mohsenSohrabi / EduSentimentAnalysis

DistilBERT-based Sentiment Analysis for Educational Reviews

This project involves fine-tuning a DistilBERT model on student reviews provided by Coursera. The model is trained to perform sentiment analysis on the reviews, classifying them based on the sentiment expressed.

Getting Started

Inference with the Trained Model

You can directly use the fine-tuned model for inference using the demo.py script. This script loads the trained model and uses it to perform sentiment analysis on a sample review. You can change the sample_review variable in the script to test the model on different reviews.

The demo.py script automatically downloads the trained model checkpoint from Google Drive, so you can use the model without having to fine-tune it yourself. To use the model, simply run the demo.py script.

Fine-Tuning the Model

If you wish to fine-tune the model yourself, you can do so by running the train_model.py script. This script fine-tunes a DistilBERT model on the review data, and saves the trained model. The fine-tuning process involves the following steps:

1. Downloading the dataset from Google Drive.
2. Transforming the data into a format suitable for Hugging Face's datasets library.
3. Tokenizing the data using a DistilBERT tokenizer.
4. Loading a pre-trained DistilBERT model for sequence classification from Hugging Face's model hub.
5. Fine-tuning the model on the review data.

You can start the fine-tuning process by running the following command:

• On Windows: py train_model.py
• On other platforms: python train_model.py

Model Performance

The accuracy of the model after fine-tuning was about 79%. This means that the model correctly predicts the sentiment of the reviews 79% of the time. Happy coding! 😊