工业级深度学习推荐系统框架详解-技术圈

","24:"BEUK""],[20,"1、抖音日活用户破 6 亿，推荐系统是怎么做到的？"],[20,"
","24:"DxzF""],[20,"2、你想什么它就推送什么，大厂的推荐系统为什么这么牛？"],[20,"
","24:"mXC7""],[20,"3、从 0 搭建一个工业级推荐系统。"],[20,"
","24:"ZOYb""],[20,"4、天猫千亿交易额，推荐系统都做了哪些事情？"],[20,"
","24:"bhaJ""],[20,"5、年薪百万的推荐工程师，需要具备哪些能力？"],[20,"
","24:"QYiE""],[20,"
","24:"feQq""],[20,"摘要：搭建一个完整的深度学习推荐系统","0:"%23000000"|27:"12"|8:1"],[20,"
","24:"P1wc""],[20,"作者：王喆","0:"%23333333"|27:"12"|8:1"],[20,"
","24:"FMKg""],[20,"
","24:"sFes""],[20,"推荐系统从没像现在这样，影响着我们的生活。当你上网购物时，天猫、京东会为你推荐商品；想了解资讯，头条、知乎会为你准备感兴趣的新闻；想消遣放松，抖音、快手会为你奉上让你欲罢不能的短视频。","27:"12""],[20,"
","24:"RAO9""],[20,"
","24:"4kpi""],[20,"而驱动这些巨头进行推荐服务的，都是基于深度学习的推荐模型。","27:"12""],[20,"
","24:"IPLu""],[20,"
","24:"Pjbo""],[20,"2019 年阿里的千人千面系统，促成了天猫”双 11“ 2684 亿成交额。假设通过改进商品推荐功能，使平台整体的转化率提升 1%，就能在 2684 亿成交额的基础上，再增加 26.84 亿。这就是推荐工程师的最大魅力，也是它支撑起百万年薪的主要原因。","27:"12""],[20,"
","24:"7ooZ""],[20,"
","24:"URR8""],[20,"但在一个成熟的推荐系统上，找到提升的突破点并不容易——不能满足于协同过滤、矩阵分解这类传统方法，而要建立起完整的深度学习推荐系统知识体系，加深对深度学习模型的理解和大数据平台的熟悉程度，才能实现整体效果上的优化。","27:"12""],[20,"
","24:"epEP""],[20,"
","24:"LcLC""],[20,"上半年，因为疫情抽空看了本书叫《深度学习推荐系统》，对我启发很大，豆瓣评分也挺高的 ","27:"12""],[20,"9.3","0:"%23000000"|27:"12"|8:1"],[20,"。作者是王喆，他是 Roku 资深机器学习工程师，推荐系统架构负责人，从业这些年，他一直深耕于推荐系统、计算广告领域，经验非常丰富。","27:"12""],[20,"
","24:"vFHD""],[20,"
","24:"ebyy""],[20,"所以，当得知他推出了专栏","27:"12""],[20,"《深度学习推荐系统实战》","0:"%23000000"|27:"12"|8:1"],[20,"，我第一时间就订阅了，跟着学下来，真是受益匪浅，之前尝试过很多深度学习模型，但效果始终没有提升。直到遇到这门课，让我对深度学习推荐系统的认知到了一个新高度，很想把它推荐给你。","27:"12""],[20,"
","24:"BohH""],[20,"
","24:"Pbkm""],[20,"在专栏中，他讲解了深度学习推荐系统的经典架构设计，带你掌握 Embedding 技术的主要实现方法，构建完整的推荐系统评估体系路径，并搭建出一个工业级的深度学习推荐系统。","27:"12""],[20,"
","24:"zOr6""],[20,"
","24:"yKJr"|7:1"],[20,"（插入海报小卡片）","1:"%23f9eda6""],[20,"
","24:"PF4F"|7:1"],[20,"?扫码免费试读","27:"12"|31:2"],[20,"
","24:"ajtC"|7:1"],[20,"拼团+口令「study2020」","27:"12"|31:2"],[20,"立省 ¥30","0:"%23ff0000"|27:"12"|31:2"],[20,"
","24:"HM1h"|7:1"],[20,"原价 ¥99，口令仅限「","27:"12"|31:2"],[20,"前 50 人","0:"%23ff0000"|27:"12"|31:2"],[20,"」有效","27:"12"|31:2"],[20,"
","24:"Y6iX"|7:1"],[20,"新人首单 ¥19.9","0:"%23ff0000"|27:"12""],[20,"
","24:"jQkl"|7:1"],[20,"
","24:"8lcr"|7:1"],[20,"他是怎样讲解这门课程的？","0:"%23000000"|27:"16"|8:1"],[20,"
","24:"NnD6"|32:2"],[20,"
","24:"7UUQ""],[20,"在课程设置上，他遵循了经典推荐系统的框架，将课程分为 6 部分，通过 30+ 深度学习推荐系统问题，带你串联起深度学习推荐系统的知识体系，并收获了一套他实践过的深度学习推荐系统开源代码，实现一个工业级的深度学习推荐系统。","27:"12""],[20,"
","24:"ffbq""],[20,"
","24:"gjGB""],[20,"可以看看专栏里的","27:"12""],[20,"学习图谱","0:"%23000000"|27:"12"|8:1"],[20,"，方便你了解这门课的设计以及用到的技术。","27:"12""],[20,"
","24:"Ksqt""],[20,"
","24:"T08N""],[20,{"gallery":"https://uploader.shimo.im/f/jYciwK93hcOxNtI4.png!thumbnail"},"0:"%23000000"|27:"12"|29:0|30:0|3:"1454"|4:"auto"|8:1|crop:""|frame:"none"|ori-height:"1406"|ori-width:"1454""],[20,"
","24:"1AbL""],[20,"
","24:"IzA1""],[20,"基础架构篇","0:"%23000000"|27:"12"|31:2|8:1"],[20,"：","27:"12"|31:2|8:1"],[20,"从推荐系统要解决的主要问题入手，讲解我们要从 0 开始实现的推荐系统， Sparrow RecSys 的主要功能和技术架构，也会用到 Spark、Flink、TensorFlow 等业界最流行的机器学习和大数据框架。","27:"12"|31:2"],[20,"
","24:"bX8g""],[20,"
","24:"w9r6""],[20,"特征工程篇","0:"%23000000"|27:"12"|31:2|8:1"],[20,"：","27:"12"|31:2|8:1"],[20,"讨论推荐系统会用到的特征，以及主要的特征处理方式，并将其实践在 Spark 上。此外，还有深度学习中非常流行的 Embedding、Graph Embedding 技术，并带你实现 Sparrow Recsys 中的相似电影推荐功能。","27:"12"|31:2"],[20,"
","24:"E9MK""],[20,"
","24:"qyUW""],[20,"线上服务篇","0:"%23000000"|27:"12"|31:2|8:1"],[20,"：","27:"12"|31:2|8:1"],[20,"在这部分，他会带你地搭建一个推荐服务器，包括服务器、存储、缓存、模型服务等模块和相关知识，","27:"12"|31:2"],[20,"涉及 Jetty Server， Spark、Redis 的使用，带你初步掌握推荐工程师在工程领域的核心技能。","27:"12""],[20,"
","24:"FS6B""],[20,"
","24:"rNlN""],[20,"推荐模型篇","0:"%23000000"|27:"12"|8:1"],[20,"：","27:"12"|8:1"],[20,"带你学习深度学习推荐模型的原理和实现方法，包括 ","27:"12"|31:2"],[20,"Embedding+MLP ，Wide&Deep，PNN ","27:"12""],[20,"等深度学习模型的架构和 ","27:"12"|31:12"],[20,"TensorFlow ","27:"12""],[20,"实现，以及注意力机制、序列模型、增强学习等相关领域的前沿进展。","27:"12"|31:12"],[20,"
","24:"MaPL""],[20,"
","24:"eTet""],[20,"效果评估篇","0:"%23000000"|27:"12"|31:2|8:1"],[20,"：","27:"12"|31:2|8:1"],[20,"学习","27:"12""],[20,"效果评估的主要方法和指标，建立起包括线下评估、线上 AB 测试、评估反馈闭环等整套的评估体系，真正能够用业界的方法而不是实验室的指标来评价一个推荐系统。","27:"12"|31:2"],[20,"
","24:"0oMS""],[20,"
","24:"4zDE""],[20,"前沿拓展篇","0:"%23000000"|27:"12"|31:2|8:1"],[20,"：","27:"12"|31:2|8:1"],[20,"将业界巨头们的深度学习推荐系统方案进行融汇贯通，重点讲解 YouTube、阿里巴巴、微软、Pinterest 等一线公司的深度学习应用，帮你追踪业界发展的最新趋势，并找到自己技术道路的方向。","27:"12"|31:2"],[20,"
","24:"DC7j""],[20,"
","24:"0dNq""],[20,"具体内容可以看看目录：","27:"12"|31:2"],[20,"
","24:"cKGk""],[20,"
","24:"vVhj""],[20,{"gallery":"https://uploader.shimo.im/f/dov25Fg1evD5Mcno.png!thumbnail"},"0:"%23000000"|27:"12"|29:0|30:0|3:"751"|4:"auto"|8:1|crop:""|frame:"none"|ori-height:"3073"|ori-width:"751""],[20,"
","24:"FSAV"|7:1"],[20,"
","24:"kkDx"|7:1"],[20,"订阅福利","0:"%23000000"|27:"12""],[20,"
","24:"GB3L"|32:2"],[20,"拼团+","27:"12""],[20,"口令","0:"%23ff0000"|27:"12""],[20,"「study2020」","27:"12""],[20,"立省 ¥30","0:"%23ff0000"|27:"12""],[20,"
","24:"rRvu"|7:3"],[20,"原价 ¥99，口令仅限","27:"12""],[20,"「前 50 人」","0:"%23ff0000"|27:"12""],[20,"有效","27:"12""],[20,"
","24:"JUeB"|7:3"],[20,"新人首单 ¥19.9","0:"%23ff0000"|27:"12""],[20,"
","24:"vKXG"|7:3"],[20,"
","24:"EfyG""],[20,"
","24:"trFR""],[20,"（插入号主海报）","1:"%23ffd900"|27:"12""],[20,"
","24:"bvsN"|7:1"],[20,"?扫码免费试读"],[20,"
","24:"ResZ"|7:1"],[20,"
","24:"Q3RD""],[20,"?点击「阅读原文」，","27:"12""],[20,"
","24:"uc6S""],[20,"输入","27:"12""],[20,"优惠口令","0:"%23ff0000"|27:"12""],[20,"「study2020」","27:"12""],[20,"
","24:"Cu4Q""],[20,"立省 ¥30","0:"%23ff0000"|27:"12""],[20," ","0:"%23ff7800"|27:"12""],[20,"入手，仅限","27:"12""],[20,"「前 50 人」","0:"%23ff0000"|27:"12""],[20,"有效","27:"12""],[20,"
","24:"bQo3""],[20,"新人首单 ¥19.9","0:"%23ff0000"|27:"12""],[20,"
","24:"bKNn""],[20,"
","24:"jhCr""],[20,"阅读原文+两个跳转图片，均跳转海报链接","1:"%23ffd900"|27:"12""],[20,"
","24:"9FeJ""],[20,"
","24:"Re4J""]]">

推荐系统从没像现在这样，影响着我们的生活。当你上网购物时，天猫、京东会为你推荐商品；想了解资讯，头条、知乎会为你准备感兴趣的新闻；想消遣放松，抖音、快手会为你奉上让你欲罢不能的短视频。

而驱动这些巨头进行推荐服务的，都是基于深度学习的推荐模型。

2019 年阿里的千人千面系统，促成了天猫”双 11“ 2684 亿成交额。假设通过改进商品推荐功能，使平台整体的转化率提升 1%，就能在 2684 亿成交额的基础上，再增加 26.84 亿。这就是推荐工程师的最大魅力，也是它支撑起百万年薪的主要原因。

但在一个成熟的推荐系统上，找到提升的突破点并不容易——不能满足于协同过滤、矩阵分解这类传统方法，而要建立起完整的深度学习推荐系统知识体系，加深对深度学习模型的理解和大数据平台的熟悉程度，才能实现整体效果上的优化。

上半年，因为疫情抽空看了本书叫《深度学习推荐系统》，对我启发很大，豆瓣评分也挺高的 9.3。作者是王喆，他是 Roku 资深机器学习工程师，推荐系统架构负责人，从业这些年，他一直深耕于推荐系统、计算广告领域，经验非常丰富。

所以，当得知他推出了专栏《深度学习推荐系统实战》，我第一时间就订阅了，跟着学下来，真是受益匪浅，之前尝试过很多深度学习模型，但效果始终没有提升。直到遇到这门课，让我对深度学习推荐系统的认知到了一个新高度，很想把它推荐给你。

在专栏中，他讲解了深度学习推荐系统的经典架构设计，带你掌握 Embedding 技术的主要实现方法，构建完整的推荐系统评估体系路径，并搭建出一个工业级的深度学习推荐系统。

?扫码免费试读

拼团+口令「study2020」立省 ¥30

原价 ¥99，口令仅限「前 50 人」有效

新人首单 ¥19.9

他是怎样讲解这门课程的？

在课程设置上，他遵循了经典推荐系统的框架，将课程分为 6 部分，通过 30+ 深度学习推荐系统问题，带你串联起深度学习推荐系统的知识体系，并收获了一套他实践过的深度学习推荐系统开源代码，实现一个工业级的深度学习推荐系统。

可以看看专栏里的学习图谱，方便你了解这门课的设计以及用到的技术。

基础架构篇：从推荐系统要解决的主要问题入手，讲解我们要从 0 开始实现的推荐系统， Sparrow RecSys 的主要功能和技术架构，也会用到 Spark、Flink、TensorFlow 等业界最流行的机器学习和大数据框架。

特征工程篇：讨论推荐系统会用到的特征，以及主要的特征处理方式，并将其实践在 Spark 上。此外，还有深度学习中非常流行的 Embedding、Graph Embedding 技术，并带你实现 Sparrow Recsys 中的相似电影推荐功能。

线上服务篇：在这部分，他会带你地搭建一个推荐服务器，包括服务器、存储、缓存、模型服务等模块和相关知识，涉及 Jetty Server， Spark、Redis 的使用，带你初步掌握推荐工程师在工程领域的核心技能。

推荐模型篇：带你学习深度学习推荐模型的原理和实现方法，包括 Embedding+MLP ，Wide&Deep，PNN 等深度学习模型的架构和 TensorFlow 实现，以及注意力机制、序列模型、增强学习等相关领域的前沿进展。

效果评估篇：学习效果评估的主要方法和指标，建立起包括线下评估、线上 AB 测试、评估反馈闭环等整套的评估体系，真正能够用业界的方法而不是实验室的指标来评价一个推荐系统。

前沿拓展篇：将业界巨头们的深度学习推荐系统方案进行融汇贯通，重点讲解 YouTube、阿里巴巴、微软、Pinterest 等一线公司的深度学习应用，帮你追踪业界发展的最新趋势，并找到自己技术道路的方向。

具体内容可以看看目录：

订阅福利

拼团+口令「study2020」立省 ¥30

原价 ¥99，口令仅限「前 50 人」有效

新人首单 ¥19.9

?扫码免费试读

?点击「阅读原文」，

输入优惠口令「study2020」

立省 ¥30入手，仅限「前 50 人」有效

新人首单 ¥19.9