This paper examines factors hindering data sharing and circulation and explores institutional designs to promote them. It begins by analyzing the obstacles to data sharing and circulation stemming from a lack of legal protection. It specifically presents the Data Circulation Limitation Hypothesis, which argues that contractual protection alone is insufficient to promote data circulation. The paper also addresses the misconception that granting rights to data hinders its circulation. Based on this analysis, the paper proposes institutional designs leveraging a Data Income (DI) system to promote data sharing and circulation.

Finding related tables is a key component for gaining insights from multiple datasets and for designing data-centric AI systems. The goal is to discover tables that are semantically related to a given table of interest from data repositories. Recently, data repositories have become increasingly cross-organizational and large-scale, as multiple public and private organizations continuously generate and manage tables independently, without any unified metadata (i.e., data descriptions) or centralized control. As a result, it has become increasingly challenging to discover and utilize relevant tables for analysis efficiently. In this paper, we propose a novel framework for discovering related tables over cross-organizational and large-scale data repositories. Our method leverages pretrained foundation models to extract semantic features directly from tables, enabling efficient and effective data discovery without the need for additional training or metadata. Specifically, our method (a) extracts semantic representations directly from the table itself through a pretrained large language model in a practical yet fully zero-shot manner, (b) does not require unified metadata or centralized model training on data repositories, and (c) allows online augmentation of data repositories. Experimental results on real-world datasets demonstrate that our method achieves more accurate and efficient related-table discovery compared with state-of-the-art baselines.

Modern supply chains produce diverse and heterogeneous data streams originating from enterprise systems, IoT devices, and open data sources. However, the integration and contextualization of these signals for proactive risk management remains a persistent challenge. This study proposes a metadata?driven semantic integration framework that unifies raw data and contextual metadata within a shared knowledge base to facilitate early risk detection and explainable decision making. The framework aligns heterogeneous data?including Enterprise Resource Planning (ERP) and Supply Chain Management (SCM) logs, IoT telemetry, and external open data?through a common semantic representation. A graph- augmented large language model (LLM) agent framework was employed to integrate multi-source data retrieval and propagation analysis, leveraging a knowledge graph to support context-aware risk inference. The effectiveness of the framework is demonstrated through three representative use cases: (1) detecting supplier disruptions using open data, (2) identifying the propagation of delivery delays across ERP and SCM data, and (3) uncovering compliance and quality anomalies through linked manufacturing metadata.

Conventional automatic semantic segmentation achieves reasonable accuracy but often fails to reflect user intentions or task-specific goals. In practical applications, regions of interest vary depending on the purpose or domain, making uniform segmentation insufficient. To address this, we propose a user-oriented system that allows intuitive input of user preferences and generates training data for neural network-based adaptive segmentation. The system enables high-precision segmentation that aligns with user-defined guidance. Its versatility allows application across diverse fields, including safety-critical areas and domains requiring improved efficiency, demonstrating both technological and societal significance.

Distributed data use that links datasets across fields and organizations has drawn attention as a source of innovation. However, evidence remains limited regarding which factors promote such data use on co-creation platforms. Using large-scale archival data from Kaggle, one of the world's largest data analysis platforms, this study employs multiple regression analysis to examine how data characteristics and data owner characteristics affect the degree of data use. The results show that peer evaluations (upvotes) and explicit owner rankings are positively associated with both data viewing and downstream use. These findings offer practical implications for the governance of data platforms and for designing policies that effectively foster data use.