- Mark as New
- Bookmark
- Subscribe
- Mute
- Subscribe to RSS Feed
- Permalink
- Report Inappropriate Content
06-21-2023 01:19 AM
Hi Experts ,
When tying any junk value e.g - hhghgh, ggghhh , it is still showing the flow contents instead of the error message.
Can some one please suggest how to fix this issue ?
Thanks in advance.
Solved! Go to Solution.
- Mark as New
- Bookmark
- Subscribe
- Mute
- Subscribe to RSS Feed
- Permalink
- Report Inappropriate Content
06-21-2023 05:59 AM - edited 06-21-2023 05:59 AM
Hey good day!
When typing junk utterances, you should be getting the Fallback topic, not the Error Message
This is due to the fallback flow in the Virtual Agent - first it tries Topic Discovery with NLU, then Keyword. If neither return a topic, then if AI Search is enabled on your instance, VA tries an AI Search. If no results come back from that, then it will push you straight to the Fallback topic, which OOTB sends you to Capabilities for 'rapid reprompt'.
First, I would test from the 'Test Active Topics' button in the Designer home page (where you see all the tiles for the topics). This will provide you feedback on whether your junk utterances are matching an Intent, and if so, which one?
From there, I would recommend that you improve the quality of your Utterances in each Intent. This means making sure that your Utterances are clear, concise, and representative of how your users would ask the Virtual Agent. Utterances should NOT be keyword-oriented, instead they should be using Natural Language (E.g. 'The printer isn't working' instead of 'printer issue').
Finally, I would raise the Confidence Threshold of the Model to around 75%. This means that NLU needs to be more confident in it's prediction in order to return a topic, thereby reducing the chance of utterances erroneously matching to topics.
Once you have a well-tuned and configured NLU Model, you should stop seeing this issue!
- Mark as New
- Bookmark
- Subscribe
- Mute
- Subscribe to RSS Feed
- Permalink
- Report Inappropriate Content
06-21-2023 01:43 AM
Hi @Neeta1,
Set the NLU model threshold of up to 70 in the associate topic.
Please refer to the ServiceNow Doc
I hope this helps!
Regards,
Hemant
- Mark as New
- Bookmark
- Subscribe
- Mute
- Subscribe to RSS Feed
- Permalink
- Report Inappropriate Content
06-21-2023 05:59 AM - edited 06-21-2023 05:59 AM
Hey good day!
When typing junk utterances, you should be getting the Fallback topic, not the Error Message
This is due to the fallback flow in the Virtual Agent - first it tries Topic Discovery with NLU, then Keyword. If neither return a topic, then if AI Search is enabled on your instance, VA tries an AI Search. If no results come back from that, then it will push you straight to the Fallback topic, which OOTB sends you to Capabilities for 'rapid reprompt'.
First, I would test from the 'Test Active Topics' button in the Designer home page (where you see all the tiles for the topics). This will provide you feedback on whether your junk utterances are matching an Intent, and if so, which one?
From there, I would recommend that you improve the quality of your Utterances in each Intent. This means making sure that your Utterances are clear, concise, and representative of how your users would ask the Virtual Agent. Utterances should NOT be keyword-oriented, instead they should be using Natural Language (E.g. 'The printer isn't working' instead of 'printer issue').
Finally, I would raise the Confidence Threshold of the Model to around 75%. This means that NLU needs to be more confident in it's prediction in order to return a topic, thereby reducing the chance of utterances erroneously matching to topics.
Once you have a well-tuned and configured NLU Model, you should stop seeing this issue!