Revisiting Financial Sentiment Analysis: A Language Model Approach

corresponding code to the paper: https://arxiv.org/abs/2502.14897

Easy access

• Neptune AI results for language model experiments can be found here including tables, confusion matrixes and more.
• the main notebook on Kaggle called tweet-classification can be found here
• the final implementation and optimization of Triple Barrier Labeling can be found in the notebook next_day_prediction

• backtesting experiments and results can be found here

How to run Experiments

use poetry to install the packages with poetry install. for more information go to poetry docs

then run with python src/run.py [Experiment ID]

Project Folders and structure

Here are the folders and what they contain:

• raw: unprocessed data
• dataset: processed data
• notebook: notebooks
• src: contains the source code for experiments

Overall Architecture

Summary of the Backtesting results