[
https://issues.apache.org/jira/browse/TIKA-3711?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17516141#comment-17516141
]
Sam Stephens commented on TIKA-3711:
------------------------------------
I guess the question is what are the semantics of this operation? When I ask
for the text of a document, what does that actually mean?
As an end user, I'd argue the semantics that are most useful to end users is
that getting the text of a document provides the closest possible
representation of the text a user would read when reading the document in its
native format.
By this argument, the image filenames should not be there, because I wouldn't
see image filenames if I was reading the Word document from within Word.
I don't think more information for its own sake is necessarily good. If I argue
this from a reductio ad absurdum perspective, I'd then say that adding text
describing all document formatting is useful. Adding the words "Heading 1" each
time there's a heading, "Bold" and "Unbold" each time a bolded section occurs.
This is clearly more information, but it's also clear that adding this
information would rapidly make the text you extracted from a Word document
unusable.
>From an end user perspective, I'm using this text extraction so I can put
>documents in a search index. Having the terms "image1", "image2" etc show up
>in my index for documents that contain images is not useful behavior, unless
>that actually occurs in the real text of the document.
The image filenames are metadata. If I wanted that metadata, I can engage with
the full XHTML representation of the document to get it. But my take is that
BodyContentHandler should give me text, not metadata.
> Image file names included in parsed Word Document text
> ------------------------------------------------------
>
> Key: TIKA-3711
> URL: https://issues.apache.org/jira/browse/TIKA-3711
> Project: Tika
> Issue Type: Improvement
> Components: parser
> Affects Versions: 2.3.0
> Reporter: Sam Stephens
> Priority: Minor
> Attachments: word-doc-with-image.docx
>
>
> The attached Word document includes nothing but a single image. Running it
> through the Tika 2.2.0 AutoDetectParser correctly returns null. Running it
> through the Tika 2.3.0 AutoDetectParser returns the text:
> {{image1.png}}
>
--
This message was sent by Atlassian Jira
(v8.20.1#820001)
