From: Barnett, J. <jef...@ya...> - 2008-04-24 19:48:04
|
Andrew, the change to weighted ANDs certainly improves the specificity of the standard query, and may make my librarians happy, but technically there is still tight coupling going on (I think). Ignore what I said about tokenizing. I was imagining an ad hoc work around involving ... $query .= " OR " . $token[i] ... The point is that the current mlt() logic passes the whole $record array to solr and *depends* on the individual words being OR'd to produce a result set. As I have observed, changing the schema default operator (not the index) to AND breaks this dependency. While this might not be desireable for other reasons, I don't think it should break. -----Original Message----- From: Andrew Nagy [mailto:and...@vi...] Sent: Thursday, April 24, 2008 2:27 PM To: Barnett, Jeffrey; vuf...@li... Subject: RE: Tight coupling: tsk, tsk Jeffrey - the changes in 0.8.2 were to the standard search queries (the search query generator in SOLR.php) - not all search queries across the board. With 0.8.2 the searches do a AND'd search as well as an OR's search and weight the AND'd results higher. This has in no way any effect on the "Similar Items" query. If you did local modifications to the index, this is probably a result of this. I highly suggest keeping the default search operator as "OR". The "title" field in the index is tokenized, the "titleStr" field is not. Am I understanding your issues correctly? Andrew > -----Original Message----- > From: vuf...@li... [mailto:vufind- > gen...@li...] On Behalf Of Barnett, Jeffrey > Sent: Thursday, April 24, 2008 1:51 PM > To: vuf...@li... > Subject: [VuFind-General] Tight coupling: tsk, tsk > > (just kidding, but I think this might be an unintended consequesce of > same) > > If you remember the default AND/OR discussion a while back, some > institutions including Yale were/are of the opinion that the broadening > effect of OR defaults, particularly in a large collection could be > distracting. As a result I made the appropriate change to schema.xml > and local librarians we satisfited with the result. Recently however > (can't say if it was release 0.8.1 or .2), the "similar items" listings > started showing no hits, or only other editions of the same title. Are > the bells ringing yet? By using the item title to seed the "more like > this" query, the default behavior of AND is eliminating all but exact > matches. While OR is clearly needed for mlt() to work, I don't think > it is proper to depend on this being the default as well. Could the > title be tokenized and the OR's inserted explicitly? Or is there a > flag that can be set for a single query? > > > ----------------------------------------------------------------------- > -- > This SF.net email is sponsored by the 2008 JavaOne(SM) Conference > Don't miss this year's exciting event. There's still time to save $100. > Use priority code J8TL2D2. > http://ad.doubleclick.net/clk;198757673;13503038;p?http://java.sun.com/ > javaone > _______________________________________________ > VuFind-General mailing list > VuF...@li... > https://lists.sourceforge.net/lists/listinfo/vufind-general |