Conditional Hypothesis Generation for LLM-Based Text Analysis with Researcher-Specified Covariates

In three linesConditional hypothesis generation framework for LLM-based text analysis incorporating researcher-specified covariates. Addresses stratum imbalance and sign reversal through feature-covariate interactions and inverse-frequency reweighting. Validated on synthetic and real-world datasets in computational social science.

Summary generated by Claude — human-verified