This is the implementation refers to the paper Towards Bankruptcy Prediction: Deep Sentiment Mining to Detect Financial Distress from Business Management Reports.
Runs the model with the german word2vec of Andreas Müller (devmount, https://github.com/devmount). Please cite the author when using the data.
Code is written in Python 3.6 and requires
- Keras (2.1) with Theano or Tensorflow
- Django (2.0)
- NLTK (3.2) with German and English language packs
- scikit-learn (0.19)
Open root/webhighlighting/settings.py and set the correct paths where you find the # TODO comments.
If you want to train english models instead of german models, you have to upload the english word2vec dataset to the
directory root/trainmodel/data/ and call the file english_300dim.model. We recommend the google word2vec dataset
by Tomas Mikolov et al. (https://code.google.com/archive/p/word2vec/).
Check the variable mp_ in root/highlight/__init__.py and set the correct path to your model. Leave the extension
out. This tool will expect in the set directory files with the extensions yourfilename.patterns,
yourfilename.model and yourfilename.data. These files are generated in the training process of this tool.
You start the server by running python3 manage.py runserver in your console. In the beginning django will load the
model that is used for highlighting, this may take a while (probably up to 2h, but this is only needed initially).
After the model is loaded, you will see a URL in the console (most likely http://127.0.0.1:8000/). Open this URL in your
browser to access the tool.