Improved Fast-moving Consumer Goods Recommendation Algorithm Integrating Lightweight Graph Convolution

Abstract

This paper proposes an optimization algorithm that integrates lightweight graph convolutional network (GCN) and Transformer for recommending fast-moving consumer goods products in e-commerce. The method captures user-item interaction information through Lightweight GCN and optimizes the algorithm using the factorization machine recommendation mechanism. Experimental results show that, under the full settings, the model's logarithmic loss is reduced to 0.358, while the top 50 hit rate and the cumulative gain of the top 50 normalized loss are increased to 0.681 and 0.241, respectively. In experiments comparing recommendations for different fast-moving consumer goods categories and resource consumption, the model demonstrates better practicality in terms of repeat purchase hit rate, average ranking, and computational efficiency. Experimental outcomes indicate that the introduced approach can improve the performance of e-commerce fast-moving consumer goods recommendation systems and has good application prospects.