AI Search Suggestions not matching Search Results

Erik Nelson
Kilo Sage

Hello!

 

It's been brought to my attention that our AI Search Suggestions are not matching the returned results. For example, searching for "Decomm" displays three Catalog Items and two KB Articles. When completing the search though, only one result shows (one of the KB Articles).  I've already validated the items have associated taxonomy/topics and I've also reindexed the Catalog Item and Knowledge Base sources. 

 

I've checked here and the Support site and haven't found anything similar to this. Has anyone else experienced this and did you get a resolution for it? I would expect, in this case that at a minimum, the 5 total suggested items would appear in the full search results.

 

Thanks!

Erik

1 ACCEPTED SOLUTION

Erik Nelson
Kilo Sage

To anyone else running into this, I opened a case with ServiceNow and was provided with https://support.servicenow.com/kb?id=kb_article_view&sysparm_article=KB0687540.

 

TL;DR - This is the expected behavior of the platform, as the text search engine only uses complete indexed words or word stems. The third-party (Porter) adopted word-stemming procedure does not equalize to word truncation, such as to expect returning results implying a wildcard.

 

Because it's just not feasible to train every single person on how to use * or % in wildcard searches (this all came up because an IT person can't figure out how to refine searches), the alternative that I've tested and works is to add to your synonym dictionary to cover those partial words. Probably doesn't work for everything, but it works for things like "decomm" vs "decommission".

View solution in original post

3 REPLIES 3

Tanushree Maiti
Kilo Patron

Hi @Erik Nelson 

 

This seems to be a knowns issue.

This problem is expected to be resolved in the following ticket: PRB1691600.

 

The support team recommends turn off the tab count as a workaround. The counter will not appear for each tab, but the results will not change when changing tabs.

 

To turn off the tab count, go to your Search Application and uncheck the "Show Search Results Count" checkbox.

 

Note: AI Search ignores attachments larger than 25 MB in size. Standard indexing only indexes content from the first 1 MB of the attachment's data, while semantic vector indexing indexes the attachment's full content.

( not found it in any SN doc but confirmed in this post

https://www.servicenow.com/community/intelligence-ml-forum/ai-search-is-there-a-limit-on-how-many-wo....)

 

 

Refer: https://www.servicenow.com/community/architect-forum/ai-search-on-serviceportal-not-giving-correct-r...

 

https://www.findbugzero.com/operational-defect-database/vendors/sn/defects/PRB1691600

 

 

Please mark this response as Helpful & Accept it as solution if it assisted you with your question.
Regards
Tanushree Maiti
ServiceNow Technical Architect
Linkedin:

Thanks, but this is not the issue I'm experiencing. It isn't about the counts or attachments, it's about the displayed results.

Erik Nelson
Kilo Sage

To anyone else running into this, I opened a case with ServiceNow and was provided with https://support.servicenow.com/kb?id=kb_article_view&sysparm_article=KB0687540.

 

TL;DR - This is the expected behavior of the platform, as the text search engine only uses complete indexed words or word stems. The third-party (Porter) adopted word-stemming procedure does not equalize to word truncation, such as to expect returning results implying a wildcard.

 

Because it's just not feasible to train every single person on how to use * or % in wildcard searches (this all came up because an IT person can't figure out how to refine searches), the alternative that I've tested and works is to add to your synonym dictionary to cover those partial words. Probably doesn't work for everything, but it works for things like "decomm" vs "decommission".