The paper discusses the importance of augmenting large language models (LLMs) with user-specific knowledge for real-world applications like personal AI assistants. The authors focus on prompt-to-triple (P2T) generation, exploring methods like zero-shot prompting, few-shot prompting, and fine-tuning. They assess these methods using a specialized synthetic dataset. The paper emphasizes the role of knowledge graphs in this process, given their clear structures and capacity for factual reasoning. The authors aim to enhance the interactive capabilities of LLMs by improving their ability to learn from user-provided information.
Publication date: 1 Feb 2024
Project Page: https://github.com/HaltiaAI/paper-PTSKC
Paper: https://arxiv.org/pdf/2402.00414
