Searches results and 'weights'

[katherinejones]

I have looked at several portions of this WIKI for this answer and not found anything that specifically addresses my question. Usually that means that what I want is not possible, but I will ask anyway.

When performing searches (Knowledge Searches in particular) and looking at relevance, we have the weight thing (Number, Short Description and Metadata) and we have the Search method used when searching Knowledge from a task or directly in the Knowledge Base (glide.knowman.search.operator) -- OR vs. AND then OR.

Is there any way to add relevance when a multi-word search includes the words in order or within close proximity?   For instance

Search term = apple pie

KB 1 has this phrase in the Meta: Given a choice of peach, apple, or pumpkin pie, my favorite is my mom's apple with cinnamon.

KB 2 has this in the Meta: My mom's recipe for apple pie is the bomb!

KB 3 has this in the Meta:   word pie word word word table word word word bullet-list apple.

I want KB 2 to have the most weight because it contains the term 'apple pie'.

I want KB 1 to have some weight because the terms apple and pie are in close proximity of each other

I want KB 3 to have the least weight because there is a lot of 'stuff' between the word pie and apple.

Thanks!

[rgm276]

Katherine

You should be able to play games with the ts_weights and use of the meta tags, i.e. you can put as many variations of a word (or phrase) in the meta tag, and each 'matched' variation should increase the relevancy score of that kb. So, in your example, if I have 3 KB's regarding various pies, by having multiple meta tags in the Apple Pie KB of apple, apple pie, applepie, pie, strawberry, peach.. then various searches will make each of the 'wanted' values appear at the top of the search list. Do the same for each of the KB's that you want to vary the 'relevance'.

you can temporarily turn on the search relevancy by changing the system property

glide.knowman.search.show_relevancy

so, an example: apple pie search

if you have the following description and meta tag values, and if you want 'similar' once to show up, then put that value in the other KB article meta tags only once, so that they will not overtake the one you want to be on top of the results

you can change the relevancy rules by reviewing this

Administering Zing Text Search - ServiceNow Wiki

finally, I know that the knowledge group has been looking into "best bets" for search relevance, but not yet available, if you want to read more, you can see that blog info here

Using ts_weight to impact search relevancy in Knowledge

[katherinejones]

This does work, but only at the expense of adding words or phrases to the meta data that are not logically part of the 'content' of the KB.   That said, I did find this information, which implies that the relevance points are higher when a multi-word search includes the words in order (even if other words are between them):

2 Frequency Points

Points are awarded to a document for the frequency (one point per occurrence) in which the search terms appear anywhere in the document.

• For example, in searching for "wild purple hyenas", a document that contains "wild" three times, "purple" five times, and "hyena" 17 times would have a score of 25.
  • 
    

3 Sequence Points

A document receives more points for containing the search terms in the order they were typed. With a higher number of search terms in sequence in the document comes an exponentially higher score.

• Following our example above, for each time the string "wild hyenas" appears in a given document, it is awarded 100 (10^2) additional points. Likewise, for each time "wild purple hyenas" appears in the document, it is awarded an additional 1000 (10^3) points.
• The points awarded are 10^x, where x is the number of search terms appearing in sequence.

[katherinejones]

All,

I did find this information, which implies that the relevance points are higher when a multi-word search includes the words in order (even if other words are between them):

2 Frequency Points

Points are awarded to a document for the frequency (one point per occurrence) in which the search terms appear anywhere in the document.

• For example, in searching for "wild purple hyenas", a document that contains "wild" three times, "purple" five times, and "hyena" 17 times would have a score of 25.

3 Sequence PointsA document receives more points for containing the search terms in the order they were typed. With a higher number of search terms in sequence in the document comes an exponentially higher score.

• Following our example above, for each time the string "wild hyenas" appears in a given document, it is awarded 100 (10^2) additional points. Likewise, for each time "wild purple hyenas" appears in the document, it is awarded an additional 1000 (10^3) points.
• The points awarded are 10^x, where x is the number of search terms appearing in sequence.