How does Knowledge Base search work?

[C Avila]

I want to properly train our knowledge article contributors to

A.  properly document meta and to 

B.  understand how the relevancy score is calculated and to

C.  understand how different fields effect the relevancy score in regards to the specified fields with assigned weights and those without weights

but I myself do not understand these things.  So I need some clarification please.  

1. What I’ve seen in our organization is in an increase in the repetition of the same search term in the meta field is this necessary?
2. I do not clearly understand if the knowledge search capability is the same as global search. 
3. Nor, do I understand if the KB search engine understands phrases or just words.
4. If it understands phrases, what is the distinction between one phrase and another?  a hard return, a comma, a semicolon etc.
5. How KB search interprets spaces, underscores, etc
6. Do we need multiple versions of the same word? Ex: Login, Log in, Logins, Logging in, Can’t login, Log on
7. I'm assuming caps vs lower case letters don't matter but can't say for certain.  I see both upper versions of acronyms and then the full lower case version of the same acronym in the meta.  Is this necessary?

Thank you in advance for any information you can provide!

[Leri Andrews · Accepted solution]

There's a fair bit of documentation on how Zing works. e.g. https://docs.servicenow.com/en-US/bundle/sandiego-platform-administration/page/administer/search-administration/concept/c_ZingTextSearch.html

I don't have answers to your specific questions because I agree with you that it's a bit of a dark art, not a science (from the knowledge manager point of view). 

The best advice I can give is to use the meta very sparingly and concentrate on well-written articles and upskilling your authors.  

For example, you'd think it would be easy to write an article about ordering a company car and have it 'found' but no.  If you're not careful the article will be titled 'Fleet ordering procedure' and refer to 'vehicles' thoughout in passive HR jargonese e.g. "The manager must ensure that the vehicle is appropriate for the incumbent....".   You'll be adding 'car' and 'company car' into the meta and considering a synonym dictionary when what is really required is an article called "How to order your company car" which refers to 'car' several times. 

If your well-written article is STILL not appearing very highly, then it's time to add a couple of choice keywords in the meta and perhaps review your weighting in ZING.

You can fiddle with your weightings if you want to give meta a higher weighting.  You can allow the field to be amended without going through a publishing approval stage and you could set up a catalog item for meta to be approved before being added - to weed out the times people will add the word 'form' or 'HR' which are simply meaningless noise.  We found that words at the front of the meta seemed more weighty and also you can add a word several times if necessary. We did end up with 'company car, car, company, vehicle, fleet, fleet vehicle' and some used spaces and some used commas and some did a line return - they all worked.  I don't think capitals mattered. 

Final word of caution, AI search doesn't even index meta by default (though people are asking for it, and you can if you want) so if you're moving that way it will be a bit of a waste putting your effort into meta.  I'm in the position now where portal uses AI search and back end still uses Zing so we're still adding the meta (a compulsory field) but it does nothing for employees.

[Barry_W]

Hi Leri,

Thanks for the detailed explanation, very useful. How are you finding the AI search in comparison to Zing? It seems logical to make use of AI where available, however we've invested some time in using meta data so I'm concerned how losing the value of that system will affect search results. I've found it quite difficult to measure search effectiveness in the past and am obviously keen not to make search results less relevant.

[Leri Andrews]

Hi Barry

We attempted some empirical testing before we launched to compare Zing and Ai search (tricky) and found it varied between languages used but in all cases it was better than zing. English is the best - you get extra features such as genius results.  Chinese is the worst - there's no tokenisation support so it has no idea where 'words' start and end in a long phrase without spaces.  Documentation from SN is somewhat vague, so we're finding out stuff as we go along. For example, our employees have a tendency to still think in terms of Zing keywords and will type in a single word like 'benefits' and wonder why nothing useful comes up.  Investigation and interpretation of https://docs.servicenow.com/bundle/sandiego-platform-administration/page/administer/ai-search/concep... has shown that it will look at the words you have used and if there are more than 1 it will look for word1 AND word2 AND word3 AND word4 and if it has no luck it will have another go and look for word1 OR word2 OR word3 OR word4.  If you use just one word it’s going to return lots and not even bother to go for another look. 

We're still working blind until we enable the analytics so it's essentially training itself and doing its own thing without any oversight. We did make some early decisions to help it out - we've indexed English + Local Language together (English + German, English + French, (English + Swedish) so that people who have chosen Swedish aren't prevented from seeing articles only written in English return in results. We've got user criteria at the article level restricting to country so that there's no chance the Ai is going to stick a Poland article on the same topic at the top of the list because of some invisible decision it's made. 

Trying to explain to people who have chosen English as their portal language (they think) but are actually instructing AI search to only look in the English 'bucket of words' for a Polish word they typed is quite tricky though.  Also it really highlights issues with catalog items which are far less rich in terms of words and often surprisingly poorly named.  For example, a user will type in 'order new mobile' and not get the 'Peripheral device requisition form' becasue it doesn't mention the word 'mobile' anywhere.....

Best of luck anyway!

Leri

[Barry_W]

Thanks Leri. Sounds like good progress, but also a challenge in terms of having your user base change their behaviour slightly.  I've written various guides for our customer base on best practise knowledge base searching, but struggle to have users read it. I guess the best knowledge base shouldnt need a guide though!

 ".....finding out stuff as we go along" - I know this feeling very well!

Thanks again,

Barry.