Expert-Informed Topic Models for Document Set Discovery

The first step in many text-as-data studies is to find documents that address a specific topic within larger document set. Researchers often rely on simple keyword searches do this, even though this may introduce considerable selection bias. Such bias be greater when researchers lack the domain knowledge required make informed search decisions, for example, cross-national research or unfamiliar social contexts. We propose expert-informed modeling (EITM) as hybrid approach tackle problem. EITM combines validity of external captured through expert surveys with probabilistic models help identify subsets cover initially unknown domain-specific topics, such events and debates, belong researcher-defined master topic. flexible efficient thematic from large text corpora further study. benchmark validate method by discovering blog posts public role religion Australian, Swiss, Turkish provide complete workflow guide application their own work.