A Cross-Lingual Real-Time E-Commerce Recommendation Method Based on Siamese Graph Convolutional Network and Bilinear...

Area of Science:

• Artificial Intelligence
• Machine Learning
• Recommender Systems

Background:

• Real-time recommendation systems face challenges in balancing low latency and high accuracy, particularly in cross-lingual scenarios.
• Existing methods struggle to efficiently handle diverse user-item interactions across different languages.

Purpose of the Study:

• To propose a novel two-stage method for cross-lingual real-time recommendation that optimizes both latency and accuracy.
• To enhance entity representation learning and feature interaction for improved recommendation performance.

Main Methods:

• Abstracting the cross-language scenario as a heterogeneous graph.
• Utilizing a Siamese Graph Convolutional Network for entity representation learning.
• Employing an efficient bilinear attention mechanism for deep feature interaction.

Main Results:

• The Siamese Graph Convolutional Network achieved stable normalized discounted cumulative gain (NDCG) values over 7.8% (20% higher than other models) for a recommendation list length of 30.
• The bilinear attention mechanism demonstrated superior convergence, reaching a mean average value of 12.7% in the 100th round, 1.9 percentage points higher than the standard bilinear mechanism.
• The proposed recommendation method achieved a 46.5% hit rate in increasing the sales rate of "long-tail products".

Conclusions:

• The proposed two-stage method effectively balances low latency and high accuracy in cross-lingual real-time recommendation.
• The approach shows significant potential for improving recommendation systems in cross-border e-commerce and other multilingual applications.
• The study highlights the effectiveness of graph convolutional networks and attention mechanisms for complex recommendation tasks.