Journal of Engineering, Project, and Production Management, 2026, 16(7), 2025-309
DLAGNN: A Dual-Layer Attention Graph Neural Network Framework with Dynamic Feature Fusion and Multi-Task Learning for Customer Relationship Management in E-Commerce
Full-time Faculty, School of Artificial Intelligence, Chongqing Jianzhu College, Chongqing,400072, China, E-mail: y.yjunyi@outlook.com
Project Management
Received December 8, 2025; revised January 22, 2026; accepted July 2, 2026
Available online July 20, 2026
Abstract: The customer-centric transformation of e-commerce makes it difficult for traditional customer relationship management to meet the needs of enterprises to accurately tap into customer value. Therefore, the research aims to build an intelligent model that integrates multi-source data, accurately predicts customer needs, and guides personalized marketing. The study utilizes the Alibaba Tianchi dataset, comprising 100,000 user interaction logs, to validate the proposed Dual-Layer Attention Graph Neural Network (DLAGNN) framework. Distinct labels were constructed for multi-task learning: customer churn was defined by a 30-day inactivity threshold, while Customer Lifetime Value (CLV) was calculated from historical transaction amounts. The framework employs a dual-layer attention mechanism to capture long-term and short-term preferences, respectively, integrates dynamic features such as timeliness and profitability, and simultaneously optimizes preference prediction, churn warning, and CLV prediction. Results indicate that the model converged efficiently and achieved superior predictive accuracy compared to baseline models such as GNN-SR and TA-RNN. Its customer churn warning could be issued significantly in advance of the actual event, extending the retention window. In the simulation test, the Return on Investment (ROI) in churn intervention and CLV improvement demonstrated the model’s practical effectiveness. This model effectively solves data sparsity and model commercial adaptability, providing a feasible path for applying deep learning in Customer Relationship Management (CRM).
Keywords: E-commerce personalized marketing; customer relationship management; data mining; graph neural network; dynamic feature fusion.
Copyright © Journal of Engineering, Project, and Production Management (EPPM-Journal).
This work is licensed under a Creative Commons Attribution-NonCommercial-NoDerivs 3.0 Unported License.
Requests for reprints and permissions at eppm.journal@gmail.com.
Citation: Yang, J. (2026). DLAGNN: A Dual-Layer Attention Graph Neural Network Framework with Dynamic Feature Fusion and Multi-Task Learning for Customer Relationship Management in E-Commerce. Journal of Engineering, Project, and Production Management, 16(7), 2025-309.
