DEVELOPMENT OF A PRODUCT RECOMMENDATION SYSTEM BASED ON ASSOCIATION RULES AND IMAGE-BASED PRODUCT SEARCH

Abstract

In the rapidly evolving e-commerce landscape, enhancing user experience through personalized recommendations and intelligent search is crucial. This thesis presents the design and implementation of an Online Supermarket System featuring two advanced intelligent modules. First, the system employs the FP-Growth algorithm to analyze historical transaction data, generating association rules with a minimum confidence of 0.65 and lift greater than 1.0 to suggest "frequently bought together" items in real-time. Second, a visual search engine is integrated using a pre-trained EfficientNet-B4 model combined with image preprocessing techniques to extract feature vectors, enabling accurate product retrieval via Cosine Similarity. The system is built upon a Microservices-oriented architecture utilizing the MEVN stack (MongoDB, Express, Vue.js, Node.js) for the core application and Python (Flask) for AI services. Experimental results demonstrate that the system operates stably, delivering relevant recommendations and precise image search results, thereby offering a practical solution for the Vietnamese e-commerce market.