Generating Ethnographic Models from Communities' Online Data

In this paper, we describe computational ethnography studies to demonstrate how machine learning techniques can be utilized to exploit bias resident in language data produced by communities with online presence. Specifically, we leverage the use of figurative language (i.e., the choice of metaphors) in online text (e.g., news media, blogs) produced by distinct communities to obtain models of community worldviews that can be shown to be distinctly biased and thus different from other communities' models. We automatically construct metaphor-based community models for two distinct scenarios: debates on gun rights and marriage equality. We then conduct a series of experiments to validate the hypothesis that the metaphors found in each community's online language convey the bias in the community's worldview.