Lessons Learned Building Natural Language Processing Systems with David Talby
💻 Abstract:
Lessons Learned Building Natural Language Processing Systems
🔊 Speaker bio:
David Talby CTO of John Snow Labs
David Talby is a chief technology officer at John Snow Labs, helping fast-growing companies apply NLP and AI to solve real-world problems in healthcare, life science, and related fields.
David has extensive experience in building and operating web-scale data science and business platforms, as well as building world-class, agile, distributed teams.
Previously, he led business operations for Bing Shopping in the US and Europe with Microsoft’s Bing Group and built and ran distributed teams that helped scale Amazon’s financial systems with Amazon in both Seattle and the UK.
David holds a Ph.D. in computer science and master’s degrees in both computer science and business administration.
If you enjoyed this talk, visit us at https://mlopsworld.com/ and come participate in our next gathering! 💼
Would you like to receive email summaries of these talks? Join our newsletter FREE here: http://bit.ly/MLOps_Summaries 📧
Timestamps:
0:00 Intro
0:11 Introduction of the host
2:05 Contents
2:37 AI versus Doctors, 2017
3:49 AI versus Doctors, 2018
5:01 Progress in reading comprehension, 2019
5:53 State of the art NLP
6:34 In the real-world, Production systems
7:34 Healthcare has hundreds of languages
Examples
7:57 Scanned Medical Records
9:03 Chat/Messaging Language
10:16 Academic Language
11:32 Pathology Reports
12:47 Other Common Formats
14:43 Healthcare NLP models don't generalize
15:16 Why can't I reuse an off-the-shelf NLP model?
Example
16:03 Emergency Room Language
18:18 NLP models rarely generalize, even on the same task
21:00 Train models on your own data
21:33 Spark NLP for healthcare: Reusable vs. Custom Assets
❓ Q&A ❓
23:25 How do you factor in the problem of badly transcribed patient records?
25:27 How many layers do you go back when fine tuning in these kind of problems?
33:18 What areas of healthcare are most amenable to NLP analysis?
36:36 Do your entity-recognition models need to be fully retrained as terminologies are updated, or does transfer learning work?
38:25 Closing remarks