ai & personalised experiences
TRANSCRIPT
![Page 1: AI & Personalised Experiences](https://reader035.vdocuments.net/reader035/viewer/2022062306/5a64ab527f8b9a5d568b50a5/html5/thumbnails/1.jpg)
AI & Personalised Experiences@neal_lathia, Senior Data Scientist, Skyscanner
November 29, 2017
![Page 2: AI & Personalised Experiences](https://reader035.vdocuments.net/reader035/viewer/2022062306/5a64ab527f8b9a5d568b50a5/html5/thumbnails/2.jpg)
![Page 3: AI & Personalised Experiences](https://reader035.vdocuments.net/reader035/viewer/2022062306/5a64ab527f8b9a5d568b50a5/html5/thumbnails/3.jpg)
![Page 4: AI & Personalised Experiences](https://reader035.vdocuments.net/reader035/viewer/2022062306/5a64ab527f8b9a5d568b50a5/html5/thumbnails/4.jpg)
A variety of machine learning problems
● Price accuracy & caching● Flight itinerary search● Destination inspiration & recommendation● Advertisement relevance ranking● Growth forecasting & customer value● Conversations● ...many more.
![Page 5: AI & Personalised Experiences](https://reader035.vdocuments.net/reader035/viewer/2022062306/5a64ab527f8b9a5d568b50a5/html5/thumbnails/5.jpg)
![Page 6: AI & Personalised Experiences](https://reader035.vdocuments.net/reader035/viewer/2022062306/5a64ab527f8b9a5d568b50a5/html5/thumbnails/6.jpg)
Personalised Experiences
![Page 7: AI & Personalised Experiences](https://reader035.vdocuments.net/reader035/viewer/2022062306/5a64ab527f8b9a5d568b50a5/html5/thumbnails/7.jpg)
Tailoring the product to specific customers
![Page 8: AI & Personalised Experiences](https://reader035.vdocuments.net/reader035/viewer/2022062306/5a64ab527f8b9a5d568b50a5/html5/thumbnails/8.jpg)
3 Examples 1. Destination Recommendation2. Itinerary Recommendation3. Contextual Support
From current experiments
![Page 9: AI & Personalised Experiences](https://reader035.vdocuments.net/reader035/viewer/2022062306/5a64ab527f8b9a5d568b50a5/html5/thumbnails/9.jpg)
1. Destination Recommendation
![Page 10: AI & Personalised Experiences](https://reader035.vdocuments.net/reader035/viewer/2022062306/5a64ab527f8b9a5d568b50a5/html5/thumbnails/10.jpg)
![Page 11: AI & Personalised Experiences](https://reader035.vdocuments.net/reader035/viewer/2022062306/5a64ab527f8b9a5d568b50a5/html5/thumbnails/11.jpg)
![Page 12: AI & Personalised Experiences](https://reader035.vdocuments.net/reader035/viewer/2022062306/5a64ab527f8b9a5d568b50a5/html5/thumbnails/12.jpg)
Can we do better?
● Historical focus on cheapness: price is only 1 thing that matters● Sparse user data - travel is infrequent● Destinations are relative - London from Edinburgh is not the same as
London from New York.
● … without imposing any new burden on our app users?
![Page 13: AI & Personalised Experiences](https://reader035.vdocuments.net/reader035/viewer/2022062306/5a64ab527f8b9a5d568b50a5/html5/thumbnails/13.jpg)
Recommending destinations based on unsupervised learning
Popular, Localised, Trending
![Page 14: AI & Personalised Experiences](https://reader035.vdocuments.net/reader035/viewer/2022062306/5a64ab527f8b9a5d568b50a5/html5/thumbnails/14.jpg)
Recommending destinations
![Page 15: AI & Personalised Experiences](https://reader035.vdocuments.net/reader035/viewer/2022062306/5a64ab527f8b9a5d568b50a5/html5/thumbnails/15.jpg)
2. Itinerary Recommendation
![Page 16: AI & Personalised Experiences](https://reader035.vdocuments.net/reader035/viewer/2022062306/5a64ab527f8b9a5d568b50a5/html5/thumbnails/16.jpg)
Itineraries as a ranking problem
![Page 17: AI & Personalised Experiences](https://reader035.vdocuments.net/reader035/viewer/2022062306/5a64ab527f8b9a5d568b50a5/html5/thumbnails/17.jpg)
Can we do better?
● Historical focus on cheapness: price is only 1 thing that matters● Sparse user data - travel is infrequent● Itineraries are complicated to trade-off against one another
● … without imposing any new burden on our app users?
![Page 18: AI & Personalised Experiences](https://reader035.vdocuments.net/reader035/viewer/2022062306/5a64ab527f8b9a5d568b50a5/html5/thumbnails/18.jpg)
Itinerary ranking as a supervised learning problem
![Page 19: AI & Personalised Experiences](https://reader035.vdocuments.net/reader035/viewer/2022062306/5a64ab527f8b9a5d568b50a5/html5/thumbnails/19.jpg)
Recommendations as overlaid results
![Page 20: AI & Personalised Experiences](https://reader035.vdocuments.net/reader035/viewer/2022062306/5a64ab527f8b9a5d568b50a5/html5/thumbnails/20.jpg)
3. Contextual Support
![Page 21: AI & Personalised Experiences](https://reader035.vdocuments.net/reader035/viewer/2022062306/5a64ab527f8b9a5d568b50a5/html5/thumbnails/21.jpg)
From search results into result controls
![Page 22: AI & Personalised Experiences](https://reader035.vdocuments.net/reader035/viewer/2022062306/5a64ab527f8b9a5d568b50a5/html5/thumbnails/22.jpg)
From search results into result controls
![Page 23: AI & Personalised Experiences](https://reader035.vdocuments.net/reader035/viewer/2022062306/5a64ab527f8b9a5d568b50a5/html5/thumbnails/23.jpg)
Can we do better?
● We know that some search tools are helpful in some situations, and less helpful in other situations
● Thousands of different search combinations - how can we manage the complexity of figuring out when a specific search tool is helpful?
● Many ideas for new search tools, tips, and messages - how can we manage the complexity of adding new types of results without any historical data?
● … without imposing any new burden on our app users?
![Page 24: AI & Personalised Experiences](https://reader035.vdocuments.net/reader035/viewer/2022062306/5a64ab527f8b9a5d568b50a5/html5/thumbnails/24.jpg)
Multi-armed bandits to learn what support works in what contexts
Multi-armed bandits work by automating the process of exploring various layouts -- and then being able to exploit the layout that has worked best across each specific context.
![Page 25: AI & Personalised Experiences](https://reader035.vdocuments.net/reader035/viewer/2022062306/5a64ab527f8b9a5d568b50a5/html5/thumbnails/25.jpg)
Multi-armed bandits to learn what support works in what contexts
![Page 26: AI & Personalised Experiences](https://reader035.vdocuments.net/reader035/viewer/2022062306/5a64ab527f8b9a5d568b50a5/html5/thumbnails/26.jpg)
Lessons Learned
![Page 27: AI & Personalised Experiences](https://reader035.vdocuments.net/reader035/viewer/2022062306/5a64ab527f8b9a5d568b50a5/html5/thumbnails/27.jpg)
Machine Learning for Product Managers. (Medium)
![Page 28: AI & Personalised Experiences](https://reader035.vdocuments.net/reader035/viewer/2022062306/5a64ab527f8b9a5d568b50a5/html5/thumbnails/28.jpg)
The AI Hierarchy of Needs. M. Rogati.
![Page 29: AI & Personalised Experiences](https://reader035.vdocuments.net/reader035/viewer/2022062306/5a64ab527f8b9a5d568b50a5/html5/thumbnails/29.jpg)
Rules of Machine Learning: Best Practices for ML Engineering. M. Zinkevich.
![Page 30: AI & Personalised Experiences](https://reader035.vdocuments.net/reader035/viewer/2022062306/5a64ab527f8b9a5d568b50a5/html5/thumbnails/30.jpg)
The State of Data Science and Machine Learning 2017. (Kaggle)
![Page 31: AI & Personalised Experiences](https://reader035.vdocuments.net/reader035/viewer/2022062306/5a64ab527f8b9a5d568b50a5/html5/thumbnails/31.jpg)
AI & Personalised Experiences@neal_lathia