Digital object systems face persistent challenges in standardization, data generation, and performance evaluation, as the evolution of Internet architecture demands efficient information management frameworks. Traditional approaches struggle with insufficient available objects, a lack of unified metadata standards, small sample datasets, and complex event relationship modeling. Recent research published in Procedia Computer Science introduces comprehensive frameworks that integrate international metadata standards with machine learning algorithms to fundamentally transform digital object generation through Bayesian network optimization and data diffusion strategies, validated on multi-category event datasets.
The analytical foundation addresses fundamental limitations through a unified digital object generation framework combining structured and unstructured metadata extraction. By analyzing event characteristics, temporal relationships, and causal dependencies through hierarchical classification methods, the framework enables accurate information extraction from text documents. The system implements rule-driven extraction strategies for simple attributes and refined algorithms for complex elements, including keywords and themes, providing a standardized seven-element metadata structure encompassing title, author, keywords, subject, description, publisher, date, and language throughout digital object lifecycle management.
The optimization architecture leverages the K2 algorithm-based Bayesian networks, combining expert knowledge with data-driven structure learning to achieve probabilistic event modeling. By addressing small sample limitations and prediction accuracy requirements inherent in traditional approaches, the system implements improved conditional probability distributions through expert experience integration. Optimization through a two-stage training methodology incorporating causal relationship refinement prevents model errors while maintaining forecasting reliability across diverse scenarios. Experimental validation demonstrates 84% prediction accuracy improvement from a 55.21% baseline, with the network structure achieving high consistency and coverage rates across training-test partitions, validated through 100-iteration robustness testing on a 550-sample event window dataset.
Practical implementations span both data generation efficiency and computational performance. The intelligent digital object system enables automated generation, achieving 0.0013 seconds per object through optimized metadata extraction and batch packaging mechanisms. Data diffusion strategy combining LDA topic modeling with Monte Carlo sampling generates standards-compliant synthetic datasets addressing insufficient test data challenges.
Contributing to this research is Yixian Jiang, who holds a Master’s degree in Information Technology from Carnegie Mellon University’s Information Networking Institute and a Bachelor’s degree in Software Engineering from South China Agricultural University. Jiang’s professional experience spans Apple, Meta, and NVIDIA, where Jiang led large-scale machine learning infrastructure projects and intelligent system optimizations. Jiang’s technical expertise bridges academic research and enterprise-scale engineering, exemplifying how theoretical advancements in Bayesian modeling can inform practical automation systems.
This body of work represents a major step forward in bridging theoretical optimization with real-world intelligent system implementation. By integrating metadata standardization with Bayesian network algorithms for digital object systems and by advancing ML automation platforms that manage hundreds of enterprise models, Jiang’s research establishes new benchmarks for data management efficiency. These frameworks demonstrate far-reaching implications for intelligent information systems, scalable automation, and the optimization of machine learning infrastructures in increasingly complex distributed computing environments.