Understanding the transport mechanisms of shale oil in kerogen nanopores is critical for optimizing recovery from organic-rich reservoirs. Here, molecular dynamics simulations were employed to investigate the spatial distribution and diffusion behavior of various hydrocarbon fractions. Results revealed a distinct stratified distribution within the pores; asphaltenes and resins primarily occupied the interface region (7.49 nm in width), while light oil dominated the central phase (5.09 nm). Quantitatively, asphaltenes contributed 52.6% of the total contact with the kerogen matrix, significantly exceeding the 5.3% contribution from light oil. This intense interfacial interaction, characterized by π-π stacking and van der Waals forces, resulted in substantial mobility suppression, with asphaltene self-diffusion coefficients being the lowest among all fractions. Specifically, asphaltene motion was confined to localized thermal vibrations near pore walls, whereas light oil exhibited superior pore-axis diffusivity. A Quantitative Structure-Activity Relationship (QSAR) model developed via a Neural Network Algorithm effectively predicted these diffusion disparities (R2 = 0.95), identifying molecular mass and electrostatic potential as governing descriptors. The wall adsorption-molecular aggregation dual mechanism elucidated here provides a molecular-scale rationale for the selective retention of heavy fractions and offers a theoretical framework for enhancing shale oil mobility through structural targeting.