Identifying and Forecasting Shifts in the Mood of Social Media Users

• US 20130275352A1
• Filed: 01/25/2013
• Published: 10/17/2013
• Est. Priority Date: 04/17/2012
• Status: Active Grant

• Current Assignee: Mitre Corporation
• Original Assignee: Mitre Corporation
• Inventors: SERVI, Leslie David, ElSON, Sara Beth
• Granted Patent: US 9,015,089 B2
• Time in Patent Office: Days
• US Class Current: 706/21
• CPC Class Codes: G06N 3/08 Learning methods
• CPC Class Codes: G06Q 10/00 Administration; Management

Abstract

Quantitatively identifying and forecasting shifts in a mood of social media users is described. An example method includes categorizing the textual messages generated from the social media users over a selected period of time into a plurality of word categories, with each word category containing a set of words associated with the mood of social media users. A score indicating an intensity of the mood of the social media users is calculated for each word category, wherein a value of the score and its corresponding time point define a data point for the word category. Subsequently, breakpoints in the mood of social media users are determined so that the breakpoints minimize a sum of square errors representing a measurement of a consistency of all data points from inferred values of the scores of the data points derived using the breakpoints over the selected period of time. Further, space of all possible breakpoints for the word categories are searched to identify a defined number and locations of the breakpoints. Finally the breakpoints over the selected period of time are interpreted to identify the shifts in the mood of social media users and trends between breakpoints.

Claims

1. A quantitative method for identifying shifts in a mood of social media users, comprising:

categorizing textual messages generated from the social media users over a selected period of time into a plurality of word categories, wherein each word category contains a set of words associated with the mood of social media users;

for each word category, calculating a score representing an intensity of the mood of the social media users, wherein a value of the score and its corresponding time point define a data point for the word category;

determining breakpoints in the mood of social media users that minimize a sum of square errors representing a measurement of a consistency of all data points from inferred values of the scores of the data points derived using the breakpoints over the selected period of time, wherein space of all possible breakpoints for the word categories are objectively searched to identify a defined number and locations of the breakpoints; and

interpreting the breakpoints over the selected period of time to identify the shifts in the mood of social media users.

2. The method of claim 1, further comprising:

generating dictionaries that define the set of words contained in each word category.

3. The method of claim 1, identify a trend in the mood of social media users between two consecutive breakpoints.

4. The method of claim 1, wherein the determining is based on a constant model, wherein the score for each word category is modeled as constant in time within a breakpoint region.

5. The method of claim 1, wherein the determining is based on a linear model, wherein the score for each word category is modeled as linear within a breakpoint region.

6. The method of claim 1, wherein the determining is based on:

7. The method of claim 6, wherein the determining is based on:

{right arrow over (y)}_j^k*=(A_j^T({right arrow over (m)})PA_j({right arrow over (m)}))<sup>−

1</sup>A_j^T({right arrow over (m)})P{right arrow over (x)}_j^k

wherein {right arrow over (y)}_j^k* denotes a best choice of {right arrow over (y)}_j^kdefined in claim 6.

8. The method of claim 6, wherein the determining is based on a fast, recursive algorithm defined as:

9. The method of claim 8, wherein the determining is based on:

10. The method of claim 6, wherein determining is based on a real time algorithm defined as:

11. The method of claim 3, wherein identifying the trend further comprising:

identifying the trend to forecast future mood shifts of the social media users based on previous breakpoints determined from historic textual messages.

12. The method of claim 11, wherein the identifying the trend further comprises extending a derived trend line from a most recent breakpoint to determine a forecasted value.

13. The method of claim 11, wherein the identifying the trend further comprises extending a derived trend line from a most recent breakpoint with a horizontal slope to determine a forecasted value.

14. The method of claim 1, wherein the determining is based on a Dth degree polynominal model, wherein the score for each word category is modeled as a Dth degree polynominal within a breakpoint region.

15. The method of claim 1, wherein the determining is based on a finite parameter model, wherein the score for each word category is modeled as a finite parameter within a breakpoint region.

16. The method of claim 1, where in the calculating further comprises calculating a ratio of number of words in the textual messages falling into the word category to a total number of words in the textual messages.

17. The method of claim 1, wherein the textual messages are short messages generated from tweet and other social media.

18. The method of claim 1, wherein the textual messages are generated from sources other than social media.

19. A system for quantitatively identifying shifts in a mood of social media users, comprising:

a message categorizer, configured to categorize textual messages generated from the social media users over a selected period of time into a plurality of word categories, wherein each word category contains a set of words associated with the mood of social media users;

a score calculator, configured to, for each word category, calculate a score representing an intensity of the mood of the social media users, wherein a value of the score and its corresponding time point define a data point for the word category;

a breakpoint determinator, configured to determine breakpoints in the mood of social media users that minimize a sum of square errors representing a measurement of a consistency of all data points from inferred values of the scores of the data points derived using the breakpoints over the selected period of time, wherein space of all possible breakpoints for the plurality of word categories are objectively searched to identify a defined number and locations of the breakpoints; and

a breakpoint interpreter, configured to interpret the breakpoints over the selected period of time to identify the shifts in the mood of social media users.

20. The system of claim 19, further comprising:

a dictionary generator, configured to generate dictionaries that define the set of words contained in each word category.

21. The system of claim 19, further comprising:

a trend identifier, configured to identify a trend in the mood of social media users between two consecutive breakpoints.

22. The system of claim 19, wherein the breakpoint determinator is further configured to determine the breakpoints based on a constant model, wherein the score for each word category is modeled as constant in time within a breakpoint region.

23. The system of claim 19, wherein the breakpoint determinator is further configured to determine the breakpoints based on a linear model, wherein the score for each word category is modeled as linear within a breakpoint region.

24. The system of claim 19, wherein the breakpoint determinator is further configured to determine the breakpoints based on:

25. The system of claim 19, wherein a set of optimal breakpoints are calculated based on:

{right arrow over (y)}_j^k*=(A_j^T({right arrow over (m)})PA_j({right arrow over (m)}))<sup>−

1</sup>A_j^T({right arrow over (m)})P{right arrow over (x)}_j^k

wherein {right arrow over (y)}_j^k* denotes a best choice of {right arrow over (y)}_j^kdefined in claim 24.

26. The system of claim 19, wherein the trend identifier further comprises:

a mood forecaster, configured to forecast future mood shifts of the social media users based on previous breakpoints determined from historic textual messages.

27. The system of claim 26, wherein the mood forecaster is further configured to extend a derived trend line from a most recent breakpoint to determine a forecasted value.

28. The system of claim 26, wherein the mood forecaster is further configured to extend a derived trend line from a most recent breakpoint with a horizontal slope to determine a forecasted value.

29. The system of claim 19, wherein the breakpoint determinator is further configured to determine the breakpoints based on a Dth degree polynominal model, wherein the score for each word category is modeled as a Dth degree polynominal within a breakpoint region.

30. The system of claim 19, wherein the breakpoint determinator is further configured to determine the breakpoints based on a finite parameter model, wherein the score for each word category is modeled as a finite parameter within a breakpoint region.

31. The system of claim 19, where in the score calculator is further configured to calculate a ratio of number of words in the textual messages falling into the word category to a total number of words in the textual messages.

32. The system of claim 19, wherein the textual messages are short messages generated from tweet and other social media.

33. The system of claim 19, wherein the textual messages are generated from sources other than social media.

34. A non-transitory computer-readable medium storing software comprising instructions executable by one or more computers which, upon such execution, cause the one or more computers to perform operations comprising:

categorizing textual messages generated from the social media users over a selected period of time into a plurality of word categories, wherein each word category contains a set of words associated with the mood of social media users;

for each Word category, calculating a score representing an intensity of the mood of the social media users, wherein a value of the score and its corresponding time point define a data point for the word category;

determining breakpoints in the mood of social media users that minimize a sum of square errors representing a measurement of a consistency of all data points from inferred values of the scores of the data points derived using the breakpoints over the selected period of time, wherein space of all possible breakpoints for the word categories are searched to identify a defined number and locations of the breakpoints; and

interpreting the breakpoints over the selected period of time to identify the shifts in the mood of social media users and the trends between the breakpoints.