BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Talks.cam//talks.cam.ac.uk//
X-WR-CALNAME:Talks.cam
BEGIN:VEVENT
SUMMARY:A Look at Partial Projections for Regression onto Text - Matt Tadd
 y (University of Chicago)
DTSTART:20100528T150000Z
DTEND:20100528T160000Z
UID:TALK24459@talks.cam.ac.uk
CONTACT:8047
DESCRIPTION:\nAn increasingly common problem in data analysis is to infer\
 nthe relationship between text and characteristics of the speaker or\ndocu
 ment source. Various modifications of the multinomial bag-of-words\nmodel 
 are most prominent among approaches designed specifically for\ntext regres
 sion\, although many generic high-dimensional pattern\nrecognition techniq
 ues are also applicable. We investigate one such\ngeneric technique\, part
 ial least-squares (PLS)\, which is commonly used\nin engineering and physi
 cal sciences. This inquiry is motivated by the\ndiscovery that ``slant-mea
 sure''\, a heuristic from political economics\nfor regressing ideology ont
 o text\, is just the first PLS direction.\nOur goal is to provide a Bayesi
 an analysis scheme for text regression\nwhich takes advantage of the mecha
 nics (and initial economic\nmotivation) of PLS\, and to this end we devise
  model-based\ninterpretations of the algorithm and adapt these to account 
 for the\nspecifics of text-count covariate matrices. Results are provided 
 in\nthe motivating application of ideology analysis for the 109th US\nCong
 ress.\n\n\n\nhttp://faculty.chicagobooth.edu/matt.taddy/
LOCATION:MR12\, CMS\, Wilberforce Road\, Cambridge\, CB3 0WB
END:VEVENT
END:VCALENDAR