Foundations of Statistical Natural Language Processing

Christopher D. Manning and Hinrich Schütze

Chapter 16: Text Categorization

Links referred to in the text

• David D. Lewis' homepage - source of The Reuters Collection
• TIMBL, a nearest neighbor classifier
• UCI machine learning repository
• UTCS Neural Networks Research Group
• rainbow text classification system
• Adwait Ratnaparkhi's MaxEnt tagger

Data used in the chapter

• The Reuters-21578 Text Categorization Test Collection. This data set precisely defines the "Modified Apté" split used in the chapter (David Lewis' updating of Apté et al.'s (1994) split to the revised Reuters-21578 collection, which he prepared).
• Format of data files
• List of 20 features
• Document vectors (774 KB) or gzipped document vectors (109 KB)
• Scripts and reformatted data files donated by Martin Jansche. These files can be directly used as input to ML programs like Weka.

Other links

• Naive Bayes software for learning to classify text And a different set of training/testing data for text classifiers. Part of the online companion for Tom Mitchell's Machine Learning text. Based on the Rainbow/Libbow software package (by Andrew McCallum).
• AT&T How May I Help You Call Routing demo
• Packages for learning Support Vector Machines include:
  • SVM-light (Thorsten Joachims)
  • libsvm
  • TinySVM
• Packages for learning maximum entropy or conditional loglinear models include:
  • Stanford Classifer
  • YASMET (Franz Och)