展开查看详情
1 .王宇博
Head of Developer Relations GCR
© 2022, Amazon Web Services, Inc. or its affiliates. All rights reserved.
2 .无处不在的推荐服务
Dear Mr./Mrs. XXX
There is a new product…
16:30
AD
© 2022, Amazon Web Services, Inc. or its affiliates. All rights reserved.
3 .什么是推荐系统/
A recommender system or a recommendation engine is a
Wikipedia subclass of information filtering systems that seeks to predict
the “rating” or “preference” a user would give to an item
中文百科
© 2022, Amazon Web Services, Inc. or its affiliates. All rights reserved.
4 .为什么要使用推荐系统
帮助用户在海量信息/ 提高用户点击
提高用户在线时长
增加业务营收
© 2022, Amazon Web Services, Inc. or its affiliates. All rights reserved.
5 .亚马逊引领个性化推荐
20 余年的持续演进
1998
第一个个性化推荐功能上线
2021
使用机器学习跨渠道和
设备为客户提供复杂、
独特的个性化推荐体验
5
© 2022, Amazon Web Services, Inc. or its affiliates. All rights reserved.
6 .推荐系统的类型
• • /
• •
…
•
•
•
•
• /
•
•
•
非个性化推荐 个性化推荐
© 2022, Amazon Web Services, Inc. or its affiliates. All rights reserved.
7 .63%
www.business2community.com/
marketing/30-amazing-personalization-
statistics-02289044
© 2022, Amazon Web Services, Inc. or its affiliates. All rights reserved.
8 .构建推荐系统的技术挑战
基于机器学习的技术虽然表现很好
基于规则的推荐策略 低延迟实时推荐 避免始终推荐最热 单一算法不能解决 构建准确有效的推荐
往往并不有效 冷启动问题 的商品 所有推荐场景 模型需要机器学习专家
© 2022, Amazon Web Services, Inc. or its affiliates. All rights reserved.
9 .推荐系统不仅仅是机器学习模型
近线特征生成 在线模型更新 服务动态扩展
模型离线验证 特征拼接 超参调优
离线特征存储 在线特征存储 模型在线推理
实时埋点收集 推荐系统 离线模型更新
冷启动 在线特征处理 离线特征处理
特征更新 模型训练
结果干预 模型批量转换 …
© 2022, Amazon Web Services, Inc. or its affiliates. All rights reserved.
10 .基本推荐系统构成
© 2022, Amazon Web Services, Inc. or its affiliates. All rights reserved.
11 .Amazon Personalize
•
•
•
•
2020 KDD
https://www.amazon.science/blog/amazon-researchers-win-best-paper-award-at-kdd
© 2022, Amazon Web Services, Inc. or its affiliates. All rights reserved.
12 .基本推荐系统构成
Amazon Personalize
Amazon Personalize
© 2022, Amazon Web Services, Inc. or its affiliates. All rights reserved.
13 . Amazon Personalize
工作原理
/
Amazon Personalize
API
Amazon Personalize
© 2022, Amazon Web Services, Inc. or its affiliates. All rights reserved.
14 .三步操作 Amazon Personalize
数据导入 模型训练 模型推理
© 2022, Amazon Web Services, Inc. or its affiliates. All rights reserved.
15 . 1.交互数据 2.用户数据 3.项目数据
必选 可选 可选
历史积累的用户与项目之间的交互数据 用户画像数据 项目元数据
例：用户ID、年龄、性别、会员等 例：项目ID、创建时间、所属类别、
例：用户A在T时刻购买了项目B，金额X元
-> A,B,T,Buy,C
级 价格
-> A,25, Male, Golden -> B, T, 3C, 2100
必填项：USER_ID、ITEM_ID、 必填项：USER_ID、至少一个其他画 必填项：ITEM_ID、至少一个其他元
TIMESTAMP 像数据 数据
静态数据启动、更新
流式数据注入
© 2022, Amazon Web Services, Inc. or its affiliates. All rights reserved.
16 . 1
• USER_PERSONALIZATION
§
§ GetRecommendations API user_id Item list
•
§
§
§
§ /
© 2022, Amazon Web Services, Inc. or its affiliates. All rights reserved.
17 . 2
• PERSONALIZED_RANKING
§
§ GetPersonalizedRanking API user_id Item list Item list
•
§
§
§
© 2022, Amazon Web Services, Inc. or its affiliates. All rights reserved.
18 . 3
• RELATED_ITEMS
§ Item
§ GetRecommendations API item_id Item list
§ SIMS Item Item feature
•
§ X
Y X Y
§ SIMS +
© 2022, Amazon Web Services, Inc. or its affiliates. All rights reserved.
19 .Amazon Personalize Campaign
§ Campaign minimal TPS
§ Campaign 的
§ 收费 MAX(minimal TPS, real TPS) /h
§ Campaign
§ Price recommendation
© 2022, Amazon Web Services, Inc. or its affiliates. All rights reserved.
20 .Amazon Personalize
•
•
• 确
• 确
•
•
•
•
•
•
© 2022, Amazon Web Services, Inc. or its affiliates. All rights reserved.
21 .Retail Demo Store
© 2022, Amazon Web Services, Inc. or its affiliates. All rights reserved.
22 .Retail Demo Store 的架构
https://github.com/aws-samples/retail-demo-store
© 2022, Amazon Web Services, Inc. or its affiliates. All rights reserved.
23 . 务
“ Amazon Personalize ,
20%
APP
——
”
© 2022, Amazon Web Services, Inc. or its affiliates. All rights reserved.
24 .Amazon Personalize AI
业务场景
及使用 需要关注 Personalize
数据 需要关注
推荐算法设计 不需要关注 千
推荐模型训练 不需要关注 Personalize 模
机器学习/
框架 不需要关注 Personalize
计算资源 不需要关注 Personalize 托管服务
© 2022, Amazon Web Services, Inc. or its affiliates. All rights reserved.
25 .Thank you!
© 2021,
2022, Amazon Web Services, Inc. or its affiliates. All rights reserved.
热门城市
