Building new NLP solutions with spaCy and Prodigy

Loading

An introduction to NLP projects and their success and failure by the author of spaCy.

Follow to receive video recommendations   a   A


In this talk, the speaker discusses how to address some of the most likely causes of failure for new Natural Language Processing (NLP) projects. His main recommendation is to take an iterative approach: don't assume you know what your pipeline should look like, let alone your annotation schemes or model architectures. 

TimeLine Index

0:00 Introduction:

2:12 Talk begins

4:00 How to fail at NLP.

6:10 How to succeed.

11:05 Don’t assume, Iterate.

11:40 What can go wrong?

13:10 Bad Labelling example.

14:07 Better Labelling example

15:49 spaCy Annotations

17:00 Compose Generic Models.

19:03 Prodigy Annotation Tool

22:33 Avoid big annotation projects.

24 Micro-experiments

26:03 A/B Evaluation

27:36 Avoid Mechanical Turks.

30:40 Iterate, Iterate, Iterate.

31:30 Contact Info

31:35 Questions.



Editors Note:

If you like this website, please upvote my Awesome Python pull request.