Datenextraktion von Aktieneinstufungen - Data extraction of stock ratings with LLMs

Based on PDFs from analyst firms with stock ratings, structured data is extracted in JSON format and prepared for further processing. A software prototype is created that uses current Large Language Models (LLMs) to extract the desired data structure from the unstructured documents. The software prototype consists of pre-processing the documents, using the language models with suitable prompts and appropriate post-processing. Current language models are evaluated with regard to their suitability for this task and checked with test data. In particular, European and open source models are also evaluated.
Local use is also evaluated for the operation of the language models, in addition to the use of cloud services.