This folder contains a streamlit app which can be used to combine different parameters and assess the models performance on the validation data.
This part of the project is deprecated since the generation of a robust labeling set turned out to not be feasible due to ressource constraints but also due to the structure of the problem. For documentation reasons this subproject is kept in the repository. Go to our main repository to read up on our project.
The ground truth to train and test the model consists of ca. 750 texts which were annotated with the labels human_rights, environment, corruption, social_concerns and employee_concerns. Some of the texts are not unique if they have multiple labels.
The labels were created by the dev team of bizres, based on relevant extracts from sustainability reports
The following approaches are tested and validated:
Cleansing and tokenization: The spacy md language model is used for filtering stopwords and lemmatization
Text Vectorization: Bag-of-Words and vectorization using the spacy language model
Prediction Algorithm: Logistic Regression and Naive Bayes
Optimization: Optimizing the modell for overall accuracy or for recall
-
To run the application with Docker, install the Docker Engine on your development machine.
-
Navigate to the local repository
-
Run
docker build -t report-topic-classification .to build a local docker image -
Run
docker run -p 8501:8501 report-topic-classificationto run the docker image -
Visit localhost:8501 to interact with the streamlit app
- For development we use Ubuntu 20.04 with Python 3.9. But the code should also run on Windows with Python 3.9
- If necessary, install Python 3.9 and Pipenv and spawn a virtual environment with
pipenv shell - Install libraries using the Pipfile with
pipenv install - To run the streamlit app usw
streamlit run app_model_validation.py - Have fun developing ;-)