Home / Current Issue / Paper 1716019
E-Commerce Cart Prediction Using Multi-Task Learning with BI-LSTM and Graph Neural Networks
Subject area: Science,Engineering and Technology · Area of research: Data Science
DOI: https://doi.org/10.64388/IREV9I10-1716019
Abstract
Shopping cart abandonment is a persistent and financially significant challenge in digital commerce, with reported global rates exceeding 68% and cumulative revenue losses running into the hundreds of billions annually. To tackle this, the present work proposes a unified multi-task learning framework capable of concurrently solving three interrelated problems: identifying abandonment risk, recommending the next likely product, and forecasting the expected transaction value. Instead of deploying isolated models for each objective, the system derives a shared latent representation from user clickstream logs and behavioural interaction histories. Temporal dynamics within sessions are encoded using a Bidirectional Long Short-Term Memory (Bi-LSTM) network, while structural dependencies across products and categories are modelled via a Graph Neural Network (GNN). The framework operates with low latency, supporting real-time deployment of targeted retention strategies before users disengage. To enhance model transparency, SHAP and LIME attribution methods are embedded as interpretability tools accessible to operational stakeholders. Empirical validation on two publicly available datasets demonstrates consistent improvements in both predictive accuracy and estimated conversion lift over single-task competitors.
Keywords
E-Commerce; Cart Abandonment; Multi-Task Learning; Bi-LSTM; Graph Neural Networks; Explainable AI; Clickstream Analysis; Customer Behaviour Analytics
References
[1] J. Choi, H. Kim, and S. Lee, "Sequential modeling for e- commerce clickstreams with RNNs," in Proc. IEEE Int. Conf. Data Mining Workshops (ICDMW), 2017, pp. 245-252.
[2] Y. Zhang, L. Wang, and Q. Chen, "Predicting cart abandonment using machine learning techniques," IEEE Access, vol. 6, pp. 65789-65798, 2018.
[3] X. Li and Y. Zhao, "Graph-based recommendations in online retail," in Proc. ACM Conf. Recommender Systems (RecSys), 2019, pp. 312-320.
[4] R. Kumar, P. Sharma, and A. Verma, "Multi-task learning for customer behavior prediction in e-commerce," IEEE Trans. Knowledge and Data Engineering, vol. 32, no. 11, pp. 2145- 2158, 2020.
[5] T. Wang, M. Liu, and Z. Sun, "Explainable AI in e- commerce analytics," IEEE Intelligent Systems, vol. 35, no. 4,
[6] A. Singh and R. Gupta, "Deep learning approaches for purchase forecasting in online retail," IEEE Access, vol. 9, pp. 112345-112356, 2021.
[7] Y. Chen, K. Huang, and L. Zhou, "Time-aware recurrent models for clickstream data analysis," in Proc. IEEE Int. Conf. Big Data, 2022, pp. 1785-1792.
[8] D. Patel, S. Mehta, and N. Rao, "Hybrid Bi-LSTM and graph neural network models for e-commerce prediction," IEEE Trans. Neural Networks and Learning Systems, vol. 33, no. 8, pp. 3672-3684, 2022.
How to cite this paper
@article{1716019,
author = {Manimoli A T, Joshi M, Arokia Nisha S, Kayalvizhi S},
title = {E-Commerce Cart Prediction Using Multi-Task Learning with BI-LSTM and Graph Neural Networks},
journal = {Iconic Research And Engineering Journals},
year = {2026},
volume = {9},
number = {10},
pages = {697-702},
issn = {2456-8880},
url = {https://www.irejournals.com/formatedpaper/1716019.pdf},
abstract = {Shopping cart abandonment is a persistent and financially significant challenge in digital commerce, with reported global rates exceeding 68% and cumulative revenue losses running into the hundreds of billions annually. To tackle this, the present work proposes a unified multi-task learning framework capable of concurrently solving three interrelated problems: identifying abandonment risk, recommending the next likely product, and forecasting the expected transaction value. Instead of deploying isolated models for each objective, the system derives a shared latent representation from user clickstream logs and behavioural interaction histories. Temporal dynamics within sessions are encoded using a Bidirectional Long Short-Term Memory (Bi-LSTM) network, while structural dependencies across products and categories are modelled via a Graph Neural Network (GNN). The framework operates with low latency, supporting real-time deployment of targeted retention strategies before users disengage. To enhance model transparency, SHAP and LIME attribution methods are embedded as interpretability tools accessible to operational stakeholders. Empirical validation on two publicly available datasets demonstrates consistent improvements in both predictive accuracy and estimated conversion lift over single-task competitors.},
keywords = {E-Commerce; Cart Abandonment; Multi-Task Learning; Bi-LSTM; Graph Neural Networks; Explainable AI; Clickstream Analysis; Customer Behaviour Analytics},
month = {April},
doi = {https://doi.org/10.64388/IREV9I10-1716019}
}