Vasudha Varadarajan

🧚

Vasudha

Postdoc

I am Vasudha Varadarajan, a postdoc at CMU LTI (since August 2025) working in the areas of NLP: understanding and modeling the human cognition as it manifests in conversations between humans as well as with LLMs. I am interested in applying NLP techniques to problems in computational social science to study socio-psychological phenomena through language use at scale, active learning, and improving conversational AI for psychometric surveys. I am also interested in interpretability and explainability of language models, and exploring NLP for low-resource and code-mixed settings. I did my PhD under the mentorship of H. Andrew Schwartz in the State University of New York at Stony Brook. I am fortunate to be mentored by Maarten Sap and Mona Diab at LTI.

If you have similar interests and would like to work together, please feel free to reach out!

Download CV

Interests

Natural Language Processing
Computational Social Science

Education

PhD in Computer Science
Stony Brook
MSc (Hons.) Physics
Birla Institute of Technology and Science (BITS) Pilani
BE (Hons.) Computer Science
Birla Institute of Technology and Science (BITS) Pilani

📚 My Research

I am a Ph.D. candidate working on modeling cognitive styles expressed through personal discourse using large language models. I am broadly interested in using NLP and data science to study social/psychological phenomena through language use, annotations and active learning, and improving conversational AI, especially for mental health. I am always looking for opportunities to work with teams with similar interests. Please reach out to collaborate 😃

Featured Publications

Ai-Generated, Human-Generated, Content Authorship, Homogeneity of LLM Text

The Consistent Lack of Variance of Psychological Factors Expressed by LLMs and Spambots

AI-generated text, while coherent and human-like, exhibits significantly less variation in psychological traits compared to human-authored content, consistently modeling an "average" human personality. This limitation makes AI text distinguishable from human writing through analysis of fundamental human traits, revealing current AI's inability to capture the full spectrum of human psychological diversity.

Jan 1, 2025

Language of Extremism, Radicalization, Social Media, Generalized Framework

Unifying the Extremes: Developing a Unified Model for Detecting and Predicting Extremist Traits and Radicalization

A novel framework using "The Extremist Eleven" factors successfully analyzes and predicts extremist behavior across different online communities by examining verbal behavioral signatures. The method demonstrates high predictive accuracy, identifying potential extremists up to 10 months before they join extremist communities, while showing that users maintain consistent extremist traits after joining these forums.

Jan 1, 2025

Mental Health, Assessments, Adaptive, Item Response Theory, Chatbot

ALBA: Adaptive Language-Based Assessments for Mental Health

This study introduces Adaptive Language-Based Assessment (ALBA) for mental health, using limited language responses to adaptively order questions and assess psychological traits. Two methods, ALIRT and Actor-Critic, outperformed non-adaptive baselines. ALIRT proved most effective, achieving high accuracy with fewer questions, demonstrating that adaptive assessments can maintain validity with smaller language samples and lower computational costs.

Jan 1, 2024

Human-in-the-Loop, Active Learning, Large Language Models, Dissonance, Stance

Transfer and active learning for dissonance detection: Addressing the rare-class challenge

This study explores transfer- and active learning solutions for rare-class problems, focusing on detecting cognitive dissonance in social media. We propose a probability-of-rare-class (PRC) approach for selecting samples and evaluate various acquisition strategies. We find that PRC effectively guides annotations and improves model accuracy, while specific transfer-learning sequences enhance initial performance but don't benefit subsequent active learning iterations.

Jan 1, 2023

Recent Publications

Vasudha Varadarajan★, Salvatore Giorgi★, Siddharth Mangalik, Nikita Soni, David M Markowitz, H Andrew Schwartz (2025). The Consistent Lack of Variance of Psychological Factors Expressed by LLMs and Spambots. Proceedings of the Workshop on Detecting AI Generated Content at COLING 2025.

PDF Cite

Allison Lahnala★, Vasudha Varadarajan★, Lucie Flek, H Andrew Schwartz, Ryan L Boyd (2025). Unifying the Extremes: Developing a Unified Model for Detecting and Predicting Extremist Traits and Radicalization. To Appear in Proceedings of the International AAAI Conference on Web and Social Media.

PDF Cite

Vasudha Varadarajan, Syeda Mahwish, Xiaoran Liu, Julia Buffolino, Christian C. Luhmann, Ryan L. Boyd, H Andrew Schwartz (2024). Upcoming Talk @ CRA Grad Cohort, Denver 2025 on Capturing Human Cognitive Styles with Language -- Towards an Experimental Evaluation Paradigm.. Under Review at ACL-ARR October 2024.

Vasudha Varadarajan, Sverker Sikström, Oscar Kjell, H Schwartz (2024). ALBA: Adaptive Language-Based Assessments for Mental Health. Proceedings of the 2024 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 1: Long Papers).

PDF Cite Code

Vasudha Varadarajan, Allison Lahnala, Adithya v Ganesan, Gourab Dey, Siddharth Mangalik, Ana-Maria Bucur, Nikita Soni, Rajath Rao, Kevin Lanning, Isabella Vallejo, Others (2024). Archetypes and Entropy: Theory-Driven Extraction of Evidence for Suicide Risk. Proceedings of the 9th Workshop on Computational Linguistics and Clinical Psychology (CLPsych 2024).

PDF Cite Code Video

See all publications

Recent & Upcoming Talks