A diverse team of 6 colleagues from different labs: Dimitrios Alivanistos (DiscoveryLab), Selene Baéz Santamaría (CLTL, Hybrid Intelligence), Michael Cochez (DiscoveryLab), Jan-Christoph Kalo (DReaMS Lab), Emile van Krieken (L&R VU), Thiviyan Thanapalasingam (INDE Lab) collaborated, participated and won the Language Model for Knowledge Base Completion (LM-KBC) competition of ISWC2022!
The team used OPENAI’s GPT-3 model, performed prompt engineering and gained insights on how large language model perform on KBC. Their approach uses few-shot prompting with only 4 examples, hand designed for each relation type. One interesting finding: Simple triple-based prompts appear to work better than natural language prompts!
Congratulations Dimitrios, Selene, Michael and the rest of the team for the win!