The fine-grained behavior modeling framework consists of the subject correlation layer, behavior sequence layer, and attribute distribution layer. The subject correlation layer reflects the association between subjects and is coupled with a mapping relationship to the behavior sequence layer. There is a time-annotated edge with time information T = (t^o, t^d) between each subject and behavior, where time T indicates the starting time t^o and ending time t^d when the subject initiates the behavior. In the behavior sequence layer, different behaviors occur in a clear sequence, and we denote Δt = σ(T₁, T₂) as the time difference between two behaviors T₁ and T₂. A behavior is described in detail by several attributes, and their associations are reflected as the coupling mapping between the behavior sequence layer and the attribute distribution layer