弥合可负担性、熟悉感与新颖性：DoorDash的LLM辅助个性化框架

A recap of our KDD 2025 PARIS Workshop talk: “Affordability, Familiarity, and Novelty: An LLM-assisted Personalization Framework for Multi-Vertical Retail Discovery.”

我们KDD 2025巴黎研讨会演讲的回顾：“可负担性、熟悉性和新颖性：一个LLM辅助的多垂直零售发现个性化框架。”

Imagine a world where every shopping moment, from a last-minute grocery run to a weekend gifting spree, feels effortless, personalized, and just right for you. At DoorDash, this is more than a vision; it’s our daily mission. As we expand beyond restaurants into new verticals like Grocery, Convenience, Alcohol, Retail, Flowers, and Gifting, we face a fascinating challenge: how do we help customers discover what they want, or what they didn’t know they wanted,  across a catalog of hundreds of thousands of SKUs?

想象一个世界，每一个购物时刻，从临时的杂货采购到周末的赠礼狂欢，都感觉轻松、个性化，并且正合适你。在DoorDash，这不仅仅是一个愿景；这是我们的日常使命。随着我们从餐厅扩展到杂货、便利店、酒类、零售、鲜花和赠礼等新垂直领域，我们面临一个迷人的挑战：我们如何帮助客户发现他们想要的，或者他们不知道自己想要的，跨越数十万SKU的目录？

In August 2025 at KDD 2025’s PARIS Workshop in Toronto, DoorDash showcased its latest advances in personalization for multi-vertical retail. Sudeep Das, Head of New Verticals ML/AI, and Raghav Saboo, Staff Machine Learning Engineer, shared how we are reimagining discovery through a large language model-assisted personalization framework. 

在2025年8月的KDD 2025的PARIS研讨会上，DoorDash展示了其在多垂直零售个性化方面的最新进展。Sudeep Das，新垂直机器学习/人工智能负责人，以及Raghav Saboo，资深机器学习工程师，分享了我们如何通过大型语言模型辅助的个性化框架重新构想发现过程。

Our work blends traditional machine learning with large language models (LLMs) to dynamically balance three core value dimensions for consumers:

我们的工作将传统机器学习与大型语言模型（LLMs）相结合，以动态平衡消费者的三个核心价值维度：

• Familiarity – surfacing the items you already love and trust
• 熟悉度 – 显示您已经喜爱和信任的商品
• Affordability – meeting you at your price preferences with the right deals
• 可负担性 – 以合适的交易满足您的价格偏好
• Novelty – introducing you to new, complementary, and exciting products
• 新颖性 – 向您介绍新的、互补的和令人兴奋的产品

Figure 1: Our Search and Personalization Framework is aimed at enhancing discovery by balancing three user value dimensions: familiarity, affordability, and novelty

图1：我们的搜索和个性化框架旨在通过平衡三个用户价值维...