To add to what Erick said, that *quantifying* is hugely important! How do you measure your search relevance improvements? How are you currently measuring it? How will you see, after you apply any changes, whether relevance was improved and how much? How will you know whether, even test queries you are using to evaluate relevance, the end users also see the same sort of improvements or whether you improved your test queries, but made no difference overall or maybe even made things worse? ...
Have a look at: * http://www.slideshare.net/sematext/tag/analytics * http://sematext.com/search-analytics/index.html - it's free, and we regularly use it with our clients with great success Otis -- Solr & ElasticSearch Support -- http://sematext.com/ Performance Monitoring -- http://sematext.com/spm On Tue, Jul 23, 2013 at 7:50 AM, Erick Erickson <erickerick...@gmail.com> wrote: > Another thing I've seen people do is something like > text:(test AND pdf)^10 text:(test pdf). > > so docs with both terms in the text field get boosted a lot, but docs > with either one will still get found. > > But as Jack says, you have to demonstrate a problem before you propose > a solution. > > You say " a lot people are concerned about improving relevance.". Just > get them to define a good set of search results. Bet they can't except > by looking at specific result lists and saying "I like that one more > than this one". You gotta quantify this somehow, do A/B testing, > whatever or you'll go mad. > > Erick > > On Mon, Jul 22, 2013 at 12:47 PM, Jack Krupansky > <j...@basetechnology.com> wrote: >> Again, you haven't indicated what the problem is. I mean, have you actually >> confirmed that a problem exists? Add debugQuery=true to your query and >> examine the "explain" section if you believe that Solr has improperly >> computed any document scores. >> >> If you simply want to boost a term in a query, use the "^" operator, which >> applies to the preceding term. a boost of 1.0 means no change, 2.0 means >> double, 0.5 means cut in half. >> >> But, you don't need to boost. Relevancy is based on the data in the >> documents themselves. >> >> BTW, q=text%3Atest+pdf does not search for "pdf" in the "text" field - >> field- qualification only applies to a single term, but you can use >> parentheses: q=text%3A(test+pdf) >> >> >> -- Jack Krupansky >> >> -----Original Message----- From: eShard >> Sent: Monday, July 22, 2013 12:34 PM >> To: solr-user@lucene.apache.org >> Subject: Re: how to improve (keyword) relevance? >> >> >> Sure, let's say the user types in test pdf; >> we need the results with all the query words to be near the top of the >> result set. >> the query will look like this: /select?q=text%3Atest+pdf&wt=xml >> >> How do I ensure that the top resultset contains all of the query words? >> How can I boost the first (or second) term when they are both the same field >> (i.e. text)? >> >> Does this make sense? >> >> Please bear with me; I'm still new to the solr query syntax so I don't even >> know if I'm asking the right question. >> >> Thanks, >> >> >> >> -- >> View this message in context: >> http://lucene.472066.n3.nabble.com/how-to-improve-keyword-relevance-tp4079462p4079502.html >> Sent from the Solr - User mailing list archive at Nabble.com.
