Class Meeting for 10-707 9/22/2010

This is one of the class meetings on the schedule for the course Information Extraction 10-707 in Fall 2010.

Linear-chain CRFs

• Slides
• Supplement to Sha & Pereira's paper - a more detailed derivation of the CRF gradient.

Required Readings

• Shallow parsing with conditional random fields, by F. Sha, F. Pereira. In Proceedings of HLT-NAACL, 2003.
• Conditional structure versus conditional estimation in NLP models, by D. Klein, C. D Manning. In Proceedings of the ACL-02 conference on Empirical methods in natural language processing-Volume 10, 2002.

Optional Readings

• Hidden Markov Models for Labeled Sequences, Krogh 1994. The method of this paper appears to be equivalent to linear-chain CRFs - so why didn't it catch on?
• Gradient tree boosting for training CRFs, Dietterich et al, ICML 2004. A very different training method for CRFs, based on regression trees.
• Semi-Supervised Conditional Random Fields for Improved Sequence Segmentation and Labeling, Jiao et al, ACL 2006. A very nice paper from the UofA group on semi-supervised CRF learning.
• Accelerated Training of Conditional Random Fields with Stochastic Gradient Methods, Vishwanathan et al, ICML 2006. CRF learning methods seem complicated - this paper shows that stochastic gradient methods, a class of very simple on-line methods, can be competitive.
• Choi, Y., and C. Cardie. Hierarchical Sequential Learning for Extracting Opinions and their Attributes. ACL-2010 (short paper)
• Lavergne, T., O. Cappé, T. ParisTech, and F. Yvon. ractical very large scale CRFs. ACl-2010. Full of detailed implementation-oriented tricks.

Background

• Conditional Random Fields: Probabilistic Models for Segmenting and Labeling Sequence Data, Lafferty et al, 2001. The original CRF paper.
• An Introduction to Conditional Random Fields for Relational Learning. A longish tutorial overview of CRFs.