Moderating online discussion using graphical text analysis

1. A computer program product, comprising:

a computer readable storage medium having program instructions embodied therewith, the program instructions readable by a processing circuit to cause the processing circuit to perform a method of moderating an online discussion between a plurality of participants, the method comprising:

receiving expressions from a participant during an online discussion;

generating a graph of expressions of the participant, wherein the graph comprises nodes that represent tokens, and edges that represent temporal precedence in the expressions, wherein each node possesses a feature vector, and wherein the feature vector is a direct sum of a syntactic vector, a semantic vector, and a predetermined vector for an identity of the participant;

performing a graphical text analysis on the graph to determine a likelihood of the participant to have a problematic behavior based at least in part on a number of edges between two nodes of the graph of expressions; and

performing a moderating action responsive to the likelihood exceeding a threshold likelihood, wherein the moderating action prevents the participant from posting an additional expression.

2. The computer program product of claim 1 , wherein the method further comprises:

generating at least one more graph of expressions of at least one more participant;

combining the graphs into a single graph; and

performing a graphical text analysis on the combined graph in order to identify which one or more of the participants have a problematic behavior.

3. The computer program product of claim 1 , wherein the performing the graphical text analysis comprises determining a likelihood of the graph to belong to one of a plurality of clusters of previously generated graphs, each cluster representing one or more different problematic behaviors.

4. The computer program product of claim 3 , wherein a problematic behavior represented by a cluster previously caused a problem in an online discussion.

5. The computer program product of claim 1 , wherein the performing the moderating action comprises sending a message to a human moderator of the discussion in order to cause the human moderator to take an appropriate action to influence a course of the discussion.

6. The computer program product of claim 1 , wherein the performing the moderating action comprises sending a message to the participant or limiting the participant's ability to participate in the discussion.

7. The computer program product of claim 1 , wherein the online discussion comprises one of a sequence of comments or posted messages in an online bulletin board, a sequence of comments under a post in a social network, a sequence of comments under an online article, a sequence of group emails, a live conversation in an online chat room or in an online video game, and text in systems that allow a participant to upload and download.

8. The computer program product of claim 1 , wherein the method further comprises:

combining the graph with another graph for another participant; and

performing a graphical text analysis on the combined graph to increase the likelihood.

9. A computer system for moderating an online discussion between a plurality of participants, the system comprising:

a memory having computer readable instructions; and

a processor configured to execute the computer readable instructions, the instructions comprising:

receiving expressions from a participant during an online discussion;

generating a graph of expressions of the participant, wherein the graph comprises nodes that represent tokens, and edges that represent temporal precedence in the expressions, wherein each node possesses a feature vector, and wherein the feature vector is a direct sum of a syntactic vector, a semantic vector, and a predetermined vector for an identity of the participant;

performing a graphical text analysis on the graph to determine a likelihood of the participant to have a problematic behavior based at least in part on a number of edges between two nodes of the graph of expressions; and

performing a moderating action responsive to the likelihood exceeding a threshold likelihood, wherein the moderating action prevents the participant from posting an additional expression.

10. The system of claim 9 , wherein the performing the graphical text analysis comprises determining a likelihood of the graph to belong to one of a plurality of clusters of previously generated graphs, each cluster representing one or more different problematic behaviors.

11. The system of claim 10 , wherein a problematic behavior represented by a cluster previously caused a problem in an online discussion.

12. The system of claim 9 , wherein the performing the moderating action comprises sending a message to a human moderator of the discussion in order to cause the human moderator to take an appropriate action to influence a course of the discussion.

13. The system of claim 9 , wherein the performing the moderating action comprises sending a message to the participant or limiting the participant's ability to participate in the discussion.

14. The system of claim 9 , wherein the online discussion comprises one of a sequence of comments or posted messages in an online bulletin board, a sequence of comments under a post in a social network, a sequence of comments under an online article, a sequence of group emails, a live conversation in an online chat room or in an online video game, and text in systems that allow a participant to upload and download.

15. A computer program product, comprising:

a computer readable storage medium having program instructions embodied therewith, the program instructions readable by a processing circuit to cause the processing circuit to perform a method of moderating an online discussion between a plurality of participants, the method comprising:

receiving expressions from a participant during an online discussion;

generating a graph of expressions of the participant, wherein the graph comprises nodes that represent tokens, and edges that represent temporal precedence in the expressions, wherein each node possesses a feature vector, and wherein the feature vector is a direct sum of a syntactic vector, a semantic vector, and a predetermined vector for an identity of the participant;

performing a graphical text analysis on the graph to determine a likelihood of the participant to have a problematic behavior based at least in part on a number of edges between two nodes of the graph of expressions; and

performing a moderating action responsive to the likelihood exceeding a threshold likelihood, wherein the moderating action prevents the participant from posting an additional expression,

wherein performing the graphical text analysis on the graph to determine a likelihood of the participant to have a problematic behavior is further based at least in part on a feature similarity of at least one feature.

16. The computer program product of claim 15 , wherein the at least one feature incorporates a syntactic component, a semantic component, and a dynamical component of the expression.