The first edition of the virtual Convention A takes place from 19 to 23 September 2022. On 22 September, Andreas Troxler will present on the topic „NLP for Actuarial Applications“, see agenda.
The presentation discusses the key concepts and main results of the recent article „Natural Language Processing – A tutorial for actuaries„.
We are looking forward to seeing you at the Convention A!
Abstract
This presentation demonstrates how text data can be incorporated into actuarial classification and regression tasks.
After briefly explaining the key concepts underlying relevant leading-edge natural language processing (NLP) models, the main part of the talk is dedicated to case studies demonstrating these techniques.
These case studies use a real-world dataset of mixed English and German road accident descriptions with an average length of 400 words and a dataset of short property insurance claims descriptions. The discussion addresses practical challenges related to a multi-lingual setting, long input sequences, the need to understand model output, and approaches to handling classification tasks in situations where there is little or no labeled data.
Both the datasets and the implementation are available as part of a tutorial recently published by the working group “Data Science” of the Swiss Association of Actuaries.
The results achieved by using the language-understanding skills of off-the-shelf NLP models with only minimal pre-processing and fine-tuning clearly demonstrate the power of transfer learning for practical applications.
Are you interested in the topic but not sure where to begin?
Do not hesitate to contact me.