Critiquing text analysis in social modeling: Best practices, limitations, and new frontiers

Abstract

Natural language processing (NLP) is an important contributor to the field of social modeling. Language is a social artifact; it is how people express opinions, persuade, or convey what they believe is important. It is thus rightly recognized that computational tools can automate at least some of the analytical work of reading an ever-increasing volume of textual data, reducing time and costs. Language is also, however, a complex and variegated system, creating a challenge for social modelers. In this paper, we contend NLP's full potential is commonly not being exploited, leading to unnecessary work and lower-quality results, and that social modelers using NLP should understand at a high level what NLP problems are, and are not, solved. Our findings have implications for both the practice and validation of NLP in the social modeling community. © 2013 Springer-Verlag.