Large Margin Methods in Information Extraction and Content Categorization

Thomas Hofmann
Technical University of Darmstadt and Fraunhofer IPSI

Support Vector Machines (SVMs) have been one of the major breakthroughs in
machine learning, both in terms of their practical success as well as their
learning-theoretic properties. This talk presents a generic extension of SVM
classification to the case of structured classification, i.e. the task of
predicting output variables with some meaningful internal structure. As we
will show, this approach has many interesting applications in information
extraction, information retrieval, document categorization and natural
language processing, including supervised training of Markov Random Fields
and probabilistic context-free grammars.

Joint work with Ioannis Tsochantaridis (Google), Yasemin Altun (Toyota
Technical Institute) and Thorsten Joachims (Cornell University)