Generating Recommendations with Watson ML

End of internship presentation for the IBM Watson Data Platform Developer Advocacy Team
Description of one project from the summer - process of building a demonstration recommendation engine for consumer shopping data showcasing the Watson Machine Learning service, and building a PixieApp to interactively test and display recommendations
Live Presentation - https://youtu.be/ekfemoGQPTM

Transcript

1. Generating Recommendations with Watson ML Katie Carlson, Developer Advocate Intern August 9, 2017
2. About Me
3. About Me
4. About Me
5. About Me
6. About Me
7. Tools used: Jupyter Notebooks PixieDust PixieApps
8. vishnuviswanath.com
9. Original Data Frame
10. Original Data Frame
11. Filtered Columns
12. Aggregated Rows
13. K-Means Clustering http://opencv-python-tutroals.readthedocs.io/en/latest/py_tutorials/py_ml/py_kmeans/py_kmeans_understanding/py_kmeans_understanding.html
14. K-Means Clustering http://opencv-python-tutroals.readthedocs.io/en/latest/py_tutorials/py_ml/py_kmeans/py_kmeans_understanding/py_kmeans_understanding.html z
15. K-Means Clustering http://opencv-python-tutroals.readthedocs.io/en/latest/py_tutorials/py_ml/py_kmeans/py_kmeans_understanding/py_kmeans_understanding.html
16. K-Means Clustering http://opencv-python-tutroals.readthedocs.io/en/latest/py_tutorials/py_ml/py_kmeans/py_kmeans_understanding/py_kmeans_understanding.html
17. K-Means Clustering http://opencv-python-tutroals.readthedocs.io/en/latest/py_tutorials/py_ml/py_kmeans/py_kmeans_understanding/py_kmeans_understanding.html
18. …
20. Configure WML credentials Authorize repository Save to WML instance Generate access token for WML API Create deployment for the model Load scoring endpoint for deployment Persist Model and Deploy to Cloud
21. Calculate cluster from shopping history Find most popular items for the cluster Filter products already in customer's cart Display recommendations
23. Limitations & Challenges • Randomness of data • New to tools: PixieDust, Spark • Lack of UI experience!
24. Future Additions & Improvements Future Additions & Improvements • Spherical k-means algorithm • View the similar customers’ carts in the app • More/different features: demographics, time of purchase, combination of items • Add product descriptions as features • Storing new shoppers and transactions to be able to retrain the model
25. Thank you! https://www.linkedin.com/in/katherine-carlson/ https://github.com/wdp-beta/get-started/blob/master/notebooks/ localcart-scenario-part-5.ipynb