在跨機構資料整合的場景中，不同來源的資料往往存在高度潛在的關聯性，這雖帶來了整合分析的可能性，卻也衍生出隱私洩漏的風險，特別是當攻擊者掌握部分背景資訊時，便可能透過連結攻擊（link attack），推測或高度猜測出其他機構所管理的敏感資料內容，進而造成使用者資訊的間接暴露。這樣的風險使得資料無法安心流通與交換，即便已累積大量資料，也難以真正加以整合與轉化為具體價值，成為當前資料共享發展的一大瓶頸。 本研究提出一個結合區塊鏈與聯邦學習建模的風險分析平台來實現跨機構建構決策樹的流程，區塊鏈和智能合約作為管理、協助和監督的角色，利用不可更改的特性，可以追溯平台上的所有動作，各機構可以在不直接共享原始資料的前提下，實現資料的整合與分析。同時，本研究對於決策樹建立過程中可能出現的隱私威脅提出分析並設計了保護機制，以保障機敏資料的隱私性與安全性。 ;In the context of cross-institutional data integration, data from different sources often exhibit a high degree of potential correlation. While this offers possibilities for integrated analysis, it also gives rise to privacy leakage risks. Especially when an attacker possesses partial background information, they may launch link attacks to infer or closely approximate sensitive data managed by other parties, thereby potentially exposing user information indirectly. Such risks hinder the secure flow and exchange of data. Even with large volumes of collected data, effective integration and conversion into valuable insights remain challenging, constituting a major bottleneck in the advancement of data sharing today. This study proposes a risk analysis platform that combines blockchain and federated learning modeling to enable the construction of decision trees across institutions. Blockchain and smart contracts serve the roles of management, assistance, and supervision. By leveraging their immutable characteristics, all actions on the platform can be traced. Institutions can achieve data integration and analysis without directly sharing raw data. At the same time, this study analyzes potential privacy threats that may arise during the decision tree construction process and designs protection mechanisms to safeguard the privacy and security of sensitive data.
