Enhancing documents review through Knowledge Graphs and Large Language Models

Abstract

In the era of data proliferation, efficiently navigating and extracting insights from complex document collections is paramount across various sectors. Traditional methods for document analysis are increasingly inadequate, facing challenges of interactivity, semantic interpretation, and integration. This thesis introduces a novel system that leverages the synergistic potential of Knowledge Graphs (KGs) and Large Language Models (LLMs) to radically innovate document review processes. By transforming static documents into dynamic, interactive structures, the system offers an intuitive and detailed approach to data retrieval, visualization, and analysis. Employing advanced Natural Language Processing (NLP) techniques, including KGs and LLMs like GPT, our solution facilitates semantic analysis, allowing users to interact with the content through natural language queries. This system not only overcomes the limitations of traditional methods but also enhances user experience by providing a more intuitive and engaging way to navigate and understand complex information. Our evaluation demonstrates the system's effectiveness in improving efficiency, effectiveness, and user interaction, suggesting a significant advancement in the field of document analysis and a foundation for future developments. This research contributes both at a high level, by demonstrating the application's potential in real-world scenarios, and at a low level, by detailing the technical architecture and the innovations in system design and interaction. The thesis concludes with potential future enhancements, highlighting the system's adaptability and the ongoing need for innovation in the dynamic landscape of document analysis. ​