During our second rotation, our work focused on helping Vinnova using natural language processing for exploration of text data, automatisation of various text processing tasks and as aid in decision making. Project abstracts, as part of applications received by Vinnova, were the main source of the text data we worked with. The data was either stored in Excel-files or as PDF-files in a database.

The use cases considered were mainly classification tasks of various kinds, such as whether an application concerns a life-science project, or named entity recognition as a way of extracting relevant technologies from an abstract. Our solutions heavily feature models such as BERT and Sentence Transformers, where we fine tuned the models for the specific downstream tasks. 

Additionally, we worked on various minor tasks such as analysing the gender distribution in incoming applications, clustering and topic modeling. We also explored the possibility of automatic assignment of assessors for incoming applications.