Statistics – Methodology
Scientific paper
2010-12-09
Statistics
Methodology
Software is available in the textir package for R
Scientific paper
Text data, including speeches, stories, and other document forms, is often composed with regard to sentiment variables that are of interest for research in marketing, economics, and other social research fields. It is also very high dimensional and difficult to incorporate into statistical analysis. This article introduces a straightforward framework of sentiment-preserving dimension reduction for text data. Our aim is to provide a general approach to text regression while avoiding the model complexity characterizing much of statistical learning for language. Using an inverse regression approach, we show that multinomial logistic regression of phrase counts onto document characteristics can be used to obtain low dimensional document representations that are rich in sentiment information. In addition to this text-specific work, the article introduces an estimation framework that should be generally applicable for high-dimensional logistic regression. In particular, we propose independent Laplace priors for each coefficient loading and advocate joint MAP estimation of coefficients and the associated prior scale. It is shown that this scheme yields a novel and stable approach to nonconcave penalized likelihood estimation for logistic regression. We also survey related approaches from the literature, connecting econometric methodology to partial least squares and linking both to inverse regression and related frameworks. Finally, the work is motivated through three detailed examples and we provide an out-of-sample prediction study to illustrate the effectiveness of our methods.
No associations
LandOfFree
Inverse Regression for Analysis of Sentiment in Text does not yet have a rating. At this time, there are no reviews or comments for this scientific paper.
If you have personal experience with Inverse Regression for Analysis of Sentiment in Text, we encourage you to share that experience with our LandOfFree.com community. Your opinion is very important and Inverse Regression for Analysis of Sentiment in Text will most certainly appreciate the feedback.
Profile ID: LFWR-SCP-O-77742