Pattern Recognition in Industry

A guest lecture in IF5181 Pattern Recognition course on case studies about customer segmentation, churn prediction, text mining from reviews, and search ranking

Transcript

1. Pattern Recognition in Industry Ali Akbar Septiandri
2. Work experience ‍ • Data scientist lead @ Airy • Senior data scientist @ eFishery • Data science consultant @ HappyFresh
3. Typical cases 1. Customer segmentation ‍ 2. Churn prediction 3. Text mining from reviews or customer complaints (sentiment analysis) 4. Dynamic retail pricing 5. Demand forecasting (time series analysis)
4. Specific cases 1. Object recognition 2. Information retrieval 3. Predictive maintenance 4. Vehicle routing problem 5. Named entity recognition 6. Multi-armed bandit
5. Case studies 1. Customer segmentation ‍ 2. Churn prediction 3. Text mining from reviews 4. Search ranking
6. Customer Segmentation
7. What would you do?
8. Clustering ‍‍ ‍ ‍‍ ‍♂‍♀ ‍‍
9. ...then what? ‍♀
10. Monothetic vs Polythetic Toxic Loyalist Loyalist Star Needs Attention Promising Potential Loyalist Let Go At Risk Can’t Lose Them Level of loyalty CLV low mid high mid low high
11. Customer Lifetime Value
12. Historical vs Forecast A B NOW
13. Buy ‘Til You Die
14. Buy ‘Til You Die - Beta-Geometric/Negative Binomial dist.
15. Buy ‘Til You Die
16. Action Items Reward, Retain, Reactivate, Monetise, Minimise Toxic Loyalist Loyalist Star Needs Attention Promising Potential Loyalist Let Go At Risk Can’t Lose Them Level of loyalty CLV low mid high mid low high
17. Churn Prediction
18. What would you do?
19. Classification
20. Survival analysis
21. Kaplan-Meier estimator for survival analysis
22. Buy ‘Til You Die - Beta-Geometric/Negative Binomial dist.
23. Text Mining from Reviews
24. What would you do?
25. Sentiment analysis
26. Dictionary-like vs Modelling + - o SentiWordNet Positive Negative Subjective Objective Machine Learning
27. “tempatnya sempurna, pas didepan pantai losari, namun sayang acnya kurang dingin padahal sdh lapor namun tetep tidak bisa dingin.”
28. Aspect-based sentiment analysis
29. Aspect-based sentiment analysis Aspects: kamar, kamar mandi, air panas, AC, harga Sentiments: bersih, lama dingin, oke
31. Search Ranking
32. “Given a list of hotels with review scores in city X, sort them from the best to the worst.”
33. What would you do?
34. Simple average
35. (Counter)example
36. Adding uncertainty: Bayesian search ranking
37. Bayes’ rule posterior prior normalizing constant likelihood
38. Beta distributed review scores
39. Bayesian search ranking
40. Resources
47. 1. Find similar cases from similar business 2. Pattern recognition in industry is an iterative process 3. Challenge yourself - question everything! 4. Should lead to action items Key takeaways
48. Thank you! @aliakbars pm@aliakbars.id speakerdeck.com/aliakbars