Displaying search results for correlated search terms

[stubarny]

Hello,

 

I'm developing a website that asks the user to submit a keyword for a search. The results should display matches for this keyword, but also show matches for related keywords (in order of relevenace!).

 

I'm planning on building up a library of which search terms users use in the same sessions (e.g. if someone searches for "it jobs" and "php jobs", I'll know the terms are correlated), and I'll also measure the click-through rates of the items on the results list.

 

I've been spending all weekend trying to map out the design for this but it's proving incredibly complicated and I think the solution is likely to be on the internet somewhere already?!

 

Please could someone point me in the right direction if you've come accross this problem before?

 

Thanks a million,

 

Stu

[WebStyles]

I'm planning on building up a library of which search terms users use in the same sessions (e.g. if someone searches for "it jobs" and "php jobs", I'll know the terms are correlated), and I'll also measure the click-through rates of the items on the results list.

 

hmmm... you planning on creating a list of relevant correlations based on what the users input?... Imagine this scenario:

 

a user interested in animals, types in 'Dogs', then types in 'Cats', then 'Horses'... And you store these as being related to each other. The next day, another user ONLY interested in dogs, types in 'dogs' and also gets cats & horses in their search result... I don't see how that can work without you manually filtering and deciding what's related or not, or checking against hundreds of similar entries before creating the relationship between words.

[stubarny]

I don't see how that can work without you manually filtering and deciding what's related or not, or checking against hundreds of similar entries before creating the relationship between words.

 

Yep that's the plan - manual over-rides in the short-term, then move over to the automated system. I kind of thought this would be a common problem?

[WebStyles]

ok, and is there a limited number of words that can be correlated?

like it the main word is 'color', how far are you willing to go in your search results, since you can add literally hundreds of colors (consider rgb codes, cmyk codes, pantone codes, html codes, etc...)

 

I ask this because I can imagine it being done differently if there was a limit than if there is no limit of words that can relate to the first.

 

(And how would 'relevance' come into play in this situation? don't they ALL have the same relevance to the keyword 'color' ? How are you planning or categorizing the relevance? in this case would you just pick 30 random results to show?)