[jira] [Commented] (TIKA-2108) Non-semantic extraction of supplemental categories

["Tim Allison (JIRA)" <ji...@apache.org>]

    [ https://issues.apache.org/jira/browse/TIKA-2108?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15542418#comment-15542418 ] 

Tim Allison commented on TIKA-2108:
-----------------------------------

Thank you for opening this.  Is it safe to split on comma/semi-colon for every value?  This gives me pause.

Should we limit it to metadata values with cardinality {{0..>1}} in the [spec|http://www.iptc.org/std/photometadata/specification/IPTC-PhotoMetadata], e.g. key words, phone numbers, email addresses, urls, etc.?

[~rgauss], any recommendations?

> Non-semantic extraction of supplemental categories
> --------------------------------------------------
>
>                 Key: TIKA-2108
>                 URL: https://issues.apache.org/jira/browse/TIKA-2108
>             Project: Tika
>          Issue Type: Bug
>         Environment: tika-app-1.13.jar, tika-server-1.13.jar
>            Reporter: Xavier Perseguers
>            Priority: Minor
>
> When extracting metadata for a file with categories, the comma (or semi-colon - according to the IPTC specification) -separated list of categories is extracted as a blank space-separated list of terms.
> Example with https://dl.dropboxusercontent.com/u/3177102/sample.jpg
> {code}
> java -jar tika-app-1.13.jar --xml sample.jpg | grep Category
> <meta name="Supplemental Category(s)" content="Fribourg Cathédrale Pont de la Poya"/>
> {code}
> When using exiftool:
> {code}
> exiftool sample.jpg  | grep Categories
> Supplemental Categories         : Fribourg, Cathédrale, Pont de la Poya
> {code}
> This is not a problem when using Tika with, say, Solr since stopwords such as "de" and "la" will be dropped. However, when using Tika standalone with an external tool, there is no way to fetch the *actual* list of categories, namely:
> * Fribourg
> * Cathédrale
> * Pont de la Poya
> where blank spaces in the title may be wanted instead of getting single-term topic names.



--
This message was sent by Atlassian JIRA
(v6.3.4#6332)