In human-centered Industry, human–robot collaboration (HRC) has emerged as a key research focus within the broader field of flexible manufacturing. As a core aspect of HRC, human–robot collaborative assembly (HRCA) has garnered significant attention due to its flexible manufacturing capabilities. These capabilities are integrated with deep learning techniques like Graph Neural Networks (GNNs) through Heterogeneous Graphs (HetG), as exemplified in human–robot collaborative assembly task allocation (HRCATA). Despite this, existing HetG datasets lack a suitable multi-objective optimization dataset as a foundational input for effective assembly task allocation. Therefore, this paper proposes a dataset model based on HetG that incorporates multiple optimization node- and edge features to tackle the HRCATA challenge. Furthermore, a heterogeneous graph dataset specific to an HRCA scenario involving new energy vehicle battery assembly is constructed to validate and assess the effectiveness of this approach.