基于区块链和语义增强的科研诚信智能管控平台

Intelligent Management and Control Platform for Scientific Research Integrity Based on Blockchain and Semantic Enhancement

摘要: 针对科研诚信共享“不敢、不愿、不能共享”的问题，及现有技术在架构-算法协同、检索深度、安全管控等方面的不足，本文提出并实现基于区块链与领域自适应语义增强检索的科研诚信平台。平台立足广东省科研诚信管理实际，构建五级协同架构，通过混合存储、动态权重混合共识与精细化智能合约，保障信息共享安全高效；设计专属检索算法，融合领域预训练语言模型与动态注意力机制，实现多源异构科研诚信信息的深度关联与精准检索；依托智能预警预测模块，结合时序演化特征挖掘与科研主体关联结构解析，构建多维度失信风险预测模型，整合历史行为与主体关联特征，实现科研失信行为预警与全链条精准管控。实验验证，平台核心指标显著优于传统数据库与现有区块链平台，大规模数据场景下仍高效稳定，为科研诚信信息跨域互联互通与智能化管控提供完整解决方案。

Abstract: Addressing the issues of “unwillingness, reluctance, and inability to share” in scientific research integrity data sharing, and the shortcomings of existing technologies in architecture-algorithm collaboration, retrieval depth, and security control, this paper proposes and implements a scientific research integrity platform based on blockchain and domain-adaptive semantic-enhanced retrieval. Based on the actual scientific research integrity management practices in Guangdong Province, the platform constructs a five-level collaborative architecture. Through hybrid storage, dynamic weighted hybrid consensus, and refined smart contracts, it ensures secure and efficient information sharing. A dedicated retrieval algorithm is designed, integrating a domain-pretrained language model and a dynamic attention mechanism, to achieve deep association and precise retrieval of multi-source heterogeneous scientific research integrity information. Leveraging an intelligent early warning and prediction module, combined with temporal evolution feature mining and scientific subject association structure analysis, a multi-dimensional misconduct risk prediction model is constructed, integrating historical behavior and subject association features to achieve early warning and precise control of scientific misconduct throughout the entire chain. Experimental results show that the platform’s core indicators significantly outperform traditional databases and existing blockchain platforms, remaining efficient and stable even in large-scale data scenarios, providing a complete solution for cross-domain interconnection and intelligent management of scientific research integrity information.