Automatische Inhaltsbersicht fr XMLKollektionen

1
Automatische Inhaltsübersicht für XML-Kollektionen

• Diploma thesis introduction

2
Goals

• To obtain custom summaries of specific XML
  collections.
• Develop or extend methods for XML summarization.
• Provide a prototype implementation for such XML
  summarization methods.

3
Structure

• Theoretical part
• Summarization
• XML Summarization
• Template method.
• Datatypes
• Summarization modules.
• Practical part
• A summarization tool.

4
Summarization 6

• The process of distilling the most important
  information from a source (or sources) to produce
  an abridged version for a particular user (or
  users) and task (or tasks).
• Extract vs. Abstract.
• Summary Function Indicative, Informative and
  Evaluative summaries.
• Approaches Shallower and Deeper approaches.

5
XML Summarization

• Previous work
• SqueezeX compressor 01, which does not directly
  imply XML summarization, but a lossless and lossy
  compressor.
• XML summary based on statistics of queries 03.
• Text Summarization with text tagged as XML 02a
  02b.
• All previous work cannot handle documents with
  specific characteristics.
• Each XML collection has its own characteristics
• Template method allows characterization.

6
Template method

• Describes a pseudo-structure of original
  document.
• Such structure may include explicit data type
  information.
• Adapts different summarization efforts to
  specific parts of an XML document through
  summarization modules.
• Semi-automatic way because of the need to design
  a template when looking for an optimal XML
  summarization of a characteristic XML collection.
  So, an explicit design handles specific
  characteristics.

7
INPUT
Bibliography
entry
entry
entry
entry
entry
titleb2
titlek3
title123
titleDB
titlek1
isbnxx-yy
isbnx5-6
isbnx2-y3
isbn7891
isbnxx-yy
autorXXX
autorJTK
autorXXX
autorFühr
autorJTK
date7/3/1971
date1/3/1987
date1/1/1971
date1/1/2004
date1/1/1999
Bibliography
OUTPUT
Bibliography
TEMPLATE
entry
entry
entry
entry
1)author ? merge with other entry siblings 2)date
? interval of year 3)else ? remove
autorJTK
autorXXX
autorFühr
year1987-1999
year1971
year2004
8
INPUT
a
b
b
c3
dt1
c9
dt2
TEMPLATE
OUTPUT
a
a
b
b
c12
dt3
1) any integer of b ? aggregation sum
function 2) any string of b ? merge all ?
text summarize
9
Template workflow

• Template design process
• Summarization rules creation (if needed).
• Data types creation (if needed).
• Template design. (reuses rules and data types)
• Template validation.
• Template project file saved.
• XML collection matching to existent templates.
• With optimal template found, summarize according
  to template rules and definitions.

10
Template Matching
Look for a matching Template
Analize nextDocument
Matched Documents
Summarize
Analize nextDocument
Match!
No match.
11
Data types

• Abstraction levels for data organization.
• In 102 three levels of data organization are
  distinguished the physical level, conceptual
  level and external level.
• Framework for dealing with data types in the XML
  context.
• Efforts towards an XSD data type framework.

12
Data type hierarchy example
D-
string
integer
date
text
name
journal title
English
German
person-name
inst-name
English JT
German JT
D-
13
Summarization Modules

• Summarization Module is a concept to extend and
  implement future summarization methods. By
  default there will be some summarization modules
  available. It is integrated by two parts
• An XSD complex type definition.
• An implemented AbstractSummarizationModule.
• The AbstractSummarizationModule is an abstract
  class that provides an interface for implementing
  specific summarization methods. A module may take
  a subtree and give a summarized subtree as an
  output.
• Examples
• A module that implements text summarization using
  segmentation, lexical chains and aid by a
  thesaurus like WordNet.
• A module that reuses an already created text
  summarizer Open Text Summarizer or
  Classifier4J.

14
A Summarization Tool

• Main implementation consists of two parts
• XML Summarization-Template Designer Suite
  (XSTDS), where the template is designed.
• AX Summarizer (Automatic XML Summarizer or AXS),
  where there is an automatic workflow of
  summarization.
• XSTDS has three main components
• Rules Creator
• Data type Designer
• Template Designer

15
Template relation
16
General Steps for User (I)

• User wants to summarize an XML collection with
  certain characteristics, these characteristics
  may be reflected in a DTD or XSD file. So in
  general, characteristics of a collection will aid
  in the creation of a template.
• The XSTDS program is loaded. A new project is
  created and a template editing GUI appears. It
  has the resemblance of a right-sided tree.
• (A wizard/assistant could be in future
  development)
• User has the option of creating at first a draft
  template from one or many XML samples. Under many
  samples there will be a merge of duplicate nodes
  under a same root and will try to determine data
  types and assign default behavior. With one
  sample merge still applies for duplicated nodes.
• User can add or remove nodes. It can apply
  summarization rules and assign data types and
  modify default behavior. There is a pool of
  previously defined summarization rules (with
  available summarization modules) and data types.
  The user has access to both pools that are
  categorized by the project where they were
  created or date (or any attribute that may be a
  useful indicator). The summarization modules are
  hard-coded (but maybe it will possible to have
  dynamic summarization module loading).

17
General Steps for User (II)

• User could specify a pattern of the input XML
  filenames as help to the future template matching
  process.
• When the user finishes the design it validates
  the template for consistency.
• User saves the project and closes the XSTDS
  application.
• User loads the AX Summarizer application. The AXS
  knows of the available templates (projects?)
  because of the default template folder.
• The AXS shows a GUI where the user can specify
  the source/input (a file, a folder or a recursive
  folder), the output folder and the template
  options. Under template options the user can set
  default matching or specify a specific
  template. It may be possible to customize certain
  options like very general output characteristics.
• User could save settings.
• The user can start now to summarize by pressing
  the Summarize button. It should be possible to
  cancel the process.
• A sound and a popup window indicates that there
  was success or failure, and a under failure there
  will be a informative log to assist the user of
  the problems that were encounter. Fatal and
  simple errors can be noticed. Some statistics of
  the whole process is shown.

18
The End

• Questions and discussions.

19
(No Transcript)
20
(No Transcript)
21
Bibliography

• 01 Cannataro, Comito, Pugliese (2002).
  SqueezeX Synthesis and Compression of XML Data.
  Proceedings of the International Conference on
  Information Technology Coding and Computing
  (ITCC02)
• 02a Kenneth C. Litkowski (2003). Text
  Summarization Using XML-Tagged Documents. CL
  Research DUC 2003
• 02b Kenneth C. Litkowski (2004). Summarization
  Experiments in DUC 2004. CL Research DUC 2004
• 03 Comai, Marrara and Tanca (2004). XML
  Document Summarization Using XQuery for Synopsis
  Creation. Proceedings of the 15th International
  Workshop on Database and Expert Systems
  Applications (DEXA04)
• 06 T. Maybury, I. Mani (2001). Automatic
  Summarization.
• American/European Conference on Computational
  Linguistics (ACL/EACL 2001) Toulouse, France 8
  July 2001
• 102 N. Fuhr (1999). Towards Data Abstraction
  in Networked Information Retrieval Systems.
  Information Processing and Management 35(2)