An ensemble model, PP-GLAM, is introduced to improve search relevance in e-commerce by combining language and graph models. The model outperforms baselines on real-world datasets, providing better interpretability and adoption strategies.
The content discusses the challenges of search relevance in e-commerce due to evolving architectures. It introduces PP-GLAM as a solution for bridging the gap between research and practical deployment. The model combines language models (LMs) and graph neural networks (GNNs) to capture semantic and behavioral signals for improved search relevance.
PP-GLAM uses additive explanation metrics to independently decide on including language model candidates, GNN model candidates, and inter-product behavioral signals. The approach aims to address the lack of interpretability in existing models by providing a modular framework with uniform data processing pipelines.
Extensive experiments on real-world multilingual e-commerce datasets show that PP-GLAM outperforms state-of-the-art baselines and proprietary models. The model's interpretability is analyzed along with its computational complexity. A deployment strategy is provided for practical implementation.
Overall, PP-GLAM offers a comprehensive solution for improving search relevance in e-commerce through an interpretable ensemble of graph and language models.
To Another Language
from source content
arxiv.org
Key Insights Distilled From
by Nurendra Cho... at arxiv.org 03-05-2024
https://arxiv.org/pdf/2403.00923.pdfDeeper Inquiries