Grounding & knowledge
Evidence-grounded language models
I build systems that use retrieval, knowledge graphs, and collaborating agents to find authoritative evidence. I focus on responses that are factual, traceable, and robust.
Ph.D. Student · NLP & Large Language Models
Anirban Saha Anik
University of North Texas · Denton, Texas
I study how language models use evidence, adapt to readers, and communicate reliably in health and crisis settings. My work connects evidence-grounded generation, human-adaptive responses, and rigorous evaluation to make AI support more factual, understandable, useful, and safe.
I combine retrieval-augmented generation, knowledge graphs, multi-agent systems, controllable text generation, and human evaluation, with a focus on health misinformation and crisis communication.
Working with Dr. Lingzi Hong in the Human-Centered Computing Lab.
Open to 2027 Applied Scientist, Research Scientist, and Machine Learning internships.
Research focus
Explore researchPeople & interaction
Adaptive human-AI interaction
I study when AI should ask, answer, adapt, or collaborate. Controllable generation, health literacy, and human feedback guide explanations that fit the reader while preserving factual meaning.
Measurement & reliability
Evaluation for high-stakes NLP
I combine benchmarks, automated metrics, and human studies to evaluate factuality, consistency, audience alignment, and actionability. I also test whether responses reflect the right context, place, and time.
- Trustworthy LLMs
- Knowledge Graphs & RAG
- Human-Centered AI
- Multi-Agent Systems
- LLM Evaluation
Selected publications
All publicationsEMNLP 2026
Ask or Answer: A Decision Framework for Multi-Turn Health Misinformation Intervention ↗
A decision framework that learns when a health-misinformation intervention should ask for clarification and when it should answer directly.
Authors
Xiaoying Song, Anirban Saha Anik , Jinyu Liu, Qitao Tan, Geng Yuan, and Lingzi Hong
COLM 2025Runner-up · SIG-SM 2025
Multi-Agent Retrieval-Augmented Framework for Evidence-Based Counterspeech Against Health Misinformation →
Specialized agents retrieve, summarize, generate, and refine evidence-based counterspeech using static and dynamic sources.
Authors & recognition
Anirban Saha Anik , Xiaoying Song, Elliott Wang, Bryan Wang, Bengisu Yarimbas, and Lingzi Hong
Runner-Up, SIG-SM Social Media Research Competition 2025
EMNLP 2025Finalist · SIG-SM 2025
Speaking at the Right Level: Literacy-Controlled Counterspeech Generation with RAG-RL →
A RAG and reinforcement-learning framework for generating evidence-grounded counterspeech at the reader's health-literacy level.
Authors & recognition
Xiaoying Song, Anirban Saha Anik , Dibakar Barua, Pengcheng Luo, Junhua Ding, and Lingzi Hong
Finalist, SIG-SM Social Media Research Competition 2025
EMNLP 2025
A Dynamic Fusion Model for Consistent Crisis Response ↗
A dynamic fusion framework for consistent, professional, and actionable crisis responses, together with a consistency metric.
Authors
Xiaoying Song, Anirban Saha Anik, Eduardo Blanco, Vanessa Frías-Martínez, and Lingzi Hong
Recent news
TimelineSpatiotemporal crisis-response paper at JCDL 2026
Our paper, “Can LLMs Reason Over Where and When? Benchmarking Spatiotemporal Synthesis in Crisis Response,” will appear at JCDL 2026 (October 13–16, Frisco, Texas).
Read the paperAsk or Answer accepted at EMNLP 2026
Our decision framework studies when a health-misinformation intervention should ask for clarification and when it should answer directly.
Read the paperThree systems at CLEF CheckThat! 2026
SciClaimSeekers, SourceMinds, and VeriFind addressed scientific-source retrieval, citation auditing, and numerical claim verification; VeriFind placed second in Task 2.
View 2026 publicationsStarted my Ph.D. at the University of North Texas
I began the Computer Science and Engineering program, focusing on trustworthy language systems for health and crisis communication.
View academic journeyCompleted my M.S. in Data Science
I graduated from the University of North Texas with a 4.00/4.00 GPA.
View education