How can we improve Azure Search?

Support Explain API

The explain api computes a score explanation for a query and a specific document. This can give useful feedback whether a document matches or didn’t match a specific query.

90 votes
Vote
Sign in
Check!
(thinking…)
Reset
or sign in with
  • facebook
  • google
    Password icon
    Signed in as (Sign out)
    You have left! (?) (thinking…)
    Anonymous shared this idea  ·   ·  Flag idea as inappropriate…  ·  Admin →

    4 comments

    Sign in
    Check!
    (thinking…)
    Reset
    or sign in with
    • facebook
    • google
      Password icon
      Signed in as (Sign out)
      Submitting...
      • AdminThe Azure Team on UserVoice (Admin, Microsoft Azure) commented  ·   ·  Flag as inappropriate

        Azure Search does support an analyze API which helps you understand what different analyzers, both out of box or custom, process text during indexing or query time. While not a full Explain API, this does help understand why a search term didn't return the results you expect. Doc link: https://docs.microsoft.com/en-us/rest/api/searchservice/test-analyzer

      • Anonymous commented  ·   ·  Flag as inappropriate

        In the Azure groups today a poster couldn't understand why when he queried his index for the text "@" he returned no results. The reason is that "@" is removed from a phrase no matter where it is found (when using the 2 analysers available in Azure Search.

        He would have saved himself a lot of head scratching if he had the _analyze endpoint

        Examples

        GET _analyze?analyzer=standard
        {this text has @ symbol}

        GET _analyze?analyzer=standard
        {my email address is allan@mydomain.com}

        GET _analyze?analyzer=english
        {my email address is allan@mydomain.com}

      • Harold Perry commented  ·   ·  Flag as inappropriate

        The way search results get ranked and are sorted in the results is often a crucial requirement. Azure search provides substantial flexibility in computing the rank scores but it could be difficult to know if you have them set up appropriately. As such, it would be very helpful to provide extra feedback on request along with query results that contain the details of how each rank score is computed. This could then help you confirm that your scoring profiles (the way rank scores are computed) are behaving appropriately.

        Providing such feedback could also allow a nice smart person (or team) ;-) to produce a "visual scoring profile checker" that e.g., for each ranked result in a query, shows a graphic with the resultant rank score and all of the different contributions to the score so it's easier to eyeball the different results and how their ranks are computed. Such a tool could facilitate the "rank score tuning process" by providing controls to quickly tweak the scoring profiles and test the results.

      Feedback and Knowledge Base