WEB-BASED PRODUCT RECOMMENDATION SYSTEM FOR MSMEs IN KENDARI CITY USING NEURAL COLLABORATIVE FILTERING AND K-NEAREST NEIGHBOR

Abstract

The rapid growth of e-commerce has created significant opportunities for MSMEs in Kendari City to expand their market reach digitally; however, product visibility and service personalization remain suboptimal. This study aims to design and develop a web-based product recommendation system for MSMEs by implementing Neural Collaborative Filtering (NCF) and K-Nearest Neighbor (KNN) algorithms through a switching mechanism. KNN is used in cold-start conditions for new users based on product similarity, while NCF is applied when interaction data are available to model the non-linear relationships between users and products. The system was developed using the Rational Unified Process (RUP) method with Python and Flask technologies. Evaluation was conducted through algorithm performance testing using Top-N metrics, User Acceptance Testing (UAT), and white-box testing. The results show that NCF achieved an HR@10 of 0.7692 and an NDCG@10 of 0.4691, while KNN obtained a Precision@10 of 0.9500. UAT produced a score of 82.67%, categorized as very good, and white-box testing indicated that the logical flow of KNN and NCF operated as designed.