Explainable Neural Collaborative Filtering with Customer Segmentation for Business Intelligence in E-Commerce

Abstract

Personalized recommendation systems are important for large-scale e-commerce platforms, but the limited interpretability of deep learning models remains a challenge for Business Intelligence (BI) applications. This study proposes an Explainable Neural Collaborative Filtering framework integrating Neural Matrix Factorization (NeuMF) with post-hoc customer segmentation to provide accurate recommendations and actionable BI insights. Experiments on the Amazon Electronics dataset involving 192,403 users, 63,001 items, and 1,689,188 interactions show that the proposed model achieved HR@10 of 0.5373, NDCG@10 of 0.3439, and validation AUC of 0.8649, outperforming the strongest baseline significantly (p < 0.001). PCA and MiniBatchKMeans identified two latent behavioral segments suitable for targeted marketing and customer profiling. Results demonstrate that explainable neural recommendation systems can balance predictive performance and managerial interpretability for enterprise BI applications.