Best Paper Award for ConceptFormer at GLOW WWW 2026
Congratulations to Joel Barmettler, Prof. Dr. Abraham Bernstein, and Dr. Luca Rossetto on receiving the Best Paper Award at GLOW WWW 2026 for their paper “ConceptFormer: Towards Graph-Native Grounding of Large Language Models via Latent Concept Injection”!
Retrieval-augmented generation (RAG) methodologies are often used to ground the responses of large language models (LLMs), with knowledge graphs such as Wikidata serving as popular information sources. However, using traditional graph RAG methods to incorporate highly connected nodes can quickly lead to heavy token consumption and fill up the available context.
ConceptFormer addresses this challenge by learning fixed-length concepts that represent the 1-hop neighborhoods of nodes. This drastically reduces token consumption while matching the performance of traditional methodologies.
ConceptFormer originated as a Master’s thesis at the University of Zurich. The resulting paper by Joel Barmettler, Prof. Dr. Abraham Bernstein, and Dr. Luca Rossetto was presented at the 1st Workshop on Graph-enhanced LLMs for trustworthy Web data management (GLOW), held as part of The ACM Web Conference (WWW 2026), where it received the Best Paper Award.
The paper “ConceptFormer: Towards Graph-Native Grounding of Large Language Models via Latent Concept Injection” is available online.