VaLiSTAD 2025: Workshop on Validating Social Text-as-Data Methods

VaLiSTAD 2025: the Workshop on Validating Social Text-as-Data Methods addresses a fundamental methodological challenge in computational social science: how do we know that NLP tools designed for general text actually measure what researchers think they measure when applied to social media data? Topics include systematic evaluation of text classifiers on diverse social corpora, auditing of LLMs for sociodemographic and political bias, reproducibility standards, benchmarking practices, and best practices for applying computational methods to political communication research.

The workshop is aimed at computational social scientists, political scientists, communication scholars, and NLP researchers who use or build text-as-data tools for studying society. As large language models are increasingly adopted for coding social data at scale, rigorous validation is essential to prevent systematic errors from propagating into published findings. VaLiSTAD provides a focused forum for developing community standards and sharing tools that make social text analysis more trustworthy and reproducible.

text-as-data validation social media data NLP model auditing computational social science LLM bias reproducibility political communication NLP sociodemographic bias