Advancements in Symbolic and Neurosymbolic Reasoning for Big Data and Knowledge Graphs

Symbolic reasoning with knowledge graphs has become a cornerstone of modern knowledge-based systems, offering a principled approach to capturing, managing, and inferring complex domain knowledge. As organizations increasingly seek transparency, explainability, and automation, logic-based formalisms such as Datalog± have gained renewed attention for their ability to combine expressiveness with tractable, rule-based reasoning. In parallel, the rise of Large Language Models (LLMs) has transformed how humans interact with information, bringing semantic fluency and adaptability, though often lacking consistency, explainability, and formal guarantees. This dissertation explores how such complementary paradigms can be effectively combined. It introduces new methodologies for scalable and expressive reasoning over knowledge graphs, including a unified treatment of Datalog± fragments and heuristic-guided inference within the VADALOG system, a state-of-the-art reasoner. Building on this foundation, it proposes a neurosymbolic framework that integrates LLMs at multiple levels of symbolic reasoning. The resulting techniques enable explainable inference, natural language interaction, and robust performance on noisy and partially structured data, with the goal of contributing to the development of AI systems that are not only expressive and adaptable, but also transparent, accountable, and grounded in logic.