iMoonLab/DHGNN

DHGNN is a respected research artifact from IJCAI 2019 focusing on hypergraph structures—where edges connect more than two vertices—and how those structures evolve. With 283 stars and 34 forks, it has served as a benchmark in the graph neural network (GNN) research community. However, its defensibility is low (3) because it functions primarily as a frozen-in-time reference implementation rather than an active library or tool. It lacks a package-manager presence (pip/conda) and has zero current development velocity. Frontier labs (OpenAI, Google) are unlikely to target this specific niche directly, as they favor more generalizable architectures like Transformers or standard GNNs, making the frontier risk 'low'. The primary threat comes from modern, well-maintained graph frameworks like PyTorch Geometric (PyG) or Deep Graph Library (DGL), which often absorb these specific research architectures into their core libraries. For a technical investor, the value here is in the intellectual property/algorithmic approach rather than the codebase itself, which is nearing obsolescence in the face of newer hypergraph research (e.g., UltraGCN, HyGNN).