OBJECTIVE/DESCRIPTION:

This course will provide the students with (1) an overview of the area of advanced document categorization and clustering, (2) a thorough study on classification evaluation in the context of uneven distribution of text categories, and (3) a comprehensive comparison of dimensionality reduction techniques used specifically for text. The students will be provided with enough background so that a term project can be developed in which they will implement the different dimensionality reduction techniques and study their effect on the categorization algorithms in terms of accuracy and the different measures investigated in the course. The goal is to cover most recent research published in the field and get enough knowledge about dimensionality reduction to write a survey paper summarizing the taxonomy of the recent techniques with a comparative study in the context of text categorization and clustering.

The course will mainly consist of a series of discussions on the topics listed below as a general guideline. Throughout the course recent relevant research papers will also be read/discussed.

TOPICS:

• Techniques for dimensionality reduction in text documents
• TFIDF
• Single Value Decomposition
• Pricipal Componant Analysis
• Independent Component Analysis
• Latent Semantic Analysis and Latent Semantic Indexing
• Techniques for categorization validation
• Precision, Recall and Fallout with respect to false positives and false negatives
• Breakeven point
• F1 measure  
• ROC curve
• The effect of clustering on categorization outcome

TEXTBOOK and REFERENCES:

• Research papers will be selected from a variety of journals and conference proceedings.