TFP-Net: A temporal-feature-prototypical network for CRM optimization and cold-start mitigation.

Area of Science:

• Artificial Intelligence
• Machine Learning
• Data Science

Background:

• E-commerce platforms face challenges in predicting user behavior and providing personalized recommendations due to vast user data.
• The cold-start problem, where new users or products lack sufficient data, hinders recommendation system performance.
• Existing models struggle to effectively capture temporal dynamics and complex feature interactions in user behavior.

Purpose of the Study:

• To propose TFP-Net, a novel customer relationship management model for enhanced e-commerce recommendation systems.
• To address challenges in user behavior prediction, personalized recommendations, and the cold-start problem.
• To improve the performance and efficiency of recommendation systems in dynamic e-commerce environments.

Main Methods:

• TFP-Net integrates Temporal Graph Attention Mechanism (TGAT) to capture temporal user behavior dynamics.
• Deep Feature Interaction Module (DeepFM) is employed to learn complex user-product relationships.
• Prototypical Network (ProtoNet) is utilized to mitigate data sparsity and improve cold-start recommendations.

Main Results:

• TFP-Net demonstrated superior performance over traditional baseline models on the Taobao and Amazon Product datasets.
• The model achieved accuracy of 87.8% on the Taobao dataset and 88.1% on the Amazon dataset.
• TFP-Net showed significant improvements in cold-start scenarios, outperforming existing models by 1.5% to 2.1%.

Conclusions:

• TFP-Net effectively addresses user behavior prediction and personalized recommendations, particularly in cold-start situations.
• The model offers superior computational efficiency, with reduced training time and inference latency.
• TFP-Net presents a promising new solution for personalized recommendations and customer relationship management in e-commerce.