NLP Inspiration Workshop

26 March 2021

Article written by Jeanine Schoonemann, Principal Consultant

After reading the interesting blogpost series on NLP, written by Jurriaan Nagelkerke and Wouter van Gils, a lot of colleagues wanted to know more on how we can use NLP models in a sensible and useful way in our everyday work.

 

Add predictive power

We are all Data Analysts and Data Scientists who spend a lot of our time on building pragmatic predictive models that can be easily implemented in the way of working of a company. As we could read in the blog series, adding the output of an NLP model to the input of our regular (tabular based) predictive models can add a lot of predictive power. As always, there is a downside to this, it makes the model more of a black box and thus less explainable.

 

Inspiration session NLP, planned in no time

Intrigued as we were by the blogposts, we wanted to pick the brains of Jurriaan and Wouter a bit more on this topic. Even though the blogs are quite elaborate, there is nothing better than a live (digital) conversation. Luckily for us, Wouter and Jurriaan love to share their knowledge and an inspiration session NLP was planned in no time!

 

The subjects about NLP

It was an interesting session where they took us along in the wonderful world of BERT and some of his siblings, like RoBERTta and BERTje (no Ernie was not there, but we did talk about Elmo). Some of the subjects we have touched in this inspiration session:

  • What are transformer models?
  • How performant are single-shot-models for your own use case?
  • What is BERT?
  • Who are BERTs siblings and why do they differ?
  • When is it useful to use an NLP model like BERT?
  • What do you need to consider when it comes to data preparation?
  • What is important to realize when you want to use (the output of) an NLP model?
  • Can you still use NLP models if you do not have access to a GPU?

 

Time to experiment and learn

Next to all of this, we have also spent a lot of time on discussing the questions we had regarding this topic and some of the real-world cases we were thinking adding an NLP model could be useful. Also, we spent some time on really playing around with BERT and his siblings, to get some sense on how this works and how we can influence the model by changing the value of certain parameters. This also gave us a lot of insight on how fast the training time of the model can increase by increasing even only one of these parameters (like epochs and batch size).

 

Summary

This was such a fun way to either get a first introduction into NLP models, advance on the topic after working mostly with topic modelling or learn more while touching the subject a little before. A diverse group of people, but everybody got what he was looking for!

 

Thanks Jurriaan and Wouter for this inspiration workshop on NLP!

To the workshop: Next level text mining

 

Contact

Do you want to know more about this subject? Please contact Jeanine Schoonemann using the details below

Jeanine Schoonemann, Principal Consultant

+31 6 55 89 75 12

j.schoonemann@cmotions.nl

Subscribe to our newsletter

Never miss anything in the field of advanced analytics, data science and its application within organizations!