About

Jonathan Gratch is a Research Professor of Computer Science and Psychology at USC. His work focuses on the intersection of computer science and psychology, contributing to understanding human-computer interaction, emotion modeling, and virtual agents. His research aims to develop intelligent systems that can recognize, interpret, and simulate human emotions to improve communication and collaboration between humans and machines.

Research topics

• Computer Science
• Psychology
• Social psychology
• Cognitive psychology
• Communication
• Human–computer interaction
• Sociology
• Epistemology
• Economics

Selected publications

• AI-Mediated Dispute Resolution

  Proceedings of the AAAI Symposium Series · 2025-05-28 · 1 citations

  articleOpen accessSenior author

  We examine the effectiveness of large language model (LLM) mediations in the under-studied dispute resolution domain. We first used a new corpus of dispute resolutions, KODIS, to investigate if LLMs can correctly identify whether to intervene. We find evidence that GPT as a mediator picks up on salient aspects of a dispute, such as Frustration and whether the disputants ultimately come to a resolution or stall at an impasse --- intervening significantly more so in cases of high frustration and impasse. Afterward, we ran a user study to compare GPT mediations against those of novice human mediators. We find participants agreed GPT's mediations were more likely to lead to resolution; were better positioned in the dialog; had better justification than human-crafted ones; and, on a forced choice, were generally more effective than novice human mediations.

  PublisherOA PDFDOI

• ASTRA: A Negotiation Agent with Adaptive and Strategic Reasoning via Tool-integrated Action for Dynamic Offer Optimization

  ArXiv.org · 2025-03-10

  preprintOpen access

  Negotiation requires dynamically balancing self-interest and cooperation within the flow of conversation to maximize one's own utility. Yet, existing agents struggle due to bounded rationality in human data, low adaptability to counterpart behavior, and limited strategic reasoning. To address this, we introduce principle-driven negotiation agents, powered by ASTRA, a novel framework for turn-level offer optimization grounded in two core principles: opponent modeling and Tit-for-Tat reciprocity. ASTRA operates in three stages: (1) interpreting counterpart behavior, (2) optimizing counteroffers via a tool-integrated action with a linear programming (LP) solver, and (3) selecting offers based on strategy assessment and the partner's acceptance probability. Through simulations and human evaluations, our agent effectively adapts to an opponent's shifting stance and achieves favorable outcomes through enhanced adaptability and strategic reasoning. Beyond enhancing negotiation performance, it also serves as a powerful coaching tool, offering interpretable strategic feedback and optimal offer recommendations beyond human bounded rationality, with its potential further validated through human evaluation.

  PublisherOA PDFDOI

• ASTRA: A Negotiation Agent with Adaptive and Strategic Reasoning via Tool-integrated Action for Dynamic Offer Optimization

  2025-01-01 · 1 citations

  articleOpen access

  Negotiation requires dynamically balancing self-interest and cooperation within the flow of conversation to maximize one's own utility.Yet, existing agents struggle due to bounded rationality in human data, low adaptability to counterpart behavior, and limited strategic reasoning.To address this, we introduce principle-driven negotiation agents, powered by ASTRA, a novel framework for turn-level offer optimization grounded in two core principles: opponent modeling and Tit-for-Tat reciprocity.ASTRA operates in three stages: (1) interpreting counterpart behavior, (2) optimizing counteroffers via a tool-integrated action with a linear programming (LP) solver, and (3) selecting offers based on strategy assessment and the partner's acceptance probability.Through simulations and human evaluations, our agent effectively adapts to an opponent's shifting stance and achieves favorable outcomes through enhanced adaptability and strategic reasoning.Beyond enhancing negotiation performance, it also serves as a powerful coaching tool, offering interpretable strategic feedback and optimal offer recommendations beyond human bounded rationality, with its potential further validated through human evaluation.

  PublisherOA PDFDOI

• Salience Adjustment for Context-Based Emotion Recognition

  2025-05-26

  articleSenior author

  Emotion recognition in dynamic social contexts requires an understanding of the complex interaction between facial expressions and situational cues. This paper presents a salience-adjusted framework for context-aware emotion recognition with Bayesian Cue Integration (BCI) and Visual-Language Models (VLMs) to dynamically weight facial and contextual information based on the expressivity of facial cues. We evaluate this approach using human annotations and automatic emotion recognition systems in prisoner’s dilemma scenarios, which are designed to evoke emotional reactions. Our findings demonstrate that incorporating salience adjustment enhances emotion recognition performance, offering promising directions for future research to extend this framework to broader social contexts and multimodal applications.

  PublisherDOI

• PUF-ML: Machine Learning - Based Physical Unclonable Functions For Cost Effective Integration In Smart Healthcare

  2025-07-06

  article

  Physical Unclonable Functions (PUF) are hardwareassisted security (HAS) primitives that provide robust and lightweight security in Internet of Things (IoT) environments. PUF is a hardware module that can generate natural random numbers using the manufacturing variations introduced during Integrated Circuits (IC) fabrication. However, as a hardware module, it is difficult to integrate into legacy devices, or “things” deployed in the field. This paper presents a machine learningbased PUF, PUF-ML, which uses machine learning models to generate the keys at the devices. We use XOR-based PUF design to make it more resistant to machine learning attacks. The PUF can generate keys with a uniqueness of around $49 \%$ and a reliability of $98 \%$. This can help to integrate PUF in various environments, such as blockchain, where devices or “things” are already deployed in the environments.

  PublisherDOI

• Can LLMs Generate Behaviors for Embodied Virtual Agents Based on Personality Traits?

  2025-09-16 · 2 citations

  articleOpen accessSenior author
  PublisherDOI

• Can LLMs Generate Behaviors for Embodied Virtual Agents Based on Personality Traits?

  ArXiv.org · 2025-08-27

  preprintOpen accessSenior author

  This study proposes a framework that employs personality prompting with Large Language Models to generate verbal and nonverbal behaviors for virtual agents based on personality traits. Focusing on extraversion, we evaluated the system in two scenarios: negotiation and ice breaking, using both introverted and extroverted agents. In Experiment 1, we conducted agent to agent simulations and performed linguistic analysis and personality classification to assess whether the LLM generated language reflected the intended traits and whether the corresponding nonverbal behaviors varied by personality. In Experiment 2, we carried out a user study to evaluate whether these personality aligned behaviors were consistent with their intended traits and perceptible to human observers. Our results show that LLMs can generate verbal and nonverbal behaviors that align with personality traits, and that users are able to recognize these traits through the agents' behaviors. This work underscores the potential of LLMs in shaping personality aligned virtual agents.

  PublisherOA PDFDOI

• Aware Yet Biased: Investigating Emotional Reasoning and Appraisal Bias in Large Language Models

  IEEE Transactions on Affective Computing · 2025-06-19 · 4 citations

  article

  This paper reports two studies investigating the emotional reasoning of Large Language Models (LLM). Previous research has suggested that LLMs are surprisingly accurate at predicting human emotions from text descriptions of situations and reason in a way that is consistent with appraisal theory—a leading theory of emotion. Study 1 tests this claim with a large multilingual corpus (English, French, and German) of autobiographical descriptions of emotionally charged events. We confirm that GPT-4, one of the most advanced and widely studied LLMs, shows a remarkable ability to predict emotion and appraisals. We further show this ability is language-independent, with accuracy being consistent across languages and unaffected by the language of the prompt. However, GPT-4 struggles to accurately predict certain emotions (shame, fear, and irritation) and fails to understand appraisal dimensions related to control and power. We repeat the experiments with Gemini-2.0-Flash and find a remarkably similar pattern of strengths and weaknesses, although it consistently outperforms GPT-4. Study 2 examines a possible mechanism for these failures based on the idea of cognitive appraisal bias. In psychological appraisal theory, appraisal bias is the idea that people evaluate situations in biased, often unrealistic ways. By testing both models on a set of situations designed to identify appraisal bias, we find they exhibit strong—but similar—appraisal bias; for example, evaluating situations as if they were a person high in agreeableness and low in power. We further offer evidence suggesting that LLMs could be debiased by incorporating a person's personality in the prompt. This research underscores LLMs' capabilities and limitations in emotional reasoning, though highlights one mechanism underlying this limitation and suggests an approach for addressing these limits.

  PublisherDOI

• Salience Adjustment for Context-Based Emotion Recognition

  ArXiv.org · 2025-07-17

  preprintOpen accessSenior author

  Emotion recognition in dynamic social contexts requires an understanding of the complex interaction between facial expressions and situational cues. This paper presents a salience-adjusted framework for context-aware emotion recognition with Bayesian Cue Integration (BCI) and Visual-Language Models (VLMs) to dynamically weight facial and contextual information based on the expressivity of facial cues. We evaluate this approach using human annotations and automatic emotion recognition systems in prisoner's dilemma scenarios, which are designed to evoke emotional reactions. Our findings demonstrate that incorporating salience adjustment enhances emotion recognition performance, offering promising directions for future research to extend this framework to broader social contexts and multimodal applications.

  PublisherOA PDFDOI

• Social bonding between humans, animals, and robots: Dogs outperform AIBOs, their robotic replicas, as social companions

  PLoS ONE · 2025-06-03 · 6 citations

  articleOpen accessCorresponding

  In the evolving landscape of technology, robots have emerged as social companions, prompting an investigation into social bonding between humans and robots. While human-animal interactions are well-studied, human-robot interactions (HRI) remain comparatively underexplored. Ethorobotics, a field of social robotic engineering based on ecology and ethology, suggests designing companion robots modeled on animal companions, which are simpler to emulate than humans. However, it is unclear whether these robots can match the social companionship provided by their original models. This study examined social bonding between humans and AIBOs, dog-inspired companion robots, compared to real dogs. Nineteen female participants engaged in 12 affiliative interactions with dogs and AIBOs across two counter-balanced, one-month bonding phases. Social bonding was assessed through urinary oxytocin (OXT) level change over an interaction, self-reported attachment using an adapted version of the Lexington Attachment to Pets Scale, and social companionship evaluations administering the Robot-Dog Questionnaire. To examine OXT level changes and self-reported attachment by comparing the two social companions, we conducted mixed-effects model analyses and planned follow-up comparisons. Frequency comparison, binary logistic regression, and thematic analysis were performed to analyze social companionship evaluations. Results revealed significant differences between dogs and AIBOs in fostering social bonds. OXT level change increased during interactions with dogs but decreased with AIBOs. Participants reported stronger attachment to dogs and rated them as better social companions. These findings highlight the current limitations of AIBOs in fostering social bonding immediately compared to dogs. Our study contributes to the growing HRI research by demonstrating an existing gap between AIBOs and dogs as social companions. It highlights the need for further investigation to understand the complexities of social bonding with companion robots, which is essential to implement successful applications for social robots in diverse domains such as the elderly and health care, education, and entertainment.

  PublisherDOI

Recent grants

• HCC: Building Rapport with Virtual Humans

  NSF · $450k · 2007–2011

• Advancing the Use of Automated Dialogue Systems for Teaching Communication and Interpersonal Skills

  NSF · $745k · 2018–2022

• SoCS: Achieving the Interpersonal Function of Affect in Human-Machine Collaboration

  NSF · $700k · 2012–2017

• HCC: Small: Learning-by-Explaining to a Virtual Human

  NSF · $496k · 2009–2012

Frequent coauthors

• Gale Lucas

  252 shared

• Stacy Marsella

  Universidad del Noreste

  162 shared

• David Traum

  92 shared

• Louis‐Philippe Morency

  92 shared

• Celso M. de Melo

  DEVCOM Army Research Laboratory

  89 shared

• Giota Stratou

  Keysight Technologies (United States)

  83 shared

• Stefan Scherer

  META Health

  63 shared

• Johnathan Mell

  University of Central Florida

  59 shared

Education

• Ph.D., Computer Science

  University of Southern California

  1992

• M.S., Computer Science

  University of Southern California

  1989

• B.S., Computer Science

  University of Southern California

  1986

Awards & honors

• 2008 International Conference on Intelligent Virtual Agents…
• 2008 International Conference on Multimodal Interaction Best…
• 2007 Interservice/Industry Training, Simulation & Education…
• 2005 Journal of Autonomous Agents and Multi-agent Systems Be…
• 2003 International Conference on Autonomous Agents and Multi…