EACL'03 Tutorial on Text Representation for Automatic Text Categorization

José María Gómez Hidalgo

Departamento de Inteligencia Artificial (Artificial Intelligence Departament)

Universidad Europea de Madrid

Summary  - proposal (PDF, ZIP)
Slides of the tutorial (PDF 4Mb, ZIP 1,94Mb)
References (BIB, online searchable bibliography)

Outline

The goal of this tutorial is making the audience familiar to the different ways that text is represented in Automated Text Categorization (ATC). I define and describe applications of Text Categorization, and present the general model for learning based ATC. Then I describe a number of works in which proposals for  text representation have been presented, including the usage of statistical and linguistic phrases, Information Extraction patterns, and WordNet information. Also specific application text features are presented, based on stylometry and structural text properties, for several tasks including author, language and genre identification, spam detection, etc.

The tutorial is mainly divided into two parts: an overview of ATC, and the discussion of specific text representations. In particular, the tutorial covers the following topics:

1. Automated Text Categorization (ATC)

A definition of (Automated) Text Categorization is presented, along a number of considerations. Some relevant links to this part are:

• The Bibliography on Automated Text Categorization by Fabrizio Sebastiani.
• The "Untangling Text Data Mining" paper by Marti Hearst.
• The home page of David D. Lewis.

2. Applications

A number of applications are brieftly described in this tutorial. Some interesting sites include:

• The Yahoo! Planet project page by Dunja Mladenic and others.
• The SwiftFile project at IBM Research.
• The European Commision POESIA (Public Opensource Environment for a Safer Internet Access) project.
• The TextCat language guesser demo by Gertjan van Noord.
• My web page and bibliography on machine learning for spam detection.
• The Snowfox automatic web directory.
• The "Practical Issues for Automated Categorization of Web Sites" paper by John M. Pierre.

3. A blueprint for learning-based ATC

Here I present the learning-based model for ATC. Some interesting links include:

• The WEKA (Waikato Environment for Knowledge Analysis) project and software at the University of Waikato.
• The Foundations of Statistical Natural Language Processing list of resources for ATC by Chris Manning and Hinrich Schütze.

4. Advanced document indexing

Some links on the topics covered in this part:

• The home page of Ellen Riloff.
• The lexical database Wordnet developed by the Cognitive Science Laboratory at Princeton University.
• My web page on using Wordnet for Text Categorization.

5. Task oriented features

Some links on the topics covered in this part:

• The "Authorship Attribution with Support Vector Machines" presentation by Joachim Diederich et al.
• The METER (MEasuring TExt Reuse) project at the University of Sheffield.

6. Summary

Some links on the topics covered in this part:

• My ECML/PKDD'02 Tutorial on Text Mining and Internet Content Filtering.

For more information, please visit the Tutorials home page at the EACL'03 website.

Last updated February, 28th 2003