Prompting LLMs with Knowledge Graphs for Enhanced Reasoning

As the ongoing discussion on LLMs illustrates, despite excelling in an inordinate amount of NLP tasks, they fail in practice simply because they can’t update knowledge and, as such, create misinformation while also resorting to opaque reasoning methods. This paper proposes a novel collaboration approach between KGs and LLMs that wouldn’t require any further training. The proposed approach will first include LLMs, step by step, into KGs and will take steps to extract specific knowledge subsets pertinent to the task at hand. Afterward, based on the newly extracted knowledge, the reasoning processes will be carried out, and the LLMs will illustrate which exact points were used for reasoning. This ensures more dependable knowledge-driven reasoning, allowing one to trace the steps of reasoning easily. The system resolves practical issues LLMs face by combining the best from the two worlds: that of KGs and LLMs. It is a promising way to improve the reasoning power of an LLM and increase overall knowledge-based reasoning effectiveness.