Xingmeng Zhao, PhD

Postdoctoral Fellow in Biomedical Informatics

Passionate researcher dedicated to responsible AI in healthcare, with expertise in clinical informatics, natural language processing, and human-AI collaboration. Currently advancing ethical AI systems and clinical NLP at the University of Colorado Anschutz Medical Campus.

About Me

I am currently a Postdoctoral Fellow in the Department of Biomedical Informatics at the University of Colorado School of Medicine, where I focus on developing ethical AI systems and advancing clinical natural language processing.

I completed my PhD in Information Technology from the University of Texas at San Antonio in 2025, under the guidance of Dr. Anthony Rios. My research brings together Natural Language Processing (NLP), Large Language Models (LLMs), and Human-Computer Interaction (HCI) to address real-world challenges in healthcare and society, with a focus on fairness, bias detection, and promoting equity in AI for healthcare.

My research journey began with a strong foundation in mathematics and statistics, earning my MS from the University of Colorado Denver and my BS in Statistics from Minzu University of China in Beijing. I have published in prestigious venues across NLP, biomedical informatics, and computational social science, including NAACL, COLING, ICWSM, BioNLP, Clinical NLP, and JAMIA.

Throughout my academic career, I have been recognized for excellence in research and teaching, receiving multiple awards including the PhD Student of the Year Award from UTSA Carlos Alvarez College of Business and being selected as one of 30 national Future Research Leaders in Artificial Intelligence by the University of Michigan.

What is Responsible AI for Healthcare?

Responsible AI ensures that algorithms operate fairly and transparently across all stages, from data collection to model inference to deployment, by actively detecting and reducing ethical harms such as bias, privacy issues, and unintended consequences.

Why Responsible AI Matters

In healthcare, even minor biases can lead to misdiagnoses, unequal treatment recommendations, and exacerbated health disparities. By rigorously measuring bias in data and models, and speculating about future bias and its consequences, we can reduce risks before deployment by examining how AI performs across different patient contexts, especially edge cases where misalignment with users' needs can lead to unintended harm.

My Comprehensive Framework

My main contribution is developing a comprehensive framework to identify, measure, and mitigate bias in biomedical and social applications through three main areas:

1. Clinical Informatics

We develop methods to extract Social Determinants of Health (SDoH), such as income, housing, and employment, from unstructured EHR notes and turn them into structured data. Real-world bias goes beyond race, age, and gender, and capturing these social factors helps us better understand long-term impacts and improve clinical decisions.

2. Public Health Informatics

We develop novel prompting strategies to understand how the public perceives news headlines and how hidden bias in content such as implicit framing or unbalanced reporting, can shape public opinion and diverge from real-world data. We also develop benchmarks to evaluate when, how, and why biomedical AI systems fail, especially across different identities like race or gender.

3. Human-Centered Design

Inspired by "Black Mirror," we develop a multi-agent system that uses LLMs to simulate virtual environments and character interactions. It generates user stories that reflect potential harms and benefits, helping people speculate about future bias and its consequences before AI models are developed and deployed.

AI in Healthcare

Developing responsible AI systems that prioritize ethical considerations, patient safety, and community well-being in healthcare applications.

Social Determinants of Health Extraction

Developed a marker-based neural network system to extract Social Determinants of Health (SDoH) from clinical notes, supporting context-aware and personalized patient care.

Clinical NLP

Advancing natural language processing techniques for clinical applications, focusing on information extraction, text classification, and medical report analysis.

Temporal Relation Extraction

Evaluating instruction-tuned language models for temporal relation extraction in clinical timelines, improving understanding of medical event sequences.

Medical Report Summarization

Developing multi-modal retrieval-based systems for chest X-ray report summarization and improving expert radiology reports through layperson summary prompting.

Human-AI Collaboration

Designing conversational interfaces, multi-agent dialogue systems, and human-centered AI interactions that enhance rather than replace human capabilities.

Multi-Agent Systems for Ethical AI

Developing virtual environments using large language models to simulate character interactions and generate user stories that help identify potential harms and benefits of AI systems before deployment.

Conversational AI Interfaces

Creating intuitive dialogue systems that facilitate natural human-AI interaction, focusing on understanding user intent and providing contextually appropriate responses.

Related Work:

Computational Social Science

Applying computational methods to understand social phenomena, media representation, and community dynamics through data-driven approaches.

Media Framing Analysis

Built AI systems to analyze how media framing in news headlines shapes public perception, focusing on the implicit portrayal of cyclists in news coverage.

Social Media and Financial Markets

Investigating how social communities can drive Wall Street through meme stocks and social media discourse analysis.

Mechanistic Interpretability for AI Safety

Investigating the internal mechanisms of AI models to understand their decision-making processes, ensure transparency, and develop safer, more reliable AI systems.

Model Transparency and Explainability

Research into understanding how large language models and neural networks make decisions, focusing on identifying interpretable features and causal mechanisms within model architectures.

AI Safety through Understanding

Developing methods to probe and understand AI model behavior to identify potential risks, biases, and failure modes before deployment in critical applications like healthcare.

Education

Aug 2020 – Aug 2025

PhD in Information Technology

University of Texas at San Antonio

Carlos Alvarez College of Business

Advisor: Dr. Anthony Rios

Dissertation: Responsible AI for Healthcare and Community Well-being From Ethical Design to Practical Deployment

Aug 2015 – Dec 2017

MS in Applied Mathematics and Statistics

University of Colorado Denver

Advisor: Dr. Stephanie Santorico

Aug 2008 – May 2012

BS in Statistics

Minzu University of China, Beijing

Professional Experience

Sept 2025 – present

Postdoctoral Fellow

University of Colorado Anschutz Medical Campus
  • Working on ethical AI and clinical NLP systems
Feb 2023 – Dec 2024

Graduate Teaching Assistant

The University of Texas at San Antonio
  • Instructed undergraduate Information System courses: IS 1413 (Excel for Business Information Systems), IS 1403 (Business Info Systems Fluency), and IS 2053 (Programming Languages I with Scripting)
Aug 2020 – Aug 2025

Graduate Research Assistant

The University of Texas at San Antonio
  • Designed a multi-agent system that generates benefit–harm user stories and guides red-team discussions with conversational agents to help stakeholders identify ethical risks early in AI development
  • Developed a neural network model to extract Social Determinants of Health (SDoH) from clinical notes to support context-aware, personalized care
  • Investigated gender bias in chemical Named Entity Recognition systems for drug safety, revealing underrecognition of female-associated entities
  • Built an AI system to analyze how media framing in news headlines shapes public perception of cyclists

Selected Publications

Peer-Reviewed Journal Publications

A Marker-based Neural Network System for Extracting Social Determinants of Health

Xingmeng Zhao and Anthony Rios
Journal of the American Medical Informatics Association (JAMIA), 2023
Impact Factor: 7.942 • H5-index: 74 • Quartile: Q1 • ABDC Journal Quality List: A • Cited by: 8

Peer-Reviewed Conference Publications

Bike Frames: Understanding the Implicit Portrayal of Cyclists in the News

Xingmeng Zhao, Xavier Walton, Suhana Shrestha, Anthony Rios
International AAAI Conference on Web and Social Media (ICWSM), 2025
H5-index: 56 • Acceptance Rate: 20% • Cited by: 2

Charting the Future: Using Chart Question-Answering for Scalable Evaluation of LLM-Driven Data Visualizations

James Ford, Xingmeng Zhao, Dan Schumacher, and Anthony Rios
31st International Conference on Computational Linguistics (COLING 2025)
H5-index: 65 • Cited by: 6

A Comprehensive Study of Gender Bias in Chemical Named Entity Recognition Models

Xingmeng Zhao, Ali Niazi, Anthony Rios
Conference of the 2024 North American Chapter of the Association for Computational Linguistics (NAACL 2024)
H5-index: 132 • Acceptance Rate: 23% • Cited by: 2

Publications Under Review and In Preparation

Improving Expert Radiology Report Summarization by Prompting Large Language Models with a Layperson Summary

Xingmeng Zhao, Tongnian Wang, and Anthony Rios
arXiv:2406.14500 • Under Review at ACL 2026
Cited by: 1

Towards Safer AI in Healthcare: Using Storytelling to Educate about Unintended Harm from AI-Powered Tools

Xingmeng Zhao and Anthony Rios
In preparation for Journal of Management Information Systems (JMIS)

BabyStories: Can Reinforcement Learning Teach Baby Language Models to Write Better Stories?

Xingmeng Zhao, Tongnian Wang, Sheri Osborn, Anthony Rios
Proceedings of the BabyLM Challenge at the 27th Conference on Computational Natural Language Learning (CoNLL) at EMNLP 2023
Cited by: 5

UTSA-NLP at SemEval-2022 Task 4: An Exploration of Simple Ensembles of Transformers, Convolutional, and Recurrent Neural Networks

Xingmeng Zhao and Anthony Rios
Proceedings of the 16th International Workshop on Semantic Evaluation (SemEval-2022)
Cited by: 2

Teaching Experience

Fall 2024

IS 1403 Business Info Systems Fluency

Instructor

Class Size: 95 students

Evaluations: Instructor Rating 4.4/5.0 • Course Rating 4.5/5.0

Spring 2024

IS 2053 Programming Languages I with Scripting

Instructor

Class Size: 43 students

Evaluations: Instructor Rating 3.9/5.0 • Course Rating 4.0/5.0

Spring 2023

IS 1413 Excel for Business Information Systems

Instructor

Class Size: 75 students

Evaluations: Instructor Rating 4.6/5.0 • Course Rating 4.7/5.0

Awards & Grants

2024

Future Research Leaders in Artificial Intelligence

University of Michigan (30 national recipients)
June 2025

Travel Award

University of Michigan MIDAS DAIR3 Program
March 2025

PhD Excellence Fund

UTSA Carlos Alvarez College of Business
April 2023

Ph.D. Student of the Year Awards

UTSA Carlos Alvarez College of Business
April 2023

Travel Award

University of Michigan MIDAS Future Leaders Summit