[
https://issues.apache.org/jira/browse/NUTCH-1075?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13085826#comment-13085826
]
Julien Nioche commented on NUTCH-1075:
--------------------------------------
Markus - the issue you're having with -D lang.extraction.policy=identify is due
to the score obtained for nl is higher than the threshold for
identifier.isReasonnablyCertain which is hardcoded in Tika.
We could make so that this test is ignored when the policy is identify only or
add a param for it. I haven't found an example where the value returned was
deemed reasonnably certain. WDYT?
> Delegate language identification to Tika
> ----------------------------------------
>
> Key: NUTCH-1075
> URL: https://issues.apache.org/jira/browse/NUTCH-1075
> Project: Nutch
> Issue Type: Improvement
> Components: parser
> Affects Versions: 1.4
> Reporter: Julien Nioche
> Assignee: Julien Nioche
> Fix For: 1.4
>
> Attachments: NUTCH-1075.patch
>
>
> In 2.0 the language identification is delegated to Tika and is done as part
> of the parsing step (and not during the indexing as done currently).
> The patch attached is a backport from trunk which implements this and adds a
> new parameter to determine the strategy to use
> {code:xml}
> <property>
> <name>lang.extraction.policy</name>
> <value>detect,identify</value>
> <description>This determines when the plugin uses detection and
> statistical identification mechanisms. The order in which the
> detect and identify are written will determine the extraction
> policy. Default case (detect,identify) means the plugin will
> first try to extract language info from page headers and metadata,
> if this is not successful it will try using tika language
> identification. Possible values are:
> detect
> identify
> detect,identify
> identify,detect
> </description>
> </property>
> {code}
--
This message is automatically generated by JIRA.
For more information on JIRA, see: http://www.atlassian.com/software/jira
