Multi layer chat detection and classification
Embodiments presented herein provide systems, methods and articles of manufacture for a computer-implemented method to monitor users interacting in an online multiuser environment. For each message sent by a user during a session, determining, via a first classifier, whether the message is either acceptable or unacceptable for the online multiuser environment. Upon satisfying a triggering condition for the session of the user, determining, via a second classifier, whether to escalate the session to a moderator for review.
1. A computer-implemented method for monitoring users interacting in an online multiuser environment, the method comprising:
training a first classifier with a plurality of messages labeled as either acceptable or unacceptable;
training the second classifier with a plurality of training example sessions, wherein each training example is generated from a session labeled as being either appropriate for escalation or inappropriate for escalation, and wherein each training example session includes messages from that session classified as unacceptable by the first classifier;
for each message sent by a first user during a session:
determining, via the first classifier, whether to label the message as being either acceptable or unacceptable for the online multiuser environment; and
upon satisfying a triggering condition for the session of the user, determining, via the second classifier, whether to escalate the session to a moderator for review.
2. The method of claim 1 , wherein the first classifier is a Bayesian Filter.
3. The method of claim 1 , wherein the second classifier is a Bayesian filter.
4. The method of claim 1 , wherein the triggering condition is a ratio of messages in the session classified as unacceptable and acceptable.
5. The method of claim 1 , wherein the triggering condition is a maximum number of messages in the session classified as unacceptable.
6. The method of claim 1 , wherein the second classifier determines whether to escalate the session to a moderator for review by evaluating messages in the session classified as unacceptable by the first classifier.
7. A non-transitory computer-readable storage medium comprising a program product which, when executed, is configured to perform an operation for monitoring users interacting in an online multiuser environment, the operation comprising:
training a first classifier with a plurality of messages labeled as either acceptable or unacceptable;
training the second classifier with a plurality of training example sessions, wherein each training example is generated from a session labeled as being either appropriate for escalation or inappropriate for escalation, and wherein each training example session includes messages from that session classified as unacceptable by the first classifier;
for each message sent by a first user during a session:
determining, via the first classifier, whether to label the message as being either acceptable or unacceptable for the online multiuser environment; and
upon satisfying a triggering condition for the session of the user, determining, via the second classifier, whether to escalate the session to a moderator for review.
8. The computer-readable storage medium of claim 7 , wherein the first classifier is a Bayesian Filter.
9. The computer-readable storage medium of claim 7 , wherein the second classifier is a Bayesian filter.
10. The computer-readable storage medium 7 , wherein the triggering condition is a ratio of messages in the session classified as unacceptable and acceptable.
11. The computer-readable storage medium 7 , wherein the triggering condition is a maximum number of messages in the session classified as unacceptable.
12. The computer-readable storage medium 7 , wherein the second classifier determines whether to escalate the session to a moderator for review by evaluating messages in the session classified as unacceptable by the first classifier.
13. A system, comprising:
a memory storing an application program; and
a processor which, when executing the application program is configured to perform an operation for monitoring users interacting in an online multiuser environment, the operation comprising:
training a first classifier with a plurality of messages labeled as either acceptable or unacceptable,
training the second classifier with a plurality of training example sessions, wherein each training example is generated from a session labeled as being either appropriate for escalation or inappropriate for escalation, and wherein each training example session includes messages from that session classified as unacceptable by the first classifier,
for each message sent by a first user during a session:
determining, via the first classifier, whether to label the message as being either acceptable or unacceptable for the online multiuser environment, and
upon satisfying a triggering condition for the session of the user, determining, via the second classifier, whether to escalate the session to a moderator for review.
14. The system of claim 13 , wherein the first classifier is a Bayesian Filter.
15. The system of claim 13 , wherein the second classifier is a Bayesian filter.
16. The system of claim 13 , wherein the triggering condition is a ratio of messages in the session classified as unacceptable and acceptable.
17. The system of claim 13 , wherein the triggering condition is a maximum number of messages in the session classified as unacceptable.
18. The system of claim 13 , wherein the second classifier determines whether to escalate the session to a moderator for review by evaluating messages in the session classified as unacceptable by the first classifier.