A Big Data-Driven Method for Animation Character Topology Correction and Generation Based on Autoregressive Models and Dynamic Graph Convolutional Neural Networks

The growing demand for high-fidelity digital content has increased the need for efficient and automated topology optimization methods in three-dimensional animation character generation. To overcome the limitations of manual topology correction and low-efficiency modeling workflows, this study proposes a big-data-driven framework integrating autoregressive models and dynamic graph convolutional neural networks. A large-scale three-dimensional animation dataset is first established to learn global vertex-sequence distributions and local topological evolution patterns. Autoregressive modeling is employed to capture global structural characteristics, while dynamic graph convolutional networks optimize local topology adaptively. A unified topology-correction and morphology-generation framework is subsequently developed to improve structural rationality and generation diversity. Experimental results demonstrate superior performance in topology correction accuracy, generation diversity, and computational efficiency compared with conventional approaches. The proposed framework provides methodological references for geometric representation learning, graph-based information processing, and intelligent digital-content generation.