Collaborative Research: Causal Discovery and Individualized Policy Optimization for Human Text Data

University Of Texas At Dallas, Richardson TX

Investigators

Abstract

Recent advancements in natural language processing (NLP) have led to a rapid increase in available text data, sparking research developments in precision medicine, economics, recommendation systems, and social science. While existing deep learning methods can predict outcomes accurately, it remains unclear how to disentangle, quantify, and use complex relationships among observed textual variables. Causal inference presents a solution for extracting trustworthy causal relationships and establishing counterfactual realities. This research project aims to develop statistical theories, methods, and algorithms to learn causal structure and establish causal identifications for text data. The project will impact various sectors, including the medical, financial, and health communities, promoting interdisciplinary collaboration. To tackle the challenges imposed by text data, the research project aims to solve the following tasks: (1) Establish a new approach to low-dimensional representation learning for text variables, with a primary focus on causal identifications; (2) Develop the textual causal structural learning and causal direction learning to identify the complex causal relationships between different text variables of interest, (3) Build a comprehensive analysis framework for average and heterogeneous textual causal effects that are able to accommodate textual features, textual actions, and textual outcomes, and develop their estimations with multiple robustness. (4) Construct an individualized online policy optimization framework tailored for text variables. During the involvement of the project, efficient computational algorithms that are designed to handle the challenges posed by large-scale and heterogeneous text data will be developed and implemented. In addition, the project will conduct software development for target applications in precision medicine and personalized recommendations. This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.