METHOD AND SYSTEM TO PREDICT THE LIKELIHOOD OF TOPICS

• US 20100280985A1
• Filed: 01/13/2009
• Published: 11/04/2010
• Est. Priority Date: 01/14/2008
• Status: Active Grant

• Current Assignee: United States Navy (U.S. Department Of Defense)
• Original Assignee: Aptima Inc.
• Inventors: Kramer, David, Colonna-Romano, John, Duchon, Andrew E., Salter, William J., Allopenna, Paul David, Weil, Shawn, McCormack, Robert
• Granted Patent: US 9,165,254 B2
• Time in Patent Office: Days
• US Class Current: 706/52
• CPC Class Codes: G06N 7/01 Probabilistic graphical models, e.g. probabilistic networks
• CPC Class Codes: G06Q 10/10 Office automation; Time management

Abstract

The present invention relate to a method and system to predict the likelihood of data topics that may occur from data sources. The likelihood of the data topics may be predicted over other dimensions of time or over other dimensions. In the present invention, a topic means a defining characteristic, usually represented as a data element, of a single feature, activity, subject, behavior, event or an aggregation of such defining characteristics.

Claims

1. A computed based system for predicting the value of a topic, said system comprising:

a processor capable of executing machine instructions on a set of data;

the machine instructions including;

means for transforming the set of data over at least one dimension value to create a topic profile, and

a means for predicting a predicted topic value at a predicted dimension value.

2. The computer based system of claim 1 wherein the set of data comprises news data.

3. The computer based system of claim 1 wherein the set of data comprises temporal data received by the processor from a communications network.

4. The computer based system of claim 1 wherein:

the means for transforming the set of data further comprises analyzing the set of data using latent semantic analysis to create the at least one topic; and

the topic profile comprises at least one topic value of the at least one topic.

5. The computer based system of claim 1 wherein:

the means for transforming the set of data further comprises analyzing the set of data using latent Dirichlet allocation to create the at least one topic; and

the topic profile comprises at least one topic value of the at least one topic.

6. The computer based system of claim 1 wherein:

the means for transforming the set of data further comprises analyzing the set of data using probabilistic latent semantic analysis to create the at least one topic; and

the topic profile comprises at least one topic value of the at least one topic.

7. The computer based system of claim 1 wherein the means for predicting the predicted topic value comprises using an autoregressive fractionally integrated moving average (AFIMA) model of the topic profile.

8. The computer based system of claim 1 wherein the means for predicting the predicted topic value comprises using an autoregressive conditional heteroskedasticity (ARCH) model of the topic profile.

9. The computer based system of claim 1 further comprising a user interface wherein a user can modify the set of data.

10. The computer based system of claim 1 wherein the means for predicting the predicted topic value further comprises:

identifying a base dimension value having a base topic profile;

identifying the predicted dimension value having a difference dimension value from the base dimension value;

analyzing the topic profile over at least one dimension value to identify a most similar topic profile to the base topic profile;

the dimension value at the most similar topic profile being a selected dimension value; and

selecting at least one topic value from the topic profile at a difference dimension value from the selected dimension value as the predicted topic value at the predicted dimension value.

11. The computer based system of claim 10 wherein the means to analyze the topic profile to identify the most similar topic profile further comprises:

at least two topic profiles;

each of the at least two topic profiles having at least one vector;

a similarity factor identifies the most similar topic profile; and

the similarity factor is identified by a method selected from the group consisting of;

defining a cosine of an angle between the at least two vectors as the similarity factor to identify the most similar topic profile,

defining an inverse of a Euclidian distance of the at least two vectors as the similarity factor to identify the most similar topic profile, and

defining a Minkowski distance between the at least two vectors in space as the similarity factor to identify the most similar topic profile.

12. A computer based method for predicting a topic value, said method comprising the steps of:

transforming a set of data having at least one topic over at least one dimension value to create at least one topic profile relating at least one topic value to the at least one dimension value;

identifying a base dimension value and a predicted dimension value, the predicted dimension value having a difference dimension value from the base dimension value;

identifying a base topic profile at the base dimension value;

analyzing the at least one topic profile to identify a most similar topic profile to the base topic profile;

identifying the dimension value of the most similar topic profile as a selected dimension value;

identifying the topic profile at the difference dimension value from the selected dimension value as an at least one predicted topic profile at the predicted dimension value; and

identifying at least one topic value from the predicted topic profile as an at least one predicted topic value.

13. The computer based method of claim 12 wherein the step of transforming the historical data is performed by a specially programmed computer comprising a series of machine instructions residing in a memory and executed by a processor.

14. The computer based method of claim 12 wherein the set of data comprises news data.

15. The computer based method of claim 12 wherein the set of data comprises graphic data.

16. The computer based method of claim 12 wherein the set of data comprises temporal data received by a processor from a communications network.

17. The computer based method of claim 12 further comprising receiving modifications to the set of data.

18. The computer based method of claim 12 wherein the data comprises data representing physical phenomena.

19. The computer based method of claim 12 further comprising:

analyzing the set of data with at least one of the techniques from the group consisting of latent semantic analysis, probabilistic latent semantic analysis and latent Dirichlet allocation to create the at least one topic.

20. The computer based method of claim 12 wherein the step of analyzing the at least one topic profile comprises:

at least two topic profiles;

each of the at least two topic profiles having at least one vector;

identifying the most similar topic profile with a similarity factor; and

the similarity factor is defined by a method selected from the group consisting of;

defining a cosine of an angle between the at least two vectors as the similarity factor,

defining an inverse of a Euclidian distance of the at least two vectors as the similarity factor, and

defining a Minkowski distance between the at least two vectors in space as the similarity factor.

21. A computer based method for predicting the value of a topic comprising the steps of:

transforming a set of data having at least one topic over at least one dimension value to create at least one topic profile relating at least one topic value at the at least one dimension value;

identifying a base dimension value and a predicted dimension value, the predicted dimension value having a difference dimension value from the base dimension value; and

analyzing the at least one topic profile to predict the topic profile from the base dimension value to the difference dimension value to create a predicted topic value at the predicted dimension value.

22. The computer based method of claim 21 wherein the step of transforming the historical data is performed by a specially programmed computer comprising a series of machine instructions residing in a memory and executed by a processor.

23. The computer based method of claim 21 wherein the set of data comprises news data.

24. The computer based method of claim 21 wherein the set of data comprises graphic data.

25. The computer based method of claim 21 wherein the set of data comprises temporal data received by a processor from a communications network.

26. The computer based method of claim 21 further comprising receiving modifications to the set of data.

27. The computer based method of claim 21 wherein the data comprises data representing physical phenomena.

28. The computer based method of claim 21 further comprising:

analyzing the set of data with at least one of the techniques from the group consisting of latent semantic analysis, probabilistic latent semantic analysis and latent Dirichlet allocation to create the at least one topic.

29. The computer based method of claim 21 wherein the step of analyzing the at least one topic profile comprises using a vector autoregression technique.

30. The computer-based method of claim 29 wherein the vector autoregression technique is chosen from at least one of the group consisting of:

an autoregressive fractionally integrated moving average (AFIMA) model;

an autoregressive moving average (ARMA) model;

an autoregressive integrated moving average (ARIMA) model; and

an autoregressive conditional heteroskedasticity (ARCH) model.