Deep Graph Embedding for Ranking Optimization in E-commerce.

Area of Science:

• E-commerce
• Machine Learning
• Graph Theory

Background:

• Effective buyer-seller matching is crucial for e-commerce customer satisfaction and platform profitability.
• Current ranking systems often fail due to mismatches in seller quality, negatively impacting user experience and ROI.
• Incorporating intra-group structural information, like buyer-buyer proximity, can address these limitations.

Purpose of the Study:

• To propose a novel deep learning method, Deep Graph Embedding (DEGREE), for enhanced e-commerce matching.
• To jointly exploit inter-group (buyer-seller) and intra-group (buyer-buyer) proximities for improved structural learning.
• To enhance matching performance and operational efficiency in e-commerce platforms.

Main Methods:

• Developed Deep Graph Embedding (DEGREE), a deep learning model leveraging graph embedding techniques.
• Incorporated intra-group structural information (buyer-buyer proximity) alongside traditional inter-group (buyer-seller) proximity.
• Utilized a sparse filtering technique to optimize computational resource usage.

Main Results:

• DEGREE demonstrated superior performance compared to state-of-the-art graph embedding methods on real-world e-commerce datasets.
• The method significantly improved matching performance while requiring fewer computational resources than alternative deep learning approaches.
• An online A/B test showed DEGREE boosted the average unit price in purchases by up to 11.93%.

Conclusions:

• DEGREE effectively integrates diverse structural information for superior e-commerce matching.
• The proposed method offers a computationally efficient solution for improving platform operational efficiency and user satisfaction.
• DEGREE represents a significant advancement in applying graph embedding for optimizing e-commerce ranking systems.