DataScienceLab2017_Как знать всё о покупателях (или почти всё)?_Дарина Перемот
•
0 likes•448 views
DataScienceLab, 13 мая 2017 Как знать всё о покупателях (или почти всё)? Дарина Перемот (ML Engineer at SynergyOne) Раскроем собственный ответ на вопрос "Чего же хочет покупатель?". Поделимся результатами исследований транзакций и расскажем, есть ли у вас домашний питомец. А так же, продемонстрируем, как машинное обучение уже сейчас помогает узнавать вас ближе. Все материалы: http://datascience.in.ua/report2017
Report
Share
DataScienceLab2017_Как знать всё о покупателях (или почти всё)?_Дарина Перемот
- 1. How to know (almost) everything about customers?
- 2. Magic time Choose any number
- 3. Tricky questions time Who would want to read your minds?
- 5. What would the merchant want? Tricky questions time
- 7. Tricky questions time What to sell? What is its price? When to sell it? Whom to sell?
- 8. What is the purchase formula?
- 12. Experts Ukrainian dataset 2,5 years 210 K rows 7 K customers 22 K POS American dataset 1 year 180 K rows 12 K customers 40 K POS What they have in common?
- 13. Cash Food stores Cafe & restaurants Gas stations ...
- 14. Tricky questions time Is there any “similarity” in client's behaviour?
- 15. Key points Uses less or equal than 4 MCC codes. Ukrainian dataset: 64% American dataset: 61% Uses less or equal than 6 MCC codes. Ukrainian dataset: 73% American dataset: 76%
- 16. Tricky questions time Why do we interested in the “similarity” of customers?
- 19. Benefit Targeted advertising in real time
- 20. The musketeers Find the target audience
- 21. The musketeers
- 22. Behavior pattern Behavior pattern = customer’s tag
- 23. Behavior pattern Customer’s feature connected with distribution of expenditures via budgeting categories, locations, time.
- 26. Tricky questions time Why would merchant need all this stuff?
- 27. Tricky answer time Discover unexpected target audience Get to know the competitors Find out the locals
- 28. The musketeers
- 30. DBSCAN
- 31. The musketeers
- 33. OneKarma beta
- 34. OneKarma beta
- 35. Links YaCM 2016 (about hyperlocation adds): https://events.yandex.ru/lib/talks/3469/ Sberbank Data Science Journey (MCC2Vec for Sberbank): https://www.sdsj.ru/slides/MCC2VEC.pdf https://youtu.be/0q5p7xP4cdA?t=23702
- 36. Links DBSCAN vs others: http://scikit-learn.org/stable/auto_examples/clu ster/plot_cluster_comparison.html OneKarma beta-test UI: https://onekarma.synergyone.pro Comics: https://marketoonist.com/
- 37. Thank You Darina Peremot ML engineer in SynergyOne email: dp@synergyone.pro facebook: @darina.peremot telegram: @peremot