NLU not detecting phrases in the model

- Mark as New
- Bookmark
- Subscribe
- Mute
- Subscribe to RSS Feed
- Permalink
- Report Inappropriate Content
01-25-2021 11:57 AM
NLU is not detecting phrases that previously exist in the model. As an example, one of the phrases is "return my assets". When typing the exact same phrase it doesn't return any topics in VA UI. Any advice how to improve this? Perhaps this is a glitch that came with Patch5?
Sometimes when just testing phrases, we get the following black error:
Appreciate any advice to help identify what could be causing these errors.
Thanks!
- Labels:
-
Virtual Agent
- 2,549 Views
- Mark as New
- Bookmark
- Subscribe
- Mute
- Subscribe to RSS Feed
- Permalink
- Report Inappropriate Content
02-02-2021 06:38 AM
Hi Yasmin,
What release are you?
Does the logs show anything?
I'm asking because the NLU system uses ServiceNow servers to fetch the data and return the intent.
We are in Orlando Patch 9 recently deployed and also seeing the same issue. Don't think it's an issue with the patch itself, but an issue with the NLU servers.

- Mark as New
- Bookmark
- Subscribe
- Mute
- Subscribe to RSS Feed
- Permalink
- Report Inappropriate Content
02-02-2021 06:58 AM
Hi Luiz, we are in Paris. Our developers saw the following error on the log but cannot tell what is is:
- Mark as New
- Bookmark
- Subscribe
- Mute
- Subscribe to RSS Feed
- Permalink
- Report Inappropriate Content
11-12-2021 07:37 AM
Make sure "Is Topic Discoverable" is set to true on the published version of the topic which is in "sys_cs_topic"
Refer:
Troubleshooting steps for Virtual Agent not predicting the topic when NLU is on
Please mark this comment as Helpful/Correct Answer if it was helpful.
- Mark as New
- Bookmark
- Subscribe
- Mute
- Subscribe to RSS Feed
- Permalink
- Report Inappropriate Content
02-23-2023 03:34 AM
Same issue here. Testing the nlu model with certain utterances results in good predictions above threshold. But when using VA with the same input there are none shown, also the nlu log is showing correct predictions (open_nlu_predict_log). Same topics can be found with other utterances without issue. Kind of makes the batch testing set useless as well..