Section 4.2.1 from the reference manual discusses field boosting and states the following:

'The text field will be 1.2 times more important than the isbn field'



Lucene says the following in their FAQ about index time boosting:


Not 100% sure how field boosts are implemented in Hibernate Search, but after a quick look at the source code I think the field boost of a document are set using a LuceneOptions object, which probably does the same as field.setBoost(...).

In that case, it seems kinda useless to have a @Boost annotation on a field, based on the Lucene FAQ. The boost will then only be effective when querying mutliple types of entity classes.

Or can a field boost still be effective to rank a document with [title:'foo' desc:'bar'] higher than a document with [title:'bar' desc:'foo'] when searching for 'foo' on all search fields with 'title' being index-time boosted to 10f. In that case, the Lucene FAQ seems wrong.

Hi,

I think the comment on the Lucene wiki is meant for the case where you boost every single field AND only search on this single field. In this case the boost is of course useless, because all fields have the same boost.

However, if you search across multiple fields as in the documentation or your example boosting has an effect. And there is of course also the use case where you search several entity types which have the same field. In this case you can get different results even if you search just on this field.

If you need a more flexible type of index time boosting you should look at @DynamicBoost feature. There you can assign dynamic boost values for each field value.

All this is of course theory to a certain degree. I recommend to write a little unit test and experiment yourself. You can use the explain functionality to inspect why certain documents matched and which score they received.

Hope this helps.

--Hardy

As suggested, I just tested the field boost, and it is indeed effective, so the Lucene FAQ seems wrong.

For testing I used an existing testproject with 1.3M+ publications (sry, don't have time to create a reusable unit test).
Each publication instance has a title and description field. The description is an ordinary non-boosted field, title has a field boost of 1000f:



I created 2 docs with unique terms in it:
DocA[title="barbar", description="foofoo and foofoo, but also barbar"]
DocB[title="foofoo and foofoo, but also barbar", description="barbar"]

Searching for publications for query:
'title:foofoo description:foofoo'
results in this ranked list:

• DocB[title="foofoo and foofoo, but also barbar", description="barbar"]
  
  Code:
  2511.4648 = (MATCH) sum of:
  
      2511.4648 = (MATCH) weight(title:foofoo in 0), product of:
          0.32124138 = queryWeight(title:foofoo), product of:
              14.396252 = idf(docFreq=1, maxDocs=1315068)
              0.022314237 = queryNorm
          7817.999 = (MATCH) fieldWeight(title:foofoo in 0), product of:
              1.4142135 = tf(termFreq(title:foofoo)=2)
              14.396252 = idf(docFreq=1, maxDocs=1315068)
              384.0 = fieldNorm(field=title, doc=0)
• DocA[title="barbar", description="foofoo and foofoo, but also barbar"]
  
  Code:
  3.2701366 = (MATCH) sum of:
  
      3.2701366 = (MATCH) weight(description:foofoo in 1), product of:
          0.32124138 = queryWeight(description:foofoo), product of:
              14.396252 = idf(docFreq=1, maxDocs=1315068)
              0.022314237 = queryNorm
          10.179687 = (MATCH) fieldWeight(description:foofoo in 1), product of:
              1.4142135 = tf(termFreq(description:foofoo)=2)
              14.396252 = idf(docFreq=1, maxDocs=1315068)
              0.5 = fieldNorm(field=description, doc=1)

I first tested with a field boost of 10f, which wasn't enough to push DocB to the top. With the higher boost of 1000f it's clear that boosting a field for all docs ísn't useless. I'll pook Lucene to update their FAQ.

Thanks for the update :-)