AMUSE
AMUSE - financial summarization
Unzip data.zip
Train new model:
python FinAnalyze.py --task train --start 0 --count
data/train = dir where the text files are data/gold = dir where the gold summaries are
Trains new AMUSE prediction model for given files and stores it in an .h5 file
Generate summaries with existing model:
python FinAnalyze.py --task generate-summaries --start 0 --count
Also stored:
a model trained on 3000 files named model.training.muse.3000.all.h5
If you use this code, please cite:
Litvak M, Vanetik N. Summarization of financial reports with AMUSE. In Proceedings of the 3rd Financial Narrative Processing Workshop 2021 (pp. 31-36).
@inproceedings{litvak2021summarization, title={Summarization of financial reports with AMUSE}, author={Litvak, Marina and Vanetik, Natalia}, booktitle={Proceedings of the 3rd Financial Narrative Processing Workshop}, pages={31--36}, year={2021} }