Venture capital horizon scanning

Abstract

Private Equity firms need to rely on larger and more diverse sources of data in order to make risk-aware investment decisions. Persons involved with making the investment decisions need tools that allow them to gain insight from such vast quantities of data. One part of that whole system is trying to understand what industry the company they are seeking to invest in are from and their specialities and characteristics. This Project fine tunes a Large Language Model to classify companies into their industries based on the description of the company. It is trained on data acquired from LinkedIn of over 64 million companies. It achieves a F1 score of 0.7101 and a Top-5 Accuracy of 88.83%.