TenderKG dataset opens AI research on bidder recommendation

Public procurement is a massive economic engine, but it remains overlooked by recommender systems due to a lack of structured datasets. TenderKG, introduced by researchers at IMT Atlantique and Lab-STICC, fills that gap with a large-scale knowledge graph covering French procurement activity from 2021 through 2023. The graph maps heterogeneous entities—companies, tenders, lots, and domain-specific taxonomies—connected by rich semantic and structural relationships. Notably, only winning bidders are observable in public records, creating a highly sparse signal for learning. To tackle this, TenderKG bundles extensive side information on market actors and tenders, including textual descriptions, hierarchical classifications, and geographical features, giving researchers the context needed to model competitive dynamics.

This dataset is purpose-built for knowledge-aware recommendation in constrained, competitive environments. It opens new research paths in bidder recommendation, knowledge graph-based inference, and competition-aware matching, where models must reason about which companies are likely to win—not just which are similar to past winners. The authors provide detailed statistics and sparsity analyses, making TenderKG a ready-to-use benchmark for real-world, high-stakes decision-making scenarios. For AI teams working on recommendation systems or public-sector analytics, TenderKG offers a fresh, realistic testbed that goes beyond typical e-commerce or media datasets, demanding deeper reasoning about market structure and competition.