Pinterest上广告转化优化模型的演变

A Journey from GBDT to Multi-Task Ensemble DNN

从GBDT到多任务集成DNN的旅程

Aayush Mudgal | Senior Machine Learning Engineer, Ads Ranking Conversion Modeling; Han Sun | Senior Machine Learning Engineer, Ads Ranking Conversion Modeling; Matt Meng | Senior Machine Learning Engineer, Ads Ranking Conversion Modeling; Runze Su | Machine Learning Engineer II, Ads Ranking Conversion Modeling; Jinfeng Zhuang | Staff Machine Learning Engineer, Ads Ranking Conversion Modeling

Aayush Mudgal | 高级机器学习工程师,广告排名转化建模;Han Sun | 高级机器学习工程师,广告排名转化建模;Matt Meng | 高级机器学习工程师,广告排名转化建模;Runze Su | 机器学习工程师 II,广告排名转化建模;Jinfeng Zhuang | 高级机器学习工程师,广告排名转化建模

In this blog post, we will share how we improved Pinterest’s conversion optimization performance by leveraging Deep Neural Networks (DNN), Multi-Task Learning (MTL), state-of-the-art feature interaction modules, in-model ensemble techniques, and user sequence modeling. We will also cover our transition to GPU serving, which is indispensable for large-scale complex models.

在本博客文章中,我们将分享如何通过利用深度神经网络(DNN)、多任务学习(MTL)、最先进的特征交互模块、模型集成技术和用户序列建模来改善Pinterest的转化优化性能。我们还将介绍我们向GPU服务的过渡,这对于大规模复杂模型是不可或缺的。

Background

背景

People often come to Pinterest for inspiration on their next life or shopping ideas. In fact, over half of Pinterest users think of Pinterest as a place to shop. Ads and other content from businesses are an important part of the consumer journey because they add value to the discovery process. Conversion Optimization optimizes Promoted Pins for specific consumer conversion actions, rather than just clicks. Advertisers can choose conversions as a campaign objective and inspire people to take specific actions like online checkouts, increased signups, or stronger leads. In this blog we will deep dive into some of our recent advancements in machine learning modeling to connect pinners with the most relevant ads.

人们经常来Pinterest寻找下一个生活或购物的灵感。事实上,超过一半的Pinterest用户认为Pinterest是一个购物的地方。来自企业的广告和其他内容是消费者旅程的重要组成部分,因为它们为发现过程增加了价值。转化优化针对特定的消费者转化行为优化推广的Pin,而不仅仅是点击。广告主可以选择转化作为广告活动的目标,并激励人们采取特定的行动,如在线结账、增加注册或更强的潜在客户。在本博客中,...

开通本站会员,查看完整译文。

trang chủ - Wiki
Copyright © 2011-2024 iteam. Current version is 2.137.1. UTC+08:00, 2024-11-15 15:00
浙ICP备14020137号-1 $bản đồ khách truy cập$