and the historical data and further wherein each of the one or more alarm levels is indicative of a probability of a change in a label class of an associated node in the graph stream.

20. The article of manufacture of claim 19 , wherein the alarm level is reported as a continuous time series for one or more nodes.