CA24121 - Knowledge Graphs in the Era of Large Language Models (KGELL)

Knowledge Graphs (KGs) have gained attention due to their ability to represent structured and interlinked information. KGs represent knowledge in the form of relations between entities, referred to as facts, typically grounded in formal ontological models. Such machine-readable formats enable AI systems to make decisions using clear and verifiable data. Consequently, KGs have become essential elements in web search engines, recommendation systems, etc. Large Language Models (LLMs) have revolutionized the landscape of AI and are widely utilized in various NLP tasks such as natural language understanding, question answering, etc. Despite their remarkable performance, LLMs suffer from some significant drawbacks. First, they are trained on general-purpose data and have lower performance in domain-specific tasks and low- resource languages. Secondly, they often reflect societal biases present in training data, which can result in biased outcomes. Third, LLMs sometimes produce inaccurate or made-up information, termed “hallucinations”. Finally, understanding the decision-making process of LLMs is challenging and their outputs may lack consistency. A potential solution to all these problems is to integrate LLMs with KGs, since KGs can provide factual information and the ability to perform reasoning. This would boost the LLM’s domain-specific reasoning, and interpretability, and mitigate biases and hallucinations. A notable challenge with KGs is their requirement for frequent updates, usually performed by processing and integrating information from vast textual datasets, LLMs can aid in generating and refining KGs. Therefore, combining LLMs and KGs offers a promising opportunity to advance both technologies and represents a pivotal challenge in the contemporary research landscape.