You are viewing a plain text version of this content. The canonical link for it is here.
Posted to dev@lucene.apache.org by "Fengtan (JIRA)" <ji...@apache.org> on 2015/10/05 04:57:26 UTC

[jira] [Created] (SOLR-8124) Adjust relevance based on user click/feedback

Fengtan created SOLR-8124:
-----------------------------

             Summary: Adjust relevance based on user click/feedback
                 Key: SOLR-8124
                 URL: https://issues.apache.org/jira/browse/SOLR-8124
             Project: Solr
          Issue Type: New Feature
          Components: search
            Reporter: Fengtan


It may be interesting to be able to bias the score of documents based on user feedback. For instance, if query q=solr+rocks returns 2 documents A and B, and if most people click on document B, then document B could be given a higher score given that it seems to satisfy most people.

A solution would be to have a field 'popularity' containing the number of times people clicked on that document. Search queries would then boost documents with a high 'popularity' (as described in [this presentation|http://www.slideshare.net/lucenerevolution/potter-timothy-boosting-documents-in-solr]).

The problem with the above is that the document's popularity would be global. Ideally, the document's popularity would depend on the search query (possibly parameters q and fq). For instance a document titled 'solr cookbook' might be popular for a query q=solr+rocks but not for q=cook+cake.

Hence a possible solution would be to maintain some kind of map between "document ID + query" and "popularity". The popularity could be exposed in a field so search queries could use it in boost clauses. I am guessing we would need to create a new SearchComponent.

Other search engines seem to support this kind of feature, for instance the Google Search Appliance provides a [/click protocol|http://www.google.com/support/enterprise/static/gsa/docs/admin/70/gsa_doc_set/xml_reference/advanced_search_reporting.html] and LucidWorks provides a [Click Scoring Relevance Framework|https://docs.lucidworks.com/display/lweug/Click+Scoring+Relevance+Framework].

Does this sound worthy ? I could not find any other ticket dealing with this so opening this one to check people's opinion.



--
This message was sent by Atlassian JIRA
(v6.3.4#6332)

---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@lucene.apache.org
For additional commands, e-mail: dev-help@lucene.apache.org