dsa4263-pietonium

Repository Structure

├── archive

├── data
    └── curated/reviews
        └── ...
    └── raw
        └── ...
    └── ...
    
├── deployment
    └── ...
    
├── model
    └── ...
 
├── sentiment_analysis
    └── Deep Learning # fine-tuning experiments of BERT models (files: .py, .ipynb, .txt)
        └── ...
    └── ML # training experiments of traditional machine learning algorithm, including SVM and XGBoost (files: .py, .ipynb)
        └── ...
          
├── service
    └── ...
     
├── test
    └── ...
      
 ├── topic_modelling 
    └── ...   

Folder Descriptions

1. archived - contains unused files and models for reference, not included in training or inferencing pipeline
   
2. data - contains raw and preprocessed reviews
   
3. deployment - contains modularized and python scripts needed for app deployment and traning-prediction pipeline
   
4. model - contains binary files of models for sentiment analysis, and topic modelling and classification [SVM, XGBoost, LDA, Gensim (except BERT)]
   
5. sentiment_analysis - contains jupyter notebooks used for training sentiment analysis pipeline
   
6. service - contains files needed to dockerise app
   
7. test - contains python scripts for unit testing
   
8. topic_modelling - contains jupyter notebook and python scripts for topic modelling and classification

Notes

• DistilBERT model is too large to push to GitHub (~800MB) and hence be loaded from Hugging Face Hub, which is integrated in code.