Re: [xml-dev] Structured from/within unstructured documents

["Edward C. Zimmermann" <edz@bsn.com>]

On Sat, 15 Dec 2007 18:04:41 +0000, Stephen Green wrote
> What methods are there, these days, for extracting structured data from
> unstructured documents (such as PDF)?

PDF documents are typically not really "unstructured". There is the metadata
(Title, author, creator etc.) and the pages as graphically intended to be
viewed.

Unless the contents are purely image data (scanned in) one can just render it
to text (same applies to MS Word, Postscript and a host of other formats). The
text has typically some intended structure. The rest is the old art of
auto-tagging.

In indexing PDF documents we routinely extract basic structure such as
sentence, paragraph and page-- in PDF we don't bother with line. This is
generic enough that we found it can nearly always be applied. For the intent
of search and retrieval we've indexed many millions of millions of PDF
documents over the years. The metadata we've found typically of very
limited use (almost always not correctly maintained). Sentence, paragraph
and page identification is, by contrast, given the conventions of writing.
very reliable.

For more complicated and detailed structure its less generic and one needs
to define some regular expression models and rules to define what makes for
whatever structural element one is looking for.

Typically, however, for PDF in the applications were a finer structure is
needed there are other, alternative, formats of the information available.

One can, of course, using other methods even break things down into
linguistic and grammatical elements as well as determine things like place
names, person names, inter-relationships etc.: the whole bag of textual
analysis.



--

Edward C. Zimmermann, Basis Systeme netzwerk, Munich
Office Leo (R&D):
   Leopoldstrasse 53-55, D-80802 Munich,
http://www.nonmonotonic.net