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 two innovative research directions to address these challenges across healthcare 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 representation learning. Next, I will focus on understanding societal dynamics by combining NLP and causal inference methods. I will describe a contextualized embedding-based metric that quantifies social-media polarization and examine when intergroup interactions mitigate between-group polarization. I will conclude with ongoing efforts to adapt large language models for complex reasoning tasks, with a focus on domain knowledge infusion, bias mitigation, and comprehensive evaluation.

To add to calendar:

Click on:  https://wordpress.cels.anl.gov/cels-seminars/

Enter your credentials.

Search for your seminar 

Click “Add to calendar”