Creative Commons

This dissertation, presented as a part of the MSc in Data Science at the International Hellenic University, embarks on an exploration of graph-based machine learning. The study begins by delving into the foundational aspects of graph theory, tracing its evolution and underlining its critical role in the realm of complex data representation and analysis. It highlights the capabilities of graphs in capturing relationships and structures in data, which traditional data representation methods cannot encapsulate effectively. The discourse transitions to machine learning, elucidating its principles and the groundbreaking advent of deep learning, emphasizing how these technologies have revolutionized data interpretation and prediction. The introduction of Graph Neural Networks marks a pivotal point in the dissertation, elaborating on their architecture, functional mechanisms, and evolutionary trajectory within the research community. This segment provides an in-depth analysis of various GNN architectures and their comparative strengths, offering a comprehensive understanding of how these models adapt and learn from graphstructured data. A substantial segment of the research is dedicated to the practical application of recommender system and question-answering demonstrating the implementation of these applications using Neo4j and LangChain, while providing a thorough exposition of the methodologies, processes, and tools employed with the purpose to showcase real-world scenarios and to demonstrate the efficacy of these technologies. This part of the study is particularly insightful, offering a glimpse into how graph-based machine learning can be effectively utilized in practical, commercial, and research-based settings.