3 Replies Latest reply: Dec 16, 2010 11:52 AM by Aaron Johnson RSS

    Search results and ranking

    Frank Gebhardt

      Hi,

       

      I constantly get asked how the ranking of search results works. I can point people to the Lucene documentation as per this discussion. Although, I like a few more specific answers. For example

       

      • The pop up box starts displaying search results after you typed 3 characters. It refines it when completing the search term. There is a result difference when you add a "space" at the end of the search word. See image for searching Jive for "search" and "search "

      search_popup.jpgsearch_full.jpg

       

      • When you use the full search results the system provides you with sorting options (relevance, date, etc) which you don't have in pop up box. I don't know what sorting algorithm is used in the pop up. Also, the results usually differ. See image for a full Jive search using "search" above.

       

      • When another person is using the same search pattern, the results are different. I assume access security and group membership play a significant part in sorting and offering the results. I don't know if the profile information (e.g. tags) are also used for refining search results.

       

      I appreciate if somebody has got some explanations

       

      cheers

      Frank

        • Re: Search results and ranking
          Aaron Johnson

          Hi Frank,

           

          Thanks for the well illustrated / explained question.

          The pop up box starts displaying search results after you typed 3 characters. It refines it when completing the search term. There is a result difference when you add a "space" at the end of the search word. See image for searching Jive for "search" and "search "

          Yep, subtle thing that we should probably fix long term. As an example, let's say I do one search for "police*" and one search for "police *"  (this is a real example from our own internal instance by the way).

           

          The query parser most likely uses the stemming analyzer to reduce the first query to be something like this:

          find me any and all content in the system that contains the string 'polic'

          and the second turns into this:

          find me any and all content in the system that contains the string 'polic' followed by a space

          The first one is going to match or get hits on things like this:

           

          police, policy, police_sting, police-only, policy.ls, http://policyresearch.com, policyclassloader.java

           

          and the second will only match on:

           

          policy, policy

           

          One thing you can do to help visualize the differences is to actually do a real search in Jive using the wildcard character (ie: "*"), do one with the added space and one without the space and on the results page you'll be able to see the different things that get hits with the yellow highlight.

           

          When you use the full search results the system provides you with sorting options (relevance, date, etc) which you don't have in pop up box. I don't know what sorting algorithm is used in the pop up. Also, the results usually differ. See image for a full Jive search using "search" above.

          The spotlight search uses the relevance sort.

           

          When another person is using the same search pattern, the results are different. I assume access security and group membership play a significant part in sorting and offering the results. I don't know if the profile information (e.g. tags) are also used for refining search results.

          We do not use profile information for refining your content search results, the differences are 100% about permissions and group membership.

           

          I hope that helps.  I'd be happy to chat with you and your team on the phone sometime if you think it would help.

           

          Thanks again for the great question.

           

          Cheers,

           

          AJ