lucene-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Andrew Duffy (JIRA)" <>
Subject [jira] Updated: (LUCENE-1389) SimpleSpanFragmenter can create very short fragments
Date Wed, 17 Sep 2008 13:03:44 GMT


Andrew Duffy updated LUCENE-1389:

    Attachment: tailfragments.patch

Another problem with the simple fragmenters is that they can produce a very short fragment
at the end of the stream. The attached diff from the trunk remembers
how long the text being fragmented is and doesn't make a fragment break if the token is less
than half a fragment away from the end of the text. SimpleFragmenter could easily be changed
in the same way.

> SimpleSpanFragmenter can create very short fragments
> ----------------------------------------------------
>                 Key: LUCENE-1389
>                 URL:
>             Project: Lucene - Java
>          Issue Type: Bug
>          Components: contrib/highlighter
>    Affects Versions: 2.3.2
>            Reporter: Andrew Duffy
>            Priority: Minor
>         Attachments: tailfragments.patch
> Line 74 of SimpleSpanFragmenter returns true when the current token is the start of a
hit on a span or phrase, thus starting a new fragment. Two problems occur:
> - The previous fragment may be very short, but if it contains a hit it will be combined
with the new fragment later so this disappears.
> - If the token is close to a natural fragment boundary the new fragment will end up very
short; possibly even as short as just the span or phrase itself. This is the result of creating
a new fragment without incrementing currentNumFrags.
> To fix, remove or comment out line 74. The result is that fragments average to the fragment
size unless a span or phrase hit is towards the end of the fragment - that fragment is made
larger and the following fragment shorter to accommodate the hit.

This message is automatically generated by JIRA.
You can reply to this email to add a comment to the issue online.

To unsubscribe, e-mail:
For additional commands, e-mail:

View raw message