From: Demian K. <dem...@vi...> - 2010-03-23 13:48:24
|
Hello, One of the librarians here noticed that the "morelikethis" handler in VuFind was coming up with some questionable results. I did some analysis and discovered a fairly significant flaw in the default configuration -- although we assign a very high weight to the callnumber-label field, we accept Solr's default term frequency value of "2," which ignores any terms that appear only once in a record. Since the value of callnumber-label appears only once in most records, call number is effectively ignored, with most similar item suggestions being based on matches in other, less reliable fields. There's a simple fix -- set term frequency to 1 by adding this line to the morelikethis requestHandler section of solr/biblio/conf/solrconfig.xml: <int name="mlt.mintf">1</int> I think it probably makes sense to make this the default configuration in the trunk, but I thought I would give everyone a chance to try it out and comment first. Any thoughts/objections? Semi-related: This also reminded me that by default, Dewey call numbers are not factored into the morelikethis query. I have added some notes to the Dewey configuration wiki page (http://vufind.org/wiki/call_numbers) to account for this. thanks, Demian |