We used Natural Language Processing to analyse the language from psychotherapy sessions. Natural Language Processing involves the use of computers to pick out different features of a section of text (for example, how long the sentences are, how many words related to sentiment there are, etc) and put these into numbers.

These numbers can be compared across different sections of text to see where there are differences. In ExTRAPPOLATE, we canvassed opinion on what the important language features for psychotherapy were from multiple experts in the team.

We used machine learning, including techniques such as CHAID, random forests and support vector machines, to compare these differences and see if these could be used to predict different types of interaction in the session, or to predict the level of patient activation. We then looked to see how accurate these predictions were.

You can hear more about the technical approach to the project in this video:

The results of the analyses are available in our End of Project Roadshow video.