This paper explores the problem of detecting sentence-level forum authority claims in online discussions. Using a maximum entropy model, we explore a variety of strategies for extracting lexical features in a sparse training scenario, comparing knowledge- and data-driven methods (and combinations). The augmentation of lexical features with parse context is also investigated. We find that certain markup features perform remarkably well alone, but are outperformed by data-driven selection of lexical features augmented with parse context.
展开▼