## HyperJoin: A Novel Approach for Joinable Table Discovery Effective management of data lakes requires the accurate discovery of tables that can be joined to obtain more complete information. A new study introduces HyperJoin, an innovative framework that utilizes advanced language models (LLMs) and a hypergraph representation to address this challenge. HyperJoin overcomes the limitations of existing approaches by modeling tables as hypergraphs, capturing structural interactions both within the tables themselves and between different tables. This approach allows the problem of joinable table discovery to be formulated as a link prediction problem within the graph. ## Architecture and Functionality The HyperJoin framework is based on a hierarchical interaction network (HIN) that learns expressive column representations through bidirectional message passing between columns and hyperedges. To improve the coherence of the results, the system adopts a reranking module that leverages a maximum spanning tree algorithm to eliminate noisy connections and maximize overall coherence. Experimental results demonstrate the superiority of HyperJoin compared to baseline solutions, with average improvements of 21.4% in precision and 17.2% in recall.