agent behavioral profiling

Current LLM agents lack quantifiable, psychometrically validated personality models that ensure human-comparable behavioral responses in social contexts. Method: We propose a four-stage modeling framework grounded in the Big Five Inventory (BFI), integrating semantic space alignment, scale embedding, empirical data fine-tuning, and simulated behavior generation—marking the first application of rigorous psychometric validation to LLM personality modeling. Contribution/Results: Our approach achieves high semantic consistency (0.92) in personality representation, strong correspondence with human scale responses (r = 0.87), and successfully replicates canonical associations between personality traits and risk/ethical decision-making. Critically, it establishes bidirectional interpretability between latent trait scores and observable behavioral outputs. This work advances LLMs as controllable, reproducible, and empirically verifiable tools for social science experimentation.

This study addresses a critical gap in existing large language model–based multi-agent software engineering systems, which typically differentiate agents solely by role and workflow while neglecting the impact of behavioral traits such as personality and emotion on team performance. To bridge this gap, the work introduces a psychology-driven behavioral modeling framework that integrates the Big Five personality traits, basic emotions, software engineering work styles, and task-specific roles. The authors systematically evaluate 78 agent configurations across code generation and code review tasks. Results demonstrate that heterogeneous behavioral configurations outperform the best homogeneous setups in six out of eight model–task combinations, with performance gains ranging from 7.1 to 11.3 percentage points. Although high conscientiousness or fear increases revision counts and token consumption, it does not consistently enhance performance. This research establishes behavioral diversity as a key factor in collaborative efficiency and proposes a new paradigm of mixed behavioral configurations.

This study investigates how personality traits influence agent behavior and decision-making efficacy in text-based interactive environments. To address this, we propose the PANDA framework—the first systematic approach to embedding 16 personality types derived from the Big Five model into reinforcement learning agents. PANDA jointly optimizes a differentiable personality classifier and a policy network, enabling controllable, personality-aware behavioral modeling. Our method integrates personality embeddings, policy gradient learning, and the TextWorld textual game environment, and is empirically validated across 25 diverse text games. Results demonstrate that agents with high openness achieve an average 19.3% improvement in task completion rate, confirming that personality traits significantly affect decision quality and human–AI alignment. This work establishes an interpretable, tunable, and embodied paradigm for personality-infused AI agents.

This study investigates whether large language models (LLMs) can generate multimodal agent behaviors—spanning verbal and nonverbal modalities—that are both consistent with and perceptible as reflecting specific personality traits. Method: Focusing on extraversion as a core dimension, we conduct empirical evaluations in two social scenarios—negotiation and ice-breaking—and introduce the first personality-aware prompting framework that embeds Five-Factor Inventory (BFI) trait scores into LLM inputs to jointly govern linguistic output and nonverbal actions (e.g., posture, gaze). Validation employs linguistic analysis, action-mapping modeling, and user perception experiments. Contribution/Results: Generated behaviors significantly differentiate introverted vs. extraverted tendencies (p < 0.01), and human participants reliably identify agent personality traits with high accuracy (mean 82.3%). This work establishes the first framework for personality-guided, controllable multimodal behavior generation from LLMs and empirically confirms its cross-modal consistency and social interpretability.

Current conversational agents exhibit fixed personalities after deployment, limiting their ability to adapt to users’ dynamic expectations across varying contexts. This work presents the first systematic exploration of a context-sensitive mechanism for dynamically modulating AI personality, enabling users to adjust the agent’s traits along eight dimensions in real time across informational, emotional, and evaluative tasks. Leveraging latent profile analysis to identify personality archetypes, combined with trajectory analysis and an online mixed-methods experiment, the study elucidates how user expectations form and evolve. Findings reveal significant discrepancies between initial and final personality configurations, with adjustment trajectories shaped by task context. Moreover, granting users autonomous control over personality settings enhances their perception of the AI as anthropomorphic and increases trust in the system.

This study addresses the limitations of self-report bias and high manual annotation costs in AI personality assessment by proposing the A-B-D framework, which pioneers a bottom-up quantitative approach to infer agent traits from authentic interaction behaviors. Through feature engineering and factor analysis on 340,000 trajectory records, functional and linguistic features were extracted, revealing six stable personality factors. The research uncovers a pronounced "attitude-behavior gap" in AI agents: it successfully identifies trait-level differences among models such as Kimi and demonstrates that behavioral factors exhibit negligible correlation with self-reported Big Five personality scores. These findings establish a novel paradigm for the objective evaluation of AI personality.

This work proposes a novel method to quantify changes in key behavioral traits—such as the propensity to acquire sensitive data—of AI agents during skill evolution. Behavioral traits are modeled as learnable directions in a text embedding space, and a linear model is fitted to the differences between pre- and post-modification embeddings of annotated skill updates to derive generalizable trait vectors. The approach enables evaluation of behavioral tendency shifts for arbitrary modifications via projection onto these vectors. To our knowledge, this is the first framework to support cross-agent assessment of behavioral evolution, facilitating trustworthy intermediaries in scoring skill updates based on trait alignment. Evaluated on 68 annotated instances, the method achieves 91.2% accuracy in sign classification of sensitive data acquisition propensity and a Spearman rank correlation coefficient of ρ = 0.82.

This study investigates the impact of personality prompting on task performance in large language model (LLM)-based multi-agent teams, with a particular focus on the moderating role of task structure. By manipulating personality traits such as agreeableness across three distinct task types—structured programming, open-ended research collaboration, and competitive negotiation—the authors systematically evaluate personality composition effects using a multi-agent simulation framework and cross-domain benchmarks. The findings reveal that personality cues exert minimal influence in coding tasks but significantly impair performance in open collaboration and negotiation scenarios, thereby challenging the assumption of universal efficacy for personality-based interventions. Although low agreeableness alters communication patterns, it does not affect the achievement of coding milestones. These results provide critical empirical insights for designing personality attributes in multi-agent systems.

This study addresses the unresolved question of how agent personality traits influence user trust and interaction experience. By generating virtual agents with specific personality characteristics through synthesized human voices, the research employs psychometric methods to assess user perception and human–agent synchrony. Results indicate that extraverted agents are more readily preferred by users, while the effect of users’ own personality on perception operates independently of agent personality. Furthermore, no significant effect of user–agent synchrony was observed. These findings elucidate the mechanisms through which vocal personality traits function in human–computer interaction, providing critical empirical evidence for optimizing anthropomorphic agent design.

This study addresses the limitations of current approaches that rely solely on prompting to specify Big Five personality traits, which often fail to ensure consistent personality expression in language model dialogues. Adopting an interactionist perspective, this work systematically demonstrates for the first time that personality expression emerges from the context-dependent interplay among personality settings, social roles, and expressive styles—challenging the conventional assumption that personality can be controlled through prompts alone. Through a factorial design generating English–Japanese dialogue data, combined with LLM-as-a-judge evaluation and cross-lingual comparative analysis, the study reveals that social roles significantly influence openness, expressive style predominantly shapes conscientiousness and agreeableness, and neuroticism is primarily driven by explicit personality settings. Notably, stable personality impressions can still be elicited through social roles and expressive styles even without explicit personality prompts, with highly consistent findings across both languages.