Скачать или смотреть AI-Driven Interoperability in Public Procurement Data – Ioannis Fountoukidis

Ensuring data interoperability in public procurement is essential for transparency, efficiency, and predictive decision-making. This study presents an AI-powered approach integrating graph-based data modeling, semantic ontologies, and machine learning to enhance interoperability and analytics in procurement datasets. The system processes TED data (2018–2023) using Neo4j and the European Procurement Ontology (ePO) for structured representation. To enable natural language querying, we locally deploy Mistral-7B AI, which interacts with the Neo4j database. However, further training is needed to improve the accuracy of complex Cypher queries.

Methods

1. Knowledge Graph Construction – TED procurement data is structured in Neo4j, linking contracting authorities, tenders, lots, and economic operators.
2. Semantic Interoperability – The ePO ontology enhances the semantic understanding of procurement data.
3. AI-Driven Querying – A Mistral-7B local AI model processes natural language queries, generating Cypher queries for retrieval, though additional training is required for complex queries.
4. Predictive Analytics & Machine Learning – Graph-based algorithms (Preferential Attachment, Adamic-Adar) predict procurement trends, while machine learning refines predictions by comparing past forecasts with actual data.
5. Web-Based Interaction – A web interface provides real-time access to procurement insights.

By integrating graph databases, semantic interoperability, and AI-driven analytics, our approach improves data accessibility, predictive insights, and decision-making. Refining AI-generated Cypher queries remains a priority to ensure error-free execution. Future work will focus on cross-border interoperability and multilingual support for broader adoption in public sector analytics.

Ioannis Fountoukidis

This is a live recording of the presentation given at #ENDORSE2025. See also the PowerPoint presentation and other information here: https://link.europa.eu/39KnxB