Graph-Based User-Centric Recommender System Using Neo4j, Cypher, and Jaccard Similarity in the Field of e-Commerce

This paper presents a scalable and interpretable recommender system architecture that uses a property graph model implemented in Neo4j to generate personalized product recommendations. By representing customers and products as nodes and purchases as edges, and leveraging Jaccard similarity over shared purchases, the proposed system identifies nearest neighbors and recommends items accordingly. The use of graph queries allows for explainability, flexibility and improved performance compared to traditional collaborative filtering methods. Visualization with networkx further enhances transparency. Our research contributes a novel integration of graph-based storage, similarity computation and visual analytics in the domain of recommender systems. We further propose an algorithm for building a graph-based recommendation system for e-commerce to offer personalized suggestions. It is implemented by batching relationships and using the UNWIND clause in Cypher to minimize memory and network overhead. Experimental results transaction data show that the graph-based method delivers interactive performance and high-quality recommendations, which are validated by comparing Jaccard similarity scores with results from the Surprise recommendation library.