Purchase prediction by statistical analysis (統計技術を用いた商品購買予測)
9 likes5,029 views
This document discusses using statistical analysis of purchase history data to predict users' purchase intervals and remind them when it is time to purchase an item again. It finds that for the rice category, 47% of users have relatively fixed purchase intervals with only a few outliers. By detecting these fixed intervals, a reminder system could notify users just before their next predicted purchase to prevent them from forgetting to buy an item. This has the potential to benefit users while also promoting repeated purchases in various product categories.
Purchase prediction by statistical analysis (統計技術を用いた商品購買予測)
- 1. Takashi Umeda (梅田卓志) @umekoumeda Oct. 20th , 2012 Purchase prediction by statistical analysis
- 2. 楽天技術研究所 Rakuten Institute of Technology Value Proposition Third Reality Vision Tokyo & NY & Paris Strategic R&D organization for Rakuten
- 3. Biography 3 • Takashi Umeda • Twitter : @umekoumeda Profile Work Purchase prediction >>> Users’ Benefits Prediction of purchase interval Seasonality forecasting Preference prediction
- 4. Biography 4 • Takashi Umeda • Twitter : @umekoumeda Profile Work Purchase prediction >>> Users’ Benefits Prediction of purchase interval Seasonality forecasting Preference prediction
- 5. Objective • Predict the users’ purchase interval • Focus on non-durable goods 5
- 6. Example of application 30 days 30 days 30 days Past FuturePast Future Buy Buy Buy Buy
- 7. Example of application 30 days 30 days 30 days Past FuturePast Future Buy Buy Buy Buy Remind！ We can remind users just before next purchase 30 days そろそろ、買い時では？
- 8. Users’ benefits Prevent users from forgetting to purchase Empty I forgot to purchase! It’s time to purchase! Few NG OK Notification
- 9. How can we predict purchase interval ?
- 10. Data set Purchase history in rice category Target users : Users purchasing over 4 times in one year Pick Up
- 11. Example of users with only fixed intervals BUY 30 days BUY BUY 31 days 29 days BUY Past Future
- 12. BUY 30 days BUY BUY 31 days 29 days BUY Past Future All purchase intervals are fixed. It’s about 30 days. We call those users as “Users with only fixed intervals” Example of users with only fixed intervals
- 13. Coverage of users with only fixed intervals Target Users with only fixed intervals (Predictable users) About 11% • Target : Users purchasing over 4 times in 1 year Too low coverage ! Not practical !
- 14. Example of users with a few outlier intervals BUY 31days 60 days 29 days BUY BUY BUY 30 days BUY
- 15. BUY 31days 60 days 29 days BUY BUY BUY 30 days BUY Most of purchase intervals are fixed. It’s about 30 days. Example of users with a few outlier intervals
- 16. BUY 31days 60 days 29 days BUY BUY BUY 30 days BUY A few intervals are outlier intervals Example of users with a few outlier intervals
- 17. BUY 31days 60 days 29 days BUY BUY BUY 30 days • There are a lots of users with many fixed and a few outlier intervals • We call those users as “Users with a few outlier intervals” BUY Example of users with a few outlier intervals
- 18. BUY 31days 60 days 29 days BUY BUY BUY 30 days BUY Cause for the outlier interval Why outlier intervals happened ?
- 19. 31days 60 days 29 days30 days Cause for the outlier interval 5kg 10kg 5kg 5kg 5kg • If consumer purchased more, interval had been longer • This type of users account for 22 %
- 20. Coverage of users with a few outlier intervals Predictable users 47% Users with a few outlier intervals 36% • Target : Users purchasing over 4 times in 1 year Target Users with only fixed intervals 11%
- 21. Trends in any other categories Predictable users exist in not only rice category but also any other categories. Ratio of predictable users 33.4 % 55.3 % 46.1 % 43.2 %
- 22. Items which we should show 57% In the reminding system, it’s better to show the item which has been purchased before. Users repeatedly purchase the same item at the same shop Users repeatedly purchase different items - Items sold at the same shop (10%) - Same priced items (14%) 43%
- 23. Items which we should show 21% In the reminding system, it’s better to show various kinds of items. Users repeatedly purchase the same item at the same shop Users repeatedly purchase different items - Items sold at the same shop (32%) - Same priced items (31%) 79%
- 24. Summary There are many predictable users 47% in the rice category We can remind users at the right moment ! It makes users happy ! Prevent users from forgetting to purchase item
- 25. Message There are many Fixed interval users It make users happy ! • In rice category, those users account for 47%. • Many categories have same trends By using detected fixed interval, We can remind users just before next purchase Users can avoid from forgetting to purchase regular buying items Purchase Prediction Users’ Benefits
- 26. Message There are many Fixed interval users It make users happy ! • In rice category, those users account for 47%. • Many categories have same trends By using detected fixed interval, We can remind users just before next purchase Users can avoid from forgetting to purchase regular buying items Purchase Prediction Users’ Benefits If you come up with any idea, feel free to tweet via twitter @umekoumeda