Mixed Methodology to Predict Social Meaning for Decision Support

Report No. ARL-MR-0850
Authors: Barbra E. Chin, Candace C. Ross, and Michelle T. Vanni
Date/Pages: September 2013; 30 pages
Abstract: For analysts of social content in language, popular Internet forums provide ample material for observing usage variation patterns. A single phenomenon, scrutinized through the lens of an established theoretical construct, focuses an approach to analysis, which is computationally tractable and exploitable. The contribution to ongoing work reported on here shows how diverse and complex manifestations of a style-switching variant of the code-switching phenomenon can serve profitably as data input to machine learning. On a group membership prediction task, logistic regression results for user posts containing style features were in the high 80s. A novel representation of structures in posts without style features boosted results for this group as well. We indicate how this approach may extend to popular social media sites, such as Facebook, to inter-language code-switching and diverse computational tasks.
Distribution: Approved for public release
  Download Report ( 1.054 MBytes )
If you are visually impaired or need a physical copy of this report, please visit and contact DTIC.
 

Last Update / Reviewed: September 1, 2013