6. Have a clearly defined goal
Conversion is the usual culprit.
What do you want to increase?
- Views
- Clicks
- Orders
- Shopping cart
- Sign ups
- Active sessions
7. Map business goal to ML metric
What’s the most suitable Machine
Learning (ML) evaluation metric to
measure views?
8. Start small
Models:
- Logistic Regression
- Matrix Factorization
- Simple rules: top n <action> e.g top 7 viewed|purchased|liked etc
Benefits
- Fast iteration
- Provides a baseline
9. You don’t need Big Data
Rule of thumb: If your data can fit on a moderately-sized RAM, ditch Spark
Benefits:
- Reduced complexity
- Plenty time to focus on other things
- Less money on infrastructure
10. Explain your recommendations
Examples:
- Because you <action> <item> e.g Because you bought Samsung S7
- Based on your browsing history
Benefits:
- Users understand why they are presented with the recommendations
- It fosters trust
14. UX matters
Choice overload: cognitive process in which people have a difficult time making a
decision when faced with many options.
Things to consider:
- How many items are ideal?
- What should be the quality of recommendations?
- Is there a clear winner in the recommendations?
- Are there duplicate items in the recommendations? - usually points to
engineering problem
16. Others
- Let users provide feedback
- Don’t be afraid to ask a user for his/her preferences
- A/B test
- Refresh recommendations
17. Because You Viewed This Presentation
- Lessons learned from building real-life recommender systems -
https://www.slideshare.net/xamat/recsys-2016-tutorial-lessons-learned-from-
building-reallife-recommender-systems
- Recommendation Systems -
https://www.slideshare.net/VikrantArya/recommendation-system-33379953
- Algorithmic Music Recommendations at Spotify -
https://www.slideshare.net/MrChrisJohnson/algorithmic-music-
recommendations-at-spotify?qid=36836a91-6664-4e9c-b857-
3ecdbb3c3346&v=&b=&from_search=1
- Learning A Personalized Homepage - https://medium.com/netflix-