<p>This research paper presents a novel recommendation model called light spectral graph convolutional networks (LSGCN) aimed at improving the ranking and diversity of recommended items in collaborative filtering (CF)-based Recommender Systems (RSs), particularly hybrid RSs. LSGCN is a hybrid model that combines spectral and spatial approaches, integrating simplified graph convolution and a node propagation control layer to eliminate eigen-decomposition and nonlinear activation from the previous SpectralCF design. The importance of RSs in e-commerce success and growth cannot be overstated, and conventional recommendation algorithms, including CF, tend to overlook longtail products, which account for a significant fraction of overall sales, complicating inventory management. To address this issue, the proposed LSGCN model integrates simple design and propagation control, resulting in improved spectral-based neural graph collaborative filtering for e-commerce recommendations. The proposed approach is supported by theoretical discussions, empirical experiments using three benchmark datasets, and comparison with five state-of-the-art CF models. The research results are further supported by statistical significance tests to verify the differences in performance among different CF models. Overall, the LSGCN model represents a novel contribution to the field of spectral-based and spatial CF models and e-commerce recommendations.</p>