Large language models (LLMs), such as ChatGPT and GPT4, are making new wavesin the field of natural language processing and artificial intelligence, due totheir emergent ability and generalizability. However, LLMs are black-boxmodels, which often fall short of capturing and accessing factual knowledge. Incontrast, Knowledge Graphs (KGs), Wikipedia and Huapu for example, arestructured knowledge models that explicitly store rich factual knowledge. KGscan enhance LLMs by providing external knowledge for inference andinterpretability. Meanwhile, KGs are difficult to construct and evolving bynature, which challenges the existing methods in KGs to generate new facts andrepresent unseen knowledge. Therefore, it is complementary to unify LLMs andKGs together and simultaneously leverage their advantages. In this article, wepresent a forward-looking roadmap for the unification of LLMs and KGs. Ourroadmap consists of three general frameworks, namely, 1) KG-enhanced LLMs,which incorporate KGs during the pre-training and inference phases of LLMs, orfor the purpose of enhancing understanding of the knowledge learned by LLMs; 2)LLM-augmented KGs, that leverage LLMs for different KG tasks such as embedding,completion, construction, graph-to-text generation, and question answering; and3) Synergized LLMs + KGs, in which LLMs and KGs play equal roles and work in amutually beneficial way to enhance both LLMs and KGs for bidirectionalreasoning driven by both data and knowledge. We review and summarize existingefforts within these three frameworks in our roadmap and pinpoint their futureresearch directions.
URL
Affiliations
Abstract
Translation (by gpt-3.5-turbo)
Summary (by gpt-3.5-turbo)