Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

LLM-MC-Affect: LLM-Based Monte Carlo Modeling of Affective Trajectories and Latent Ambiguity for Interpersonal Dynamic Insight

Yu-Zheng Lin, Bono Po-Jen Shih, John Paul Martin Encinas, Elizabeth Victoria Abraham Achom, Karan Himanshu Patel, Jesus Horacio Pacheco, Sicong Shao, Jyotikrishna Dass, Soheil Salehi, Pratik Satam

TL;DR

This work treats utterance-level emotion as a latent distribution $p(e|u_t)$ rather than a deterministic label, and uses stochastic LLM decoding with Monte Carlo sampling to construct uncertainty-aware sentiment trajectories. By mapping these trajectories to standardized scores and analyzing them with sequential cross-correlation and slope indicators, the framework reveals dyadic interpersonal dynamics such as leading/lagging influence and affective synchronization in teacher–student dialogues. The approach is validated on simulated educational interactions, showing robustness to decoding temperature and model variations, and yielding interpretable interaction patterns (e.g., teacher-led positive contagion) without biometric data. This probabilistic, scalable pipeline offers a generalizable tool for affect-aware instructional analytics and broader social-behavioral research.

Abstract

Emotional coordination is a core property of human interaction that shapes how relational meaning is constructed in real time. While text-based affect inference has become increasingly feasible, prior approaches often treat sentiment as a deterministic point estimate for individual speakers, failing to capture the inherent subjectivity, latent ambiguity, and sequential coupling found in mutual exchanges. We introduce LLM-MC-Affect, a probabilistic framework that characterizes emotion not as a static label, but as a continuous latent probability distribution defined over an affective space. By leveraging stochastic LLM decoding and Monte Carlo estimation, the methodology approximates these distributions to derive high-fidelity sentiment trajectories that explicitly quantify both central affective tendencies and perceptual ambiguity. These trajectories enable a structured analysis of interpersonal coupling through sequential cross-correlation and slope-based indicators, identifying leading or lagging influences between interlocutors. To validate the interpretive capacity of this approach, we utilize teacher-student instructional dialogues as a representative case study, where our quantitative indicators successfully distill high-level interaction insights such as effective scaffolding. This work establishes a scalable and deployable pathway for understanding interpersonal dynamics, offering a generalizable solution that extends beyond education to broader social and behavioral research.

LLM-MC-Affect: LLM-Based Monte Carlo Modeling of Affective Trajectories and Latent Ambiguity for Interpersonal Dynamic Insight

TL;DR

This work treats utterance-level emotion as a latent distribution rather than a deterministic label, and uses stochastic LLM decoding with Monte Carlo sampling to construct uncertainty-aware sentiment trajectories. By mapping these trajectories to standardized scores and analyzing them with sequential cross-correlation and slope indicators, the framework reveals dyadic interpersonal dynamics such as leading/lagging influence and affective synchronization in teacher–student dialogues. The approach is validated on simulated educational interactions, showing robustness to decoding temperature and model variations, and yielding interpretable interaction patterns (e.g., teacher-led positive contagion) without biometric data. This probabilistic, scalable pipeline offers a generalizable tool for affect-aware instructional analytics and broader social-behavioral research.

Abstract

Emotional coordination is a core property of human interaction that shapes how relational meaning is constructed in real time. While text-based affect inference has become increasingly feasible, prior approaches often treat sentiment as a deterministic point estimate for individual speakers, failing to capture the inherent subjectivity, latent ambiguity, and sequential coupling found in mutual exchanges. We introduce LLM-MC-Affect, a probabilistic framework that characterizes emotion not as a static label, but as a continuous latent probability distribution defined over an affective space. By leveraging stochastic LLM decoding and Monte Carlo estimation, the methodology approximates these distributions to derive high-fidelity sentiment trajectories that explicitly quantify both central affective tendencies and perceptual ambiguity. These trajectories enable a structured analysis of interpersonal coupling through sequential cross-correlation and slope-based indicators, identifying leading or lagging influences between interlocutors. To validate the interpretive capacity of this approach, we utilize teacher-student instructional dialogues as a representative case study, where our quantitative indicators successfully distill high-level interaction insights such as effective scaffolding. This work establishes a scalable and deployable pathway for understanding interpersonal dynamics, offering a generalizable solution that extends beyond education to broader social and behavioral research.
Paper Structure (29 sections, 14 figures, 7 tables)

This paper contains 29 sections, 14 figures, 7 tables.

Table of Contents

  1. Introduction
  2. Background
  3. Statistical Formulation of Affective Ambiguity
  4. Emotion as a Temporally Coupled Interpersonal System
  5. Methodology
  6. LLM-Based Affective Trajectory Modeling
  7. Probabilistic Affective Perception: Monte Carlo Estimation via Stochastic LLM Decoding
  8. Estimation of Mean Affective State and Ambiguity
  9. Sentiment Scoring Mechanism, Polarity Mapping, and Trajectory Modeling
  10. Interpersonal Affective Dynamics and Interaction Modeling
  11. Sequential Cross-Correlation Analysis
  12. Slope-Based Development Indicator
  13. From Interpersonal Affective Trajectories to Interaction Insights: Mining Dynamics in Educational Settings
  14. Experiment results and Discussion
  15. Experimental Setup
  16. ...and 14 more sections

Figures (14)

  • Figure 1: Architectural workflow of LLM-MC-Affect for distilling interaction insights. (a) Input Contextualization integrates natural dialogue with a psychometric prompt. (b) Probabilistic Affective Perception approximates the latent affective distribution through stochastic LLM decoding and Monte Carlo estimation. (c) Statistical Estimation derives the central affective states (mean) and quantified ambiguity (variance) from the resulting probability density function. (d) Affective Trajectory represents the continuous longitudinal evolution of sentiment for both interlocutors. (e) Interpersonal Dynamics Analysis synthesizes sequential lag, cross-correlation, and affective trends to extract high-level Interaction Insights into the dyadic relationship.
  • Figure 2:
  • Figure 3:
  • Figure 4:
  • Figure 6:
  • ...and 9 more figures