Abstract
<jats:p>The objective of this study is to identify and evaluate resource-efficient implementation methodologies for personalized recommendation systems in Small and Medium-sized E-Commerce Enterprises (SMEs). The research addresses key challenges faced by these enterprises, including data sparsity and limited computational resources. The study employs a staged, hybrid implementation architecture. In the initial phase, popularity-based methods and Association Rule Mining (ARM) are applied to mitigate the cold-start problem. The main recommendation mechanism relies on an optimized Hybrid Collaborative Filtering (HCF) model based on implicit feedback, utilizing lightweight frameworks such as LightFM. Evaluation is conducted through A/B testing tailored for low-traffic conditions. The results indicate that HCF models enriched with item metadata demonstrate higher hit rates under sparse data conditions and positively influence Average Order Value (AOV) and Conversion Rate (CR). Keywords: recommendation systems, e-commerce, hybrid filtering, data sparsity, cold-start</jats:p>