Bayesian Logistic Regression in Text Classification and Mining (Plus A Big New Test Collection)

Bayesian logistic regression allows incorporating task knowledge through model structure and priors on parameters. I will discuss content-based text categorization and authorship attribution using 1) priors that control sparsity and sign of parameters, 2) priors that incorporate domain knowledge from reference books and other texts, and 3) the use of polytomous (1-of-k) dependent variables. All experiments were performed with our open-source programs, BBR and BMR, which can fit models with millions of parameters. (Joint work with David Madigan, Alex Genkin, Aynur Dayanik, Dmitriy Fradkin, and Vladimir Menkov at Rutgers and DIMACS.)

I will also briefly discuss the IIT CDIP (Complex Document Information Processing) test collection, which I am developing under an ARDA subcontract to Illinois Institute of Technology. It is based on 1.5TB of scanned and OCR'd documents released in tobacco litigation, and will be a major resource for research in information retrieval, document analysis, social network analysis, and perhaps databases. (Joint work with Gady Agam, Shlomo Argamon, Ophir Frieder, Dave Grossman, and a cast of hundreds.)
Dave Lewis is based in Chicago, IL, and consults on information retrieval, data mining, and natural language processing. He previously held research positions at AT&T Labs, Bell Labs, and the University of Chicago. He received his Ph.D. in Computer Science from the University of Massachusetts, Amherst, and did his undergraduate work down the road at Michigan State.