Questions on Virtual Agent AI Search

Junyu
Tera Contributor

Dear all experts,

 

Our department is very interested in Virtual Agent AI Search and has some questions:

1. Search configuration

2. Sequencing and focus on search results

3. user rating and relevancy influence on search results

4. Connected knowledge sources & APIs

 

Can you guys provide any answers/relevant documentation/videos?

Thank you for your answer!

1 ACCEPTED SOLUTION

  1. Search results are based off whatever fields you want - there's ootb configurations for Catalog Items, Knowledge, and Users which should be sufficient for most. Note: if iirc, I think they may exclude "Meta", though you can add that to the config easily. Result order is determined by search relevance by default, which is effectively a black box, but you can configure Result Improvement Rules to boost whatever search results you want, whether as broadly as "always boost all catalog items" or as specific as "boost this one kb article if the search term includes X". NLU is used for the Genius Results only I believe.
  2. This I'm not entirely sure. I don't think user KB article ratings affect search results, but I am pretty sure that user interactions with results (i.e. which item they click) does affect search results over the long term. They recently added a "was this helpful" button to Genius Results as well, but I don't think that automatically does anything to the results - if I recall, that just gets saved as data for the search admin to review and make any manual adjustments to the config.

View solution in original post

3 REPLIES 3

Chris D
Mega Sage
Mega Sage

AI Search is pretty good, no complaints having almost a year of implementation under our belts now. While you can "set and forget" it after doing initial configuration, it does tend to lend itself towards some at least semi-regular maintenance to ensure certain search results are promoted/avoided, etc.

Take a read through ServiceNow's Docs on AI Search (servicenow.com) and come back here with any specific questions that aren't answered there.

Junyu
Tera Contributor

Hi @Chris D ,

I have two questions of search results:

1. What sort of search results are based on (e.g. keywords, NLU...) and what is the priority (e.g. VA topic, catalog item, article...)?

2. How do user ratings and relevance affect AI search results? For example, a five-star less relevant article and a one-star very relevant article. . .?

 

Thank you for your answer.

  1. Search results are based off whatever fields you want - there's ootb configurations for Catalog Items, Knowledge, and Users which should be sufficient for most. Note: if iirc, I think they may exclude "Meta", though you can add that to the config easily. Result order is determined by search relevance by default, which is effectively a black box, but you can configure Result Improvement Rules to boost whatever search results you want, whether as broadly as "always boost all catalog items" or as specific as "boost this one kb article if the search term includes X". NLU is used for the Genius Results only I believe.
  2. This I'm not entirely sure. I don't think user KB article ratings affect search results, but I am pretty sure that user interactions with results (i.e. which item they click) does affect search results over the long term. They recently added a "was this helpful" button to Genius Results as well, but I don't think that automatically does anything to the results - if I recall, that just gets saved as data for the search admin to review and make any manual adjustments to the config.