A bootstrapping approach for identifying stakeholders in public-comment corpora

A stakeholder is an individual, group, organization, or community that has an interest or stake in a consensus-building process. The goal of stakeholder identification is identifying stakeholder mentions in natural language text. We present novel work in using a bootstrapping approach for the identification of stakeholders in public comment corpora. We refine the definition of a stakeholder by categorizing stakeholders into 2 distinct stakeholder types and experiment with automatically identifying one of these two types: instances where the author identifies him/herself as a member of a particular group. An existing bootstrapping information extraction algorithm is combined individually with 3 distinct extraction pattern templates. Results show that this stakeholder group can be learned in a minimally supervised bootstrapping framework using 2 of the 3 extraction pattern templates. An experimental analysis explores the challenges in applying the third extraction pattern template to this problem. Results on all 3 extraction pattern templates provide insight on the unique and novel challenge of identifying stakeholders.