AVI-PERSONALITY: A TRAIT-ACTIVATED MULTIMODAL DATASET FOR PERSONALITY AND COMPETENCY ASSESSMENT IN ASYNCHRONOUS VIDEO INTERVIEWS

Asynchronous Video Interviews (AVIs) are increasingly common in recruitment, yet existing datasets for AI-based personality assessment often rely on task-free social media videos and crowdsourced labels that lack construct validity. This paper introduces AVI-Personality, a multimodal dataset designed to address these limitations by grounding personality and competency assessment in structured, theory-driven interview tasks.

The dataset includes 3,876 videos from 646 participants who completed a simulated management traineeship application. The interview structure is guided by Trait Activation Theory (TAT), featuring two generic questions and four personality-targeted questions designed to elicit specific behavioral expressions. The researchers collected both self-reported HEXACO personality traits and expert-rated observer reports, the latter provided by trained psychologists and recruiters using Behaviorally Anchored Rating Scales (BARS). The study validates the dataset through rigorous psychometric testing, including inter-rater reliability, construct validity, and demographic fairness analyses.

Validation results indicate that observer-rated personality traits have moderate to high reliability, with trait-activating questions yielding more diagnostic behavioral cues than generic questions. Notably, observer-rated personality traits showed significantly stronger predictive validity for job-related competencies and overall interview performance compared to self-reported personality scores. A comprehensive benchmark of AI models reveals that while text-based methods provide strong personality-relevant cues, multimodal methods achieve the best overall performance, though they only modestly outperform strong text-based baselines.

This dataset moves the field of automated personality recognition beyond superficial first impressions. By providing a standardized, psychometrically grounded framework, AVI-Personality enables researchers to develop and evaluate AI models that are more relevant to real-world recruitment, where the goal is to assess stable individual differences and work-related competencies rather than transient behavioral signals.