Amazon.com 의 개인화 추천 / 예측 기능을 우리도 써 봅시다. :: 심호진 - AWS Community Day 2019
10 likes5,172 views
The document presents an overview of Amazon Personalize and Amazon Forecast, focusing on their capabilities for personalized recommendations and predictive analytics. It discusses the application processes, underlying technologies, and challenges faced in implementing these machine learning services. Additionally, it includes practical demonstrations and emphasizes the impact of personalization on customer experience and revenue.
Amazon.com 의 개인화 추천 / 예측 기능을 우리도 써 봅시다. :: 심호진 - AWS Community Day 2019
- 1. © 2018, Amazon Web Services, Inc. or its affiliates. All rights reserved. Amazon.com 의 개인화 추천 / 예측 기능을 우리도 써 봅시다. 심호진 11번가 주식회사 steven.hj.shim@gmail.com
- 2. Agenda Amazon Personalize 개인화 및 추천에 대하여 Amazon Personalize 소개 Amazon Personalize 사용 방법 데모 - 캡쳐 화면 결론 Amazon Forecast 예측 기술에 대하여 Amazon Forecast 소개 Amazon Forecast 사용 방법 데모 - 캡쳐 화면 결론
- 3. © 2018, Amazon Web Services, Inc. or its affiliates. All rights reserved. Amazon Personalize
- 6. 개인화 및 추천이 PV, 긍정적인 고객 경험, 매출 등에 영향을 줌
- 7. © 2018, Amazon Web Services, Inc. or its affiliates. All rights reserved. 기존 환경의 개인화 추천이 어려운 점… •ML • • computing resource • storage
- 8. © 2018, Amazon Web Services, Inc. or its affiliates. All rights reserved. Amazon Personalize 소개 Amazon.com Machine Learning API • Deep Learning •AutoML / HPO • Activity / Click Event /
- 9. 다양한 분야에 적용 가능 Video / Audio on demand / streaming Retail / e-commerce News / Media Travel
- 10. Amazon Personalize 적용 과정 Data S3 / API or AutoML Workload Event /
- 11. © 2018, Amazon Web Services, Inc. or its affiliates. All rights reserved. Amazon Personalize Recipe •Recipe ? • Machine Learning •Recipe •DeepFM, HRNN, SIMS, etc •AutoML & HPO • Recipe Hyperparameter •BYOR (Bring Your Own Recipe)
- 12. Amazon Personalize 실시간 추천 구조 Amazon Personalize 실시간 Event Data 전송 Browser 추천 요청 Inference 질의 추천 결과추천 결과 Server
- 13. Amazon Personalize 데모 - 캡쳐 화면
- 15. Dataset Group
- 17. S3 Object
- 18. Solution
- 19. Recipe TPS
- 20. Solution
- 21. * AutoML Recipe
- 23. sdk cli campaign
- 25. awscli 를 이용해 추천 요청
- 26. 추천 결과를 프로덕션에 반영 46171067 90482731 85720120 38760991 45203114 29184523
- 27. © 2018, Amazon Web Services, Inc. or its affiliates. All rights reserved. Amazon Personalize 결론 • AWS • - • - • / •
- 28. © 2018, Amazon Web Services, Inc. or its affiliates. All rights reserved. Amazon Forecast
- 29. © 2018, Amazon Web Services, Inc. or its affiliates. All rights reserved. Forecast - Data , Domain
- 30. 예측 기술의 어려운 점 • • • •
- 31. © 2018, Amazon Web Services, Inc. or its affiliates. All rights reserved. Amazon Forecast 소개 Amazon.com Machine Learning Time-series • 20 ML forecast • •Amazon.com MQ-RNN , ARIAM, ETS, Prophet
- 32. Amazon Forecast 적용 과정 Data S3 / API or AutoML Forecast
- 33. © 2018, Amazon Web Services, Inc. or its affiliates. All rights reserved. Amazon Forecast - Data 준비하기 Forecast Predictor Training 3 type dataset ) , ) , Target time-series Related time-series Item metadata
- 34. © 2018, Amazon Web Services, Inc. or its affiliates. All rights reserved. Amazon Forecast - Recipe 선택하기 (1/2) • Recipe , AutoML Amazon Forecast Recipe • forecasting Choose Recipe
- 35. © 2018, Amazon Web Services, Inc. or its affiliates. All rights reserved. Amazon Forecast - Recipe 선택하기 (2/2) •ARIMA •SQF •DeepAR+ •ETS •MDN •MQ-RNN •NPTS •Prophet AutoML
- 36. © 2018, Amazon Web Services, Inc. or its affiliates. All rights reserved. Amazon Forecast - Forecasting 결과 보기
- 37. Amazon Forecast 데모 - 캡쳐 화면
- 38. 샘플 데이터 준비 •가구별 시간당 전기 사용량 데이터 •https://docs.aws.amazon.com/ forecast/latest/dg/getting- started.html
- 39. Domain
- 40. Import Data ( , , )
- 42. Recipe
- 43. Predictor
- 46. © 2018, Amazon Web Services, Inc. or its affiliates. All rights reserved. Amazon Forecast 결론 • AWS • - • - • •
- 47. © 2018, Amazon Web Services, Inc. or its affiliates. All rights reserved. 마무리
- 48. 개인 소요 시간 2 hours Data & Training 2 hours 2~10 hours+ • 510 (1.7GB) • Python (json -> csv) • Machine Learning
- 50. Thank you! © 2018, Amazon Web Services, Inc. or its affiliates. All rights reserved. 심호진 / steven.hj.shim@gmail.com AWSKRUG Slack @steven