51²è¹Ý

Abstract: Extracting actionable insights from unstructured text remains a fundamental goal of NLP, yet conventional approaches often fall short in temporal and causal reasoning and integrating domain-specific insights. This talk introduces innovative research directions to address these challenges across clinical and social science domains. First, I will introduce GraphTREx, a model that enhances clinical temporal relation extraction by integrating context-aware representations with graph neural networks. This integration enables improved timeline extraction from clinical notes, overcoming challenges posed by long documents and specialized jargon, thereby supporting more accurate patient representations. I will then turn to social data, illustrating how contextualized representations help quantify between-group polarization and use causal inference to examine when intergroup interactions mitigate it. Finally, I will present a comprehensive framework to  investigate gender bias in generative-AI-driven recruitment and highlight how behavioral alignment affects fairness.

About the Speaker: Rochana Chaturvedi is a Postdoctoral Research Fellow at Argonne National Laboratory, where she develops large language model (LLM) frameworks for scientific reasoning. Her research lies at the intersection of natural language processing, graph-based modeling, and causal machine learning, with applications spanning healthcare, computational social science, AI ethics, and climate science communication. She received her Ph.D. in Computer Science from the University of Illinois Chicago and previously served as an Assistant Professor at the University of Delhi. Her research has been published in venues such as ACL, the Web Conference (WWW), Political Analysis, and IEEE BIBM, and has been covered by major media outlets.

We look forward to you your active participation.