Method and system for correlating social media conversations

• US 10,003,560 B1
• Filed: 12/27/2012
• Issued: 06/19/2018
• Est. Priority Date: 08/31/2012
• Status: Active Grant

• Current Assignee: TPG Specialty Lending Incorporated
• Original Assignee: Sprinklr Inc.
• Inventors: Perkins, Jacob Andrew, Huddleston, Erik Lee, Renteria, Ramon, De Oliveira, John Joseph
• Primary Examiner(s): Salad, Abdullahi E
• Assistant Examiner(s): McAdams, Robert B
• Application Number:
  US13/727,991
• Time in Patent Office: 2,000 Days
• CPC Class Codes: H04L 51/216 Handling conversation history, e.g. grouping of messages in sessions or threads
• CPC Class Codes: H04L 51/52 for supporting social networking services

Abstract

A conversation correlation system is configured to associate social signals with conversations and group the conversations into topic clusters. The topic clusters for different days are linked together to form conversation streams. The conversation streams may identify conversation topics within ecosystems and identify changes in conversation topics over time. Topic labels may be generated for the topic clusters that better describe conversation topics. In one example, the topic labels may be generated using a mutual information topic distribution model.

Claims

1. A method, comprising:

identifying, by a computing device, social signal data based on social signals published using social media accounts, the social signal data including first information that includes a content of the social signals and second information that is different than the first information, wherein the second information includes metadata;

identifying, by the computing device, conversations in the social signals using said second information;

grouping, by the computing device, the conversations into topic clusters using the first information, wherein said grouping using the first information comprises;

forming a first set of topic clusters that corresponds to a first time; and

forming one or more second sets of topic clusters that correspond to one or more second different times, respectively; and

wherein, responsive to the grouping using the first information, a first one of the conversations is grouped with a second one of the conversations into a same one of the topic clusters of the first set or the one or more second sets, the first one of the conversations based on a root social signal of the social signals that is not explicitly linked, by the second information, with a different root social signal on which the second one of the conversations is based; and

linking, by the computing device, the topic clusters together into a plurality of conversation streams, wherein a first conversation stream of the plurality of conversation streams includes a topic cluster of the first set and at least one associated topic cluster of the one or more second sets, and wherein a second conversation stream of the plurality of conversation streams includes a different topic cluster of the first set and at least one associated topic cluster of the one or more second sets;

calculating, by the computing device, strengths of the different conversation streams based on a quantity of the social signals contributing to the conversation streams and a quantity and types of followers associated with the conversation streams; and

deriving, by the computing device, a score for one of the social media accounts based on the strengths of the conversation streams.

2. The method of claim 1, wherein identifying the conversations in the social signals using the second information comprises identifying ones of the socials signals associated with a same root message.

3. The method of claim 1, wherein the metadata is included in the social signals.

4. The method of claim 1, wherein the first time corresponds to a first day and the second time corresponds to a second different day.

5. The method of claim 1, wherein grouping the conversations into the topic clusters using the first information is based on similarity of terms of social signals of the conversations.

6. The method of claim 1, wherein linking the topic clusters together into the plurality of conversation streams comprises:

generating probabilities for terms associated with the topic clusters;

generating similarity values between the topic clusters based on the probabilities for the terms that match in different ones of the topic clusters; and

linking the topic clusters together into the plurality of conversation streams based on the similarity values.

7. The method of claim 6, further comprising;

identifying the topic clusters associated with different days of the week;

generating the similarity values between the topic clusters associated with the different days of the week; and

linking the topic clusters associated with the different days of the week together based on the similarity values.

8. The method of claim 1, further comprising deriving social metrics for the conversation streams.

9. The method of claim 1, further comprising:

identifying groups of the social signals associated with the conversation streams; and

generating social metrics for the groups of social signals.

10. The method of claim 1, further comprising:

identifying groups of the social signals associated with the topic clusters;

generating social metrics for the groups of social signals; and

combining the social metrics for the groups of social signals for the same topic clusters linked to the same conversation streams.

11. The method of claim 1, further comprising:

displaying, by the computing device, continuous time series image flows representing the conversation streams;

identifying, by the computing device, overall sentiments for topics associated with the conversation streams; and

displaying, by the computing device, indicators in conjunction with the time series image flows indicating changes in the overall sentiments for the topics associated with the conversation streams.

12. The method of claim 11, further comprising:

identifying, by the computing device, a quantity of the social signals associated with each of the conversation streams; and

displaying, by the computing device, different widths of the continuous time series image flows based on the quantity of the social signals associated with each of the conversation streams.

13. The method of claim 11, further comprising:

generating social metrics for the conversation streams; and

displaying the social metrics in response to detecting selection of the images representing the conversation streams.

14. The method of claim 1, further comprising:

generating a context distribution from the social signals;

generating candidate labels from the context distribution;

generating joint distributions for the candidate labels;

generating topic distributions for the topic clusters;

deriving mutual information values for the candidate labels based on the joint distributions for the candidate labels and the topic distributions for the topic clusters; and

selecting the candidate labels as topic labels for the topic clusters based on the mutual information values.

22. The method of claim 1, wherein the grouping, by the computing device, the conversations into topic clusters using the first information further comprises:

generating a first term vector for the first conversation, the first term vector for the first conversation including terms of the social signals of the first conversation;

generating a second term vector for the second conversation, the second term vector for the second conversation including terms of the socials of the second conversation; and

grouping the first and second conversations into a same one of the topic clusters of the first set or the one or more second sets based on similarity of the first and second term vectors.

15. An apparatus, comprising:

a memory configured to store social signal data based on social signals published using social media accounts, the social signal data including first information that includes a content of the social signals and second information that is different than the first information, wherein the second information includes metadata; and

a processor configured to;

identify conversations associated with the social signals using said second information;

group the conversations into clusters using the first information, wherein said grouping using the first information comprises;

forming a first set of clusters that correspond to a first time;

forming one or more second sets of clusters that correspond to one or more second different times, respectively;

wherein, responsive to the grouping using the first information, a first one of the conversations is grouped with a second one of the conversations into a same one of the clusters of the first set or the one or more second sets, the first one of the conversations based on a root social signal of the social signals that is not explicitly linked, by the second information, with a different root social signal on which the second one of the conversations is based;

identify topic distributions for the clusters by identifying probabilities for terms in the clusters;

generate a context distribution from the social signals;

identify candidate labels from the context distribution;

generate joint distributions for the candidate labels by identifying different groups of the social signals that include terms from the candidate labels and identifying probabilities for terms in the identified groups of social signals;

derive mutual information values for the candidate labels by applying the joint distributions for the candidate labels to the topic distributions for the clusters; and

select topic labels for the clusters from the candidate labels with the largest mutual information values.

16. The apparatus of claim 15, wherein the processor is further configured to link the clusters for different days together as a plurality of conversation streams based on the topic labels, wherein a first conversation steam of the plurality of conversation streams includes a cluster of the first set and at least one associated cluster of the one or more second sets, and wherein a second conversation stream of the plurality of conversation streams includes a different cluster of the first set and at least one associated cluster of the one or more second sets.

17. The apparatus of claim 15, wherein the context distribution comprises terms and probabilities of the terms occurring in the social signals of the ecosystem during a day.

18. The apparatus of claim 17, wherein processor is further configured to identify word level bigrams and trigrams from the context distribution with highest probabilities as the candidate labels.

19. The apparatus of claim 15, wherein the processor is further configured to:

identify the social signals associated with the clusters; and

generate the topic distributions based on terms and term probabilities for the social signals associated with the clusters.

20. The apparatus of claim 15, wherein the processor is further configured to:

link the clusters of social signals for different days together as conversation streams; and

display images representing the conversation streams.

21. The apparatus of claim 15, wherein the processor is further configured to display the topic labels associated with the clusters in response to detecting a selection of the conversation streams linked to the clusters.

23. An apparatus, comprising:

a memory configured to store social signal data based on social signals published using social media accounts, the social signal data including first information that includes a content of the social signals and second information that is different than the first information, wherein the second information includes metadata; and

a processor configured to;

identify conversations in the social signals using said second information;

group the conversations into topic clusters using the first information, including;

form a first set of topic clusters that corresponds to a first time; and

form one or more second sets of topic clusters that correspond to one or more second different times, respectively; and

wherein, responsive to the grouping using the first information, a first one of the conversations is grouped with a second one of the conversations into a same one of the topic clusters of the first set or the one or more second sets, the first one of the conversations based on a root social signal of the social signals that is not explicitly linked, by the second information, with a different root social signal on which the second one of the conversations is based; and

link the topic clusters together into a plurality of conversation streams, wherein a first conversation stream of the plurality of conversation streams includes a topic cluster of the first set and at least one associated topic cluster of the one or more second sets, and wherein a second conversation stream of the plurality of conversation streams includes a different topic cluster of the first set and at least one associated topic cluster of the one or more second sets;

calculate strengths of the different conversation streams based on a quantity of the social signals contributing to the conversation streams and a quantity and types of followers associated with the conversation streams; and

derive a score for one of the social media accounts based on the strengths of the conversation streams.

24. The apparatus of claim 23, wherein identifying the conversations in the social signals using the second information comprises identifying ones of the socials signals associated with a same root message.

25. The apparatus of claim 23, wherein the metadata is included in the social signals.

26. The apparatus of claim 23, wherein the first time corresponds to a first day and the second time corresponds to a second different day.

27. The apparatus of claim 23, wherein grouping the conversations into the topic clusters using the first information is based on similarity of terms of social signals of the conversations.

28. The apparatus of claim 23, wherein link the topic clusters together into the plurality of conversation streams comprises:

generate probabilities for terms associated with the topic clusters;

generate similarity values between the topic clusters based on the probabilities for the terms that match in different ones of the topic clusters; and

link the topic clusters together into the plurality of conversation streams based on the similarity values.

29. The apparatus of claim 23, wherein the processor is further configured to:

cause a continuous time series image flows representing the conversation streams to be displayed;

identify overall sentiments for topics associated with the conversation streams; and

cause indicators to be displayed in conjunction with the time series image flows to indicate changes in the overall sentiments for the topics associated with the conversation streams.

30. The apparatus of claim 29, wherein the processor is further configured to:

identify a quantity of the social signals associated with each of the conversation streams; and

cause different widths of the continuous time series image flows to be displayed based on the quantity of the social signals associated with each of the conversation streams.

31. The apparatus of claim 29, wherein the processor is further configured to:

generate social metrics for the conversation streams; and

cause the social metrics to be displayed in response to detecting selection of the images representing the conversation streams.

32. The apparatus of claim 23, wherein the processor is further configured to:

generate a context distribution from the social signals;

generate candidate labels from the context distribution;

generate joint distributions for the candidate labels;

generate topic distributions for the topic clusters;

derive mutual information values for the candidate labels based on the joint distributions for the candidate labels and the topic distributions for the topic clusters; and

select the candidate labels as topic labels for the topic clusters based on the mutual information values.

33. The apparatus of claim 23, wherein group the conversations into topic clusters using the first information further comprises:

generating a first term vector for the first conversation, the first term vector for the first conversation including terms of the social signals of the first conversation;

generating a second term vector for the second conversation, the second term vector for the second conversation including terms of the socials of the second conversation; and

grouping the first and second conversations into a same one of the topic clusters of the first set or the one or more second sets based on similarity of the first and second term vectors.