Expert Feedback Loop Issue

[johndoh]

Hello all,

Someone decided to write a novel for an utterance that the expert feedback loop found. We had this marked as not sure and accepted some other items. After is when I attempt to train the model to complete the process and I am getting:

Exception caught in submitTrainingJob method: Synchronous training failed - reason: {"status":"failure","response":{"messages":[{"type":"ERROR","message":"Enter utterances that only contain less than 200 characters and less than 25 words.","messageKey":"Enter utterances that only contain less than {0} characters and less than {1} words.","replacements":["200","25"],"

I updated the item in ml_solution, open_nlu_predict_intent_feedback, open_nlu_predict_feedback, open_nlu_predict_entry_feedback, open_nlu_predict_log, and ml_label_candidate. 

Until I updated the ml_label_candidate the feedback loop would only show the invalid version. At this point I attempted to update the item in EFL and matched it to a known topic. This still would not resolve the issue and still getting the "Enter utterances that only contain less than 200 characters and less than 25 words. Any suggestions?

Thanks John

[RenatoMendoza]

Hello @johndoh, Did you happen to find a solution to this problem? We are experiencing this as well. I am looking at all the related tables, but still no luck.

[johndoh]

deleting the item from ml_labeled_data resolved the issue.

From HIWAVE Ticket: It looks like the long text was not picked up as an "utterance" but rather as a snippet of text from the VA Chat Logs. This would explain why you cannot find it under the sys_nlu_utterance table (and likewise why the validate length business rule did not block the record in the first place). I was able to find the text under the [ml_labeled_data] table. I will admit our documentation does not make it clear that the "utterances" you see in Expert Feedback Loop are not actual Utterances but rather just snippets of text.

[RenatoMendoza]

Thank you very much @johndoh, this workaround does indeed work. After deleting the records in ml_labeled_data with more than 200 characters and records with more than 25 words, we were able to train the model once again.

[Brian Bakker]

@RenatoMendoza 

This defect was fixed a while back, so please ensure you have upgraded to the latest version of the "NLU Workbench - Advanced Features" application. It throws the error when training the NLU Model, so feedback would have been provided and the utterance that exceeds 200 chars or 25 words will be located in table [ml_labeled_data] and will need deleting to fix this issue.

For further details, log into ServiceNow Support and review my KB1633901: [NLU] Active Learning (AL) and Expert Feedback Loop (EFL) - Further insights on the utterance extraction from the Virtual Agent (VA) chat log to enable NLU Admins to provide feedback and further improve their NLU Models.

If the issue persists, log a case with ServiceNow Support, as we can use back-end queries to identify the utterances causing an error when training the NLU Model.

Hope this helps.

Regards,

Brian

[RenatoMendoza]

Thank you very much for your response @Brian Bakker, I will check that the version is updated and if the error persists, I will open a case with support.