Nathan Vandemoortele and Bram Steenwinckel have been awarded for their best performing solution in the Table Metadata to Knowledge Graph competition within Semantic Web Challenge on Tabular Data to Knowledge Graph Matching challenge at the the 23st International Semantic Web Conference held in Baltimore, USA.
Within this competition, teams were asked to match table metadata, e.g., column names, to knowledge graphs without any access to table data and content. This is a challenging task due to the limited available context that could be used by annotation systems to perform the semantic linking.
Nathan and Bram proposed a solution that employs a Retrieval-Augmented Generation (RAG) framework to conduct an initial broad search for potential matches, ensuring scalability. Matches were further refined using a Large Language Model (LLM) incorporating advanced prompting techniques such as Chain-of-Thought (CoT) and Self-Consistency (SC). The results are combined using Reciprocal Rank Fusion (RRF) to generate a final ranked list of matches More information can be found in their paper
Despite challenges like computational cost and reliance on high-quality metadata, this solution demonstrated significant improvements over traditional methods and offers a scalable alternative to human annotation.
