About

Lyle Ungar is a Professor of Computer and Information Science, Psychology, Bioengineering, Operations, Information and Decisions, and Genomics and Computational Biology at the University of Pennsylvania. His research interests encompass a broad range of topics including information economics, statistical relational learning, text mining, active learning, market-based methods for distributed scheduling and system optimization involving human and computer agents, and gene and protein expression. Ungar's work also focuses on clustering and collaborative filtering, genomics, regulatory network modeling, information extraction from biological texts and consumer data, machine learning, data mining, and computational biology, with future interests in computer go. Throughout his career, Ungar has contributed to advancing understanding in these fields through research that integrates computational methods with biological and social sciences. His work on information extraction and modeling in biological texts and genomics aims to improve data analysis in biological systems, while his investigations into market-based and active learning methods seek to optimize complex systems and decision-making processes. Ungar's interdisciplinary approach combines insights from computer science, psychology, and biology to address complex scientific and practical problems.

Research topics

• Computer Science
• Sociology
• Artificial Intelligence
• Medicine
• Psychology
• Mathematics
• Social psychology
• World Wide Web
• Psychiatry
• Nursing
• Virology
• Demography
• Gerontology
• Natural Language Processing
• Econometrics
• Economics
• Machine Learning
• Geography
• Linguistics
• Family medicine
• Environmental health
• Advertising
• Engineering
• Mathematical optimization

Selected publications

• The assessment of psychological richness, meaning, and happiness with social media text data: Predictive accuracy and distinct behavioral correlates

  PLoS ONE · 2026-01-07

  articleOpen access

  Assessing well-being with social media text data is a promising method, but besides hedonic well-being, little is known about whether additional well-being dimensions, such as psychological richness and eudaimonic well-being, can be predicted from such data. We compare the predictive accuracy for hedonic well-being, eudaimonic well-being, and the recently proposed construct of psychological richness in a large sample of Facebook users (n = 2,644), and find that the inclusion of language features incrementally improved model prediction accuracy beyond demographic features for psychological richness, but not for hedonic or eudaimonic well-being. Psychological richness had the lowest overall prediction accuracy (r = .21) followed by hedonic well-being (r = .27) and eudeomonic well-being (r = .29). The linguistic features associated with Psychological Richness were face valid, and in many instances the content and direction of the associations were unique to Psychological Richness, which provides discriminant validity evidence.

  PublisherDOI

• Self-Reported Side Effects of Semaglutide and Tirzepatide in Online Communities

  medRxiv · 2026-03-13

  articleOpen access

  ABSTRACT Social media can reveal patient experiences with glucagon-like peptide-1 receptor agonists (GLP-1 RAs) that extend beyond clinical trial data. We analyzed 410,198 Reddit posts (May 2019–June 2025) mentioning semaglutide or tirzepatide. A total of 67,008 users self-reported using these medications, and 43.5% described at least one side effect. Gastrointestinal symptoms predominated, including nausea (36.9%), fatigue (16.7%), vomiting (16.3%), constipation (15.3%), and diarrhea (12.6%). Notably, reproductive symptoms (e.g., menstrual irregularities) and temperature-related complaints (e.g., chills, hot flashes) emerged as unrecognized potential effects. These findings highlight patient concerns not well captured in current labeling or trials. Large-scale social media analysis can complement traditional pharmacovigilance by detecting emerging safety signals and expanding understanding of the real-world safety profile of GLP-1 RAs.

  PublisherOA PDFDOI

• Comparing patterns of recent mental health service use for predicting suicidal events following emergency department mental health visits in the United States: A national cohort study

  Social Psychiatry and Psychiatric Epidemiology · 2025-12-02

  articleOpen access

  OBJECTIVE: To examine patterns of recent service use to predict non-fatal suicidal events shortly following emergency department (ED) visits for mental health. METHODS: For this retrospective cohort study, we used Optum electronic health record data from 2,445,597 ED mental health episodes (2015-2022) for persons aged ≥ 10 years. We then constructed a series of logistic regression models to evaluate how six permutations of characterizing prior 180-day mental health service use predicted acute non-fatal suicidal events within 180 days of ED discharge beyond demographic characteristics and ED mental health diagnoses. Model performance was assessed by area under the receiver operating curve (AUC). RESULTS: Overall, 7.2% (n = 176,000) of episodes resulted in an acute suicidal event within 180 days. Model performance improved from demographic characteristics and ED mental health diagnoses (AUC = 0.76) when past 180-day service use variables were added, but with minimal differences between a binary variable of any mental health service use compared to monthly counts, weighted slopes, or interactions (AUC's 0.78-0.82). The final model containing demographics, ED mental diagnosis, and past 180-day service use yielded an AUC of 0.83. The most predictive past service use variables were any inpatient or ED event for self-harm or suicide ideation (OR = 5.45, 95% CI = 5.37, 5.54) and any ED mental health visit (OR = 1.87, 95% CI = 1.84, 1.90). CONCLUSIONS: As part of evaluating suicide risk in ED settings, information about recent acute care for mental health, suicidal ideation, or self-harm use significantly contributes to the short-term prediction of non-fatal suicidal events.

  PublisherOA PDFDOI

• Understanding gender and age differences in language use: cross-cultural insights from Weibo and Facebook

  Humanities and Social Sciences Communications · 2025-10-31 · 2 citations

  articleOpen accessSenior author

  This study integrates social role theory and socioemotional selectivity theory to investigate the cultural universalities and differences in language use among male and female users across different age groups on Weibo and Facebook. By analyzing social media language, we aim to understand how gender and age influence linguistic patterns and reflect broader cultural norms and societal values. Aggregated language from Weibo and Facebook users (N = 8728 per platform; 665,377 and 742,418 posts, respectively) was analyzed by both a top-down closed-vocabulary (Linguistic Inquiry and Word Count) approach and a data-driven open-vocabulary (Differential Language Analysis) approach. Our findings support and extend social role theory, showing that female users on both platforms use more communal and relational language, while male users focus on agentic and task-oriented content. Cultural dimensions, such as collectivism and individualism, modulate the expression of social roles, with Weibo users adhering more closely to traditional gender norms compared to Facebook users. Our findings also validate and extend the socioemotional selectivity theory by demonstrating how cultural frameworks shape the specific ways aging individuals pursue emotional and social goals. For example, on both platforms, age-related language patterns reveal a U-shaped trend in positive emotions, with a decline in middle age and an increase in older adulthood, reflecting a universal shift toward emotionally meaningful goals. Additionally, older users on Weibo engage more in collectivistic themes, while their Facebook counterparts focus on personal well-being and social ties. These results highlight the complex interplay between culture, gender, and age in shaping language use on social media, providing valuable insights into the cultural and societal influences on communication.

  PublisherOA PDFDOI

• Language models in digital psychiatry: challenges with simplification of healthcare materials

  NPP—Digital Psychiatry and Neuroscience · 2025-05-22 · 2 citations

  articleOpen access

  Linguistic hurdles in healthcare, such as complex language, significantly affect patient outcomes, including satisfaction with interaction, comprehension of healthcare materials, and engagement with the healthcare system. Reducing these hurdles has been a focus in healthcare delivery, as they significantly hinder patient engagement and adherence to treatments. The growing use of large language models (LLMs) in healthcare opens the possibility to reduce these linguistic hurdles. This study evaluates the ability of five prominent LLMs-GPT-3.5, GPT-4, GPT-4o, LLaMA-3, and Mistral-to simplify healthcare information to the standard recommended by the American Journal of Medicine. Our results indicate that while LLMs can approximate targeted reading levels, their outputs are inconsistent, with significant variability in reading level and deviation from the topic, making them unsuitable for deployment in healthcare settings.

  PublisherOA PDFDOI

• T-FIX: Text-Based Explanations with Features Interpretable to eXperts

  ArXiv.org · 2025-11-06

  preprintOpen access

  As LLMs are deployed in knowledge-intensive settings (e.g., surgery, astronomy, therapy), users are often domain experts who expect not just answers, but explanations that mirror professional reasoning. Yet evaluating whether an LLM "thinks like an expert" remains difficult: existing approaches rely on per-example expert annotation, making them costly, hard to scale, and tied to a single notion of correct reasoning within each domain. To address this gap, we introduce T-FIX, a unified evaluation framework that operationalizes expert alignment as a desired attribute of LLM-generated explanations. T-FIX spans seven scientific tasks across three domains, with each task evaluated against expert-defined criteria that capture domain-grounded reasoning rather than generic explanation quality. Our framework enables automatic, personalizable evaluation of expert alignment that generalizes to unseen explanations without ongoing expert involvement. Code is available at https://github.com/BrachioLab/FIX-2/.

  PublisherOA PDFDOI

• Voice-Based Chatbots for English Speaking Practice in Multilingual Low-Resource Schools: A Multi-Stakeholder Mixed-Methods Study

  2025-07-11

  preprint
  PublisherDOI

• The Impact of Language Mixing on Bilingual LLM Reasoning

  ArXiv.org · 2025-07-21

  preprintOpen accessSenior author

  Proficient multilingual speakers often intentionally switch languages in the middle of a conversation. Similarly, recent reasoning-focused bilingual large language models (LLMs) with strong capabilities in both languages exhibit language mixing-alternating languages within their chain of thought. Discouraging this behavior in DeepSeek-R1 was found to degrade accuracy, suggesting that language mixing may benefit reasoning. In this work, we study language switching in Chinese-English bilingual reasoning models. We identify reinforcement learning with verifiable rewards (RLVR) as the critical training stage that leads to language mixing. We show that language mixing can enhance reasoning: enforcing monolingual decoding reduces accuracy by 5.6 percentage points on MATH500. Additionally, a lightweight probe can be trained to predict whether a potential language switch would benefit or harm reasoning, and when used to guide decoding, increases accuracy by 2.92 percentage points. Our findings suggest that language mixing is not merely a byproduct of multilingual training, but is a strategic reasoning behavior.

  PublisherOA PDFDOI

• Knowing When Not to Answer: Lightweight KB-Aligned OOD Detection for Safe RAG

  ArXiv.org · 2025-08-04 · 1 citations

  preprintOpen access

  Retrieval-Augmented Generation (RAG) systems are increasingly deployed in high-stakes domains, where safety depends not only on how a system answers, but also on whether a query should be answered given a knowledge base (KB). Out-of-domain (OOD) queries can cause dense retrieval to surface weakly related context and lead the generator to produce fluent but unjustified responses. We study lightweight, KB-aligned OOD detection as an always-on gate for RAG systems. Our approach applies PCA to KB embeddings and scores queries in a compact subspace selected either by explained-variance retention (EVR) or by a separability-driven t-test ranking. We evaluate geometric semantic-search rules and lightweight classifiers across 16 domains, including high-stakes COVID-19 and Substance Use KBs, and stress-test robustness using both LLM-generated attacks and an in-the-wild 4chan attack. We find that low-dimensional detectors achieve competitive OOD performance while being faster, cheaper, and more interpretable than prompted LLM-based judges. Finally, human and LLM-based evaluations show that OOD queries primarily degrade the relevance of RAG outputs, showing the need for efficient external OOD detection to maintain safe, in-scope behavior.

  PublisherOA PDFDOI

• Designing Mental-Health Chatbots for Indian Adolescents: Mixed-Methods Evidence, a Boundary-Object Lens, and a Design-Tensions Framework

  ArXiv.org · 2025-11-11

  preprintOpen access

  Mental health challenges among Indian adolescents are shaped by unique cultural and systemic barriers, including high social stigma and limited professional support. We report a mixed-methods study of Indian adolescents (survey n=362; interviews n=14) examining how they navigate mental-health challenges and engage with digital tools. Quantitative results highlight low self-stigma but significant social stigma, a preference for text over voice interactions, and low utilization of mental health apps but high smartphone access. Our qualitative findings reveal that while adolescents value privacy, emotional support, and localized content in mental health tools, existing chatbots lack personalization and cultural relevance. We contribute (1) a Design-Tensions framework; (2) an artifact-level probe; and (3) a boundary-objects account that specifies how chatbots mediate adolescents, peers, families, and services. This work advances culturally sensitive chatbot design by centering on underrepresented populations, addressing critical gaps in accessibility and support for adolescents in India.

  PublisherOA PDFDOI

Recent grants

• Predicting AOD Relapse and Treatment Completion from Social Media Use

  NIH · $1.6M · 2014–2019

• Training Program in Computational Genomics

  NIH · $9.0M · 1999–2030

Frequent coauthors

• Raina M. Merchant

  University of Pennsylvania

  113 shared

• Sharath Chandra Guntuku

  University of Pennsylvania Health System

  110 shared

• Johannes C. Eichstaedt

  95 shared

• H. Andrew Schwartz

  93 shared

• David A. Asch

  86 shared

• Salvatore Giorgi

  University of Pennsylvania

  75 shared

• Brenda Curtis

  National Institute on Drug Abuse

  60 shared

• João Sedoc

  60 shared

Labs

• Operations, Information and Decisions DepartmentPI

Awards & honors

• A 680,000-Person Megastudy of Nudges to Encourage Vaccinatio…
• Megastudies Improve the Impact of Applied Behavioural Scienc…