lucene-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Robert Muir (JIRA)" <j...@apache.org>
Subject [jira] [Commented] (LUCENE-1889) FastVectorHighlighter: support for additional queries
Date Mon, 27 Jun 2011 14:31:49 GMT

    [ https://issues.apache.org/jira/browse/LUCENE-1889?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13055580#comment-13055580
] 

Robert Muir commented on LUCENE-1889:
-------------------------------------

Hi Mike, Simon has an issue open to make a lot of what you are talking about wrt positions
easier:
LUCENE-2878

In my opinion once LUCENE-2878 is resolved, we may want to then consider adding the capability
for a codec to encode the offset deltas in parallel with the positions (so its just a stream
of delta-encoded integers you read in parallel with the positions for things like highlighting).


Then, highlighting would not require term vectors anymore right? I think this would be much
faster and more efficient without the space waste of term vectors, and we could prototype
such a thing by encoding these ourselves into the payloads... which is close to the same,
but I think ultimately optionally supporting offsets this way will be better especially with
block-oriented compression algorithms.



> FastVectorHighlighter: support for additional queries
> -----------------------------------------------------
>
>                 Key: LUCENE-1889
>                 URL: https://issues.apache.org/jira/browse/LUCENE-1889
>             Project: Lucene - Java
>          Issue Type: Wish
>          Components: modules/highlighter
>            Reporter: Robert Muir
>            Priority: Minor
>         Attachments: LUCENE-1889.patch
>
>
> I am using fastvectorhighlighter for some strange languages and it is working well! 
> One thing i noticed immediately is that many query types are not highlighted (multitermquery,
multiphrasequery, etc)
> Here is one thing Michael M posted in the original ticket:
> {quote}
> I think a nice [eventual] model would be if we could simply re-run the
> scorer on the single document (using InstantiatedIndex maybe, or
> simply some sort of wrapper on the term vectors which are already a
> mini-inverted-index for a single doc), but extend the scorer API to
> tell us the exact term occurrences that participated in a match (which
> I don't think is exposed today).
> {quote}
> Due to strange requirements I am using something similar to this (but specialized to
our case).
> I am doing strange things like forcing multitermqueries to rewrite into boolean queries
so they will be highlighted,
> and flattening multiphrasequeries into boolean or'ed phrasequeries.
> I do not think these things would be 'fast', but i had a few ideas that might help:
> * looking at contrib/highlighter, you can support FilteredQuery in flatten() by calling
getQuery() right?
> * maybe as a last resort, try Query.extractTerms() ?

--
This message is automatically generated by JIRA.
For more information on JIRA, see: http://www.atlassian.com/software/jira

        

---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@lucene.apache.org
For additional commands, e-mail: dev-help@lucene.apache.org


Mime
View raw message