"NSF Scholar Heterogeneous Hypergraph Dataset"

"We present a preprocessed heterogeneous hypergraph dataset constructed from publicly available National Science Foundation (NSF) award and publication records, designed to support research on scholar collaboration recommendation and academic team formation. The dataset comprises 5,099 scholar nodes and 5,631 award hyperedges, augmented with publication co-authorship, institutional affiliation, research keyword, and venue hyperedges, yielding a rich multi-relational graph structure with five distinct edge types. Each scholar node is associated with a JSON profile encoding award history, publication records, research keywords, and institutional metadata. The hypergraph is released in a processed format, including the node-to-index mapping, edge lists with typed relations, and pre-split training, validation, and test pairs derived from award team membership. This dataset directly supports two complementary tasks: (1) individual scholar collaboration recommendation, framed as link prediction over the heterogeneous hypergraph; and (2) query-guided academic team formation, where a multi-dimensional query is used to assemble a team of scholars satisfying skill, venue, program, and collaboration diversity constraints. To our knowledge, this is the first publicly available heterogeneous hypergraph benchmark constructed from NSF award data for these tasks. The dataset is intended to facilitate reproducible evaluation of graph neural network, knowledge graph embedding, and team formation algorithms in the academic domain."