Temporal Dynamics of Emotion in Magnetoencephalography: From Dataset Validation to Quantitative Trajectory Analysis

Abstract

While emotion is fundamentally time-varying process, the dominant paradigm in brain-based emotion recognition has remained focused on static classification. This focus on single, aggregate labels, rather than on the unfolding temporal dynamics, has persisted due to two primary challenges: (1) a scarcity of high-resolution, richly labeled neural datasets, and (2) a lack of analytical frameworks to quantify the properties of emotional trajectories. This thesis addresses both challenges through a sequential, two-part study. Study 1 addressed the dataset challenge by constructing and validating a novel, large-scale Magnetoencephalography (MEG) dataset. This dataset was built using culturally optimized video stimuli for Korean young adults and employed a comprehensive assessment method combining dimensional (SAM) and fine-grained (PrEmo) labels. Validation results confirmed the dataset’s high quality: robust four-class neural classification (89.87% accuracy) was achieved, and critically, we found evidence of a brain-behavior correspondence: videos that elicited ambiguous SAM behavioral responses correspondingly yielded lower classification accuracy. Study 2 built upon this validated dataset to address the analysis challenge. The study proposed and validated a novel set of trajectory-based indices designed to quantify the dynamic properties of emotion estimation trajectories derived from the MEG data. The findings demonstrate that these temporal indices, which quantify dynamic properties such as stability and variability, successfully differentiated not only the broad emotional categories but also the more fine-grained emotional differences within those categories-distinctions missed by traditional classification. By first establishing a foundational dataset and then validating a new dynamic analysis method upon it, this thesis provides a validated pathway for enriching emotion recognition, introducing temporal trajectory indices as a new analytical dimension. This work establishes a robust methodological groundwork for the detailed and temporally precise assessment and quantification of human affective dynamics.