Customer support recommender system migrates from gradient-boosted trees to deep learning

A research paper details the migration of a production customer support recommender system from a gradient-boosted tree model to a deep recommender architecture. The migration was necessary due to evolving product catalogs and the need to incorporate multimodal signals like transcripts. The paper outlines techniques such as reformulating recommendation as pairwise binary prediction, negative sampling, and attention pooling over transcript chunks to maintain recommendation quality. The new deep recommender approach demonstrated parity at the beginning of conversations and superior performance in later stages compared to a CatBoost baseline. AI

影响 Demonstrates practical application of deep learning models for complex recommendation tasks, potentially improving customer support efficiency.

排序理由 The cluster contains a research paper detailing a practical migration of a machine learning system. [lever_c_demoted from research: ic=1 ai=1.0]

在 arXiv cs.AI 阅读 →

• arXiv
• Catboost
• CORE Recommender
• DagsHub
• Deep Recommenders
• Gradient Boosted Trees
• Hugging Face
• IArxiv Recommender

• 论文
• 产品
• 基础设施

AI 生成摘要 · Google Gemini · 来自 1 个来源。 我们如何撰写摘要 →