Hi, just a background on my setup. I'm crawling with Nutch-1.2, I used Solr-1.4 and Solr-3.5, with the same result. Solr is still using the default settings.
I found this problem just by accident. I queried "mobile broadband", page A, has 2 occurences and scores higher than page B that has 19 occurences. I found it weird and that's why I started investigating. The debug results are given below and you can see that queryWeight, idf and queryNorm are the same, tf is higher, as expected, in B but what makes the difference is clearly fieldNorm. A: 0.010779975 = (MATCH) weight(content:"mobil broadband" in 18730), product of: 1.0 = queryWeight(content:"mobil broadband"), product of: 6.2444286 = idf(content: mobil=4922 broadband=2290) 0.16014275 = queryNorm 0.010779975 = fieldWeight(content:"mobil broadband" in 18730), product of: 1.4142135 = tf(phraseFreq=2.0) 6.2444286 = idf(content: mobil=4922 broadband=2290) 0.0012207031 = fieldNorm(field=content, doc=18730) B: 8.5223187E-4 = (MATCH) weight(content:"mobil broadband" in 14391), product of: 1.0 = queryWeight(content:"mobil broadband"), product of: 6.2444286 = idf(content: mobil=4922 broadband=2290) 0.16014275 = queryNorm 8.5223187E-4 = fieldWeight(content:"mobil broadband" in 14391), product of: 4.472136 = tf(phraseFreq=20.0) 6.2444286 = idf(content: mobil=4922 broadband=2290) 3.0517578E-5 = fieldNorm(field=content, doc=14391) Remi On Wed, Jan 18, 2012 at 8:52 PM, Jan Høydahl <jan....@cominvent.com> wrote: > > I've come accros a problem where newly indexed pages almost always come > > first even when the term frequency is relatively slow. > > There is no inherent index-time boost, so this must be something else. > Can you give us an example of a query? Which query parser do you use? > > > I read the posts below on "fieldNorm" and "omitNorms" but setting > > "omitNorms=true" doesn't change anything for me on the calculation of > > fieldNorm. > > Are you sure you have spelled omitNorms="true" correctly, then restarted > Solr (to refresh config)? The effect of Norms on your score will be that > shorter fields score higher than long fields. > > Perhaps you instead can try to tell us your use-case. What kind of raning > are you trying to achieve? Then we can help suggest how to get there. > > -- > Jan Høydahl, search solution architect > Cominvent AS - www.cominvent.com > Solr Training - www.solrtraining.com
