Tone tags text classifier

Current results

• All collected and processed datasets are in their respective archives in the following folder:

.\datasets

archiving was implemented using WinRAR

• Project presentation

• Project report

Metrics

The code to retrieve the metrics is provided in the file:

.\notebooks\check_lstm_results.ipynb

Here are the best results, the results for the other datasets for the lstm model can be seen in the above file. All metrics are presented with weighted and top k 3 parameters.

LSTM Accuracy (weighted, top_k=3)

LSTM Confusion matrix in 11 epoch (weighted, top_k=3)

LSTM F1 Score (weighted, top_k=3)

LSTM Precision (weighted, top_k=3)

LSTM Recall (weighted, top_k=3)

How to try to use model from this repository

• to use this project you need to unpack model to

.\results\models\lstm_model.pt

The archived model is located in the folder:

.\results\models\lstm_model.zip

• The project uses glove vectors, downloading it manually may be faster than the code will do, so you can download glove.twitter.27B.50d from here manually (optional): Link to GloVe

• to host this app download model in directory and use this command:

python -m streamlit run streamlit_app.py

About requirements

The current requirements contain versions of the libraries to run the deployment, but are not tested for use by notebooks. Use the following guide to install torch correctly.