Unifying LLMs & Knowledge Graphs for GenAI

Knowledge Graphs (KGs)

Knowledge Graphs are graph-based data strucures that model factual knowledge by interconnecting entities and their relationships. In brief, entities are represented as nodes in the graph strucure that represent real-world objects, concepts, or events, while edges denote the relationships between them.

The Resource Description Framework (RDF) enables KGs to represent information in a structured, triple format (subject-predicate-object), making it universally interoperable and semantically rich. This structure allows for detailed representation of real-world phenomena, facilitating precise querying, reasoning, and inference. By leveraging the RDF model, KGs can effectively encode, exchange, and integrate knowledge across diverse systems, enabling advanced reasoning capabilities and supporting the creation of intelligent applications that can understand and interact with the world in more human-like ways.

Unification of Large Language Models (LLMs) and KGs

The unification of LLMs and KGs seeks to synergize the deep, contextual understanding capabilities of LLMs with the structured, factual knowledge encapsulated in KGs. This integration aims to address the limitations of each approach independently, enhancing the LLMs’ ability to generate accurate, reliable, and contextually informed outputs by leveraging the structured knowledge from KGs. Such a unification not only improves factual accuracy and consistency in LLM-generated content but also enriches the interpretability and adaptability of these models across diverse applications.

Our Research

The core mission of this research is to harness the factual knowledge from Knowledge Graphs to enhance the capabilities of Large Language Models. By integrating the structured, verified information contained within KGs, this research project aims to inform and guide the decision-making processes of LLMs, ensuring that their outputs are not only contextually relevant but also factually accurate. This synergy is intended to overcome the limitations of LLMs in handling factual information, thereby significantly improving their utility in real-world applications.

Current Research

Our research contributes to the development of a Linked Data graph utilizing the W3C RDF model. This graph archives proprietary knowledge from the University of Nicosia, including faculty research, course content, and more. This structured knowledge graph is designed to inform decisions and enhance the factual accuracy of Large Language Models, bridging the gap between dynamic, contextual language understanding and static, factual knowledge bases. The project aims to develop a web-based system to democratize access to this comprehensive knowledge base, fostering an environment of open learning and innovation. The system will be accessible to faculty, staff, students, and external paid users, promoting a culture of shared knowledge and educational growth.