Estimating Individual Personality Traits from Emotion Expression Patterns in Japanese Text

​

This study aims to establish a method for quantitatively extracting individual personality traits from emotion expression patterns in Japanese text. We focus on a psychological phenomenon widely observed in Japanese speakers — the unconscious adjustment of self-presentation according to interlocutor and context— and develop a computational framework to capture it. While prior work on text-based personality estimation typically collapses emotion into a single dominant label, we preserve Plutchik's eight emotions as a continuous probability distribution vector, extracting fine-grained individual differences from emotion gradation. We first validate our methodology on the WRIME corpus as a sandbox, then extend the approach to real-world data with recipient information , aiming to quantify recipient-specific personas and visualize latent personality traits that self-report inventories cannot reach.