[
https://issues.apache.org/jira/browse/LUCENE-3440?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13114291#comment-13114291
]
sebastian L. commented on LUCENE-3440:
--------------------------------------
bq. Patch looks great!
Thanks.
bq. 1. For the new totalWeight, add getter method and modify toString() in
WeightedFragInfo().
Okay.
bq. 2. The patch uses hard-coded DefaultSimilarity to calculate idf. I don't
think that a custom similarity can be used here, too. If so, how about just
copying idf method rather than creating a similarity object?
I played a little with log(numDocs - docFreq + 0.5 / docFreq + 0.5) but is
seems to make no difference. If I'm not mistaken there is no method
IndexReader.getSimilarity() or IndexReader.getDefaultSimilarity().
Therefore: Okay.
bq. 3. Please do not hesitate to update ScoreComparator (do not add
WeightOrderFragmentsBuilder)
Hm, I thought about something like that:
{code:xml}
<highlighting>
<fragmentsBuilder name="ordered"
class="org.apache.solr.highlight.ScoreOrderFragmentsBuilder" default="false"/>
<fragmentsBuilder name="weighted"
class="org.apache.solr.highlight.WeightOrderFragmentsBuilder" default="true"/>
</highlighting>
{code}
For Solr-users (like me). If somebody would like to use the boost-based
ordering, he could. Maybe, for some use-cases the boost-based approach is
better than the weighted one.
bq. 4 Could you update package javadoc (
https://builds.apache.org//job/Lucene-trunk/javadoc/contrib-highlighter/org/apache/lucene/search/vectorhighlight/package-summary.html#package_description
) and insert totalWeight into description and figures.
Okay.
bq. 5. use docFreq(String field, BytesRef term) version for trunk to avoid
creating Term object.
Okay.
bq. I agree. I think if there is a table so that we can compare totalBoost
(current) and totalWeight (patch) with real values, it helps a lot.
I'll write some Proof-of-concept Test-Class. But this may take some time.
I discovered a little problem with overlapping terms, depending on the
analyzing-process:
WeightedPhraseInfo.addIfNoOverlap() dumps the second part of hyphenated words
(for example: social-economics). The result is that all informations in
TermInfo are lost and not available for computing the fragments weight. I
simple modified WeightedPhraseInfo.addIfNoOverlap() a little to change this
behavior:
{code:java}
void addIfNoOverlap( WeightedPhraseInfo wpi ){
for( WeightedPhraseInfo existWpi : phraseList ){
if( existWpi.isOffsetOverlap( wpi ) ) {
existWpi.termInfos.addAll( wpi.termInfos );
return;
}
}
phraseList.add( wpi );
}
{code}
But I am not sure if there could be some unforeseen site-effects?
> FastVectorHighlighter: IDF-weighted terms for ordered fragments
> ----------------------------------------------------------------
>
> Key: LUCENE-3440
> URL: https://issues.apache.org/jira/browse/LUCENE-3440
> Project: Lucene - Java
> Issue Type: Improvement
> Components: modules/highlighter
> Affects Versions: 3.5, 4.0
> Reporter: sebastian L.
> Priority: Minor
> Labels: FastVectorHighlighter
> Fix For: 3.5, 4.0
>
> Attachments: LUCENE-3.5-SNAPSHOT-3440-3.patch,
> LUCENE-4.0-SNAPSHOT-3440-3.patch
>
>
> The FastVectorHighlighter uses for every term found in a fragment an equal
> weight, which causes a higher ranking for fragments with a high number of
> words or, in the worst case, a high number of very common words than
> fragments that contains *all* of the terms used in the original query.
> This patch provides ordered fragments with IDF-weighted terms:
> total weight = total weight + IDF for unique term per fragment * boost of
> query;
> The ranking-formula should be the same, or at least similar, to that one used
> in org.apache.lucene.search.highlight.QueryTermScorer.
> The patch is simple, but it works for us.
> Some ideas:
> - A better approach would be moving the whole fragments-scoring into a
> separate class.
> - Switch scoring via parameter
> - Exact phrases should be given a even better score, regardless if a
> phrase-query was executed or not
> - edismax/dismax-parameters pf, ps and pf^boost should be observed and
> corresponding fragments should be ranked higher
--
This message is automatically generated by JIRA.
For more information on JIRA, see: http://www.atlassian.com/software/jira
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
