You are viewing a plain text version of this content. The canonical link for it is here.
Posted to solr-user@lucene.apache.org by Wei Zhao <we...@astound.ai> on 2018/08/16 19:15:44 UTC

4.10 default ranking scorer, BM25 or classic? How to change that?

Hi,

Does anyone know what the default scorer for 4.10 is? BM25 or classic
tf-idf?

I have been trying to change that, in cloud mode. I have managed to change
the schema.xml in the zookeeper to add the following lines:

 <!-- <similarity class="solr.BM25SimilarityFactory"/> -->
  <similarity
class="org.apache.lucene.search.similarities.LMDirichletSimilarity"/>

The commented line was also tried. So I have tried different syntax, using
factory without identifying additional parameters or using default,
BM25 or LMDirichlet
similarity. However, no matter how I did that, the final matching score
never changed.

I'm using standard query with lots of boosting such as title:apple^4.

Please let me know what I missed to change the scorer. Thank you.

Re: 4.10 default ranking scorer, BM25 or classic? How to change that?

Posted by Erick Erickson <er...@gmail.com>.
Does debug=true show you that the scores are, indeed, computed exactly the same?

It could simply be that how you index doesn't contain the data you
need to see a difference. If you're searching on string fields for
instance, there isn't much "there there".

What I'd do is work from a stand-alone instance first before worrying
about Cloud, ensure that you can see the differences then move to
Cloud, just to remove one of the variables.

Best,
Erick

On Thu, Aug 16, 2018 at 12:15 PM, Wei Zhao <we...@astound.ai> wrote:
> Hi,
>
> Does anyone know what the default scorer for 4.10 is? BM25 or classic
> tf-idf?
>
> I have been trying to change that, in cloud mode. I have managed to change
> the schema.xml in the zookeeper to add the following lines:
>
>  <!-- <similarity class="solr.BM25SimilarityFactory"/> -->
>   <similarity
> class="org.apache.lucene.search.similarities.LMDirichletSimilarity"/>
>
> The commented line was also tried. So I have tried different syntax, using
> factory without identifying additional parameters or using default,
> BM25 or LMDirichlet
> similarity. However, no matter how I did that, the final matching score
> never changed.
>
> I'm using standard query with lots of boosting such as title:apple^4.
>
> Please let me know what I missed to change the scorer. Thank you.