The project aims to create a simple API that will allow communication with a fine-tuned model and classifications of the type of tweets/posts - OFFENSIVE or NOT OFFENSIVE. While working on the project, an experiment tracking tool - W&B was used to track performance of the model. Model was trained using a GPU from Google Colab.
On the validation set, the model achieved an accuracy of 81.57%, precision of 68.92 % and recall of 74.27%.
Clone the repository with the command
git clone https://github.com/jmisilo/offensive-language-classificator
Then go to the directory and install depedencies:
cd offensive-language-classificator
pip install -r requirements.txt
To run the following command:
uvicorn src.app:app --port <PORT>
Predicting the Type and Target of Offensive Posts in Social Media - Zampieri, Marcos and Malmasi, Shervin and Nakov, Preslav and Rosenthal, Sara and Farra, Noura and Kumar, Ritesh
More than a feeling: Accuracy and Application of Sentiment Analysis - Hartmann, Jochen and Heitmann, Mark and Siebert, Christian and Schamp, Christina