nlp_bert_team

Natural Language Processing

Project

Elbrus Bootcamp | Phase-2 | Team Project

🦸‍♂️ Team

1. Anton Yablokov
2. Vika Ivanova
3. Solomon Chakaev

🎯 Task

Development of a multipage application using streamlit (Streamlit service deployed on HuggingFace Spaces):

• Page 1 • Review Classification for Polyclinics

    User-entered review classification model
    Outputs prediction results by three models:
    1) Classic ML algorithm (Logistic Regression) trained on BagOfWords representation
    2) LSTM model
    3) BERT-based
    Alongside the prediction, the time it was obtained is displayed
    The page features a comparative table based on the f1-macro metric for all constructed classifiers
  

• Page 2 • Text Generation with GPT Model

    Generative model 
    Users can adjust the length of the generated sequence
    The number of generations, temperature, top-k/p, maximum length, and the number of generated sequences can be controlled
  

• Page 3 • Evaluation of User Message Toxicity. The model is also available as a Telegram bot

    Accepts user messages and assesses their toxicity
    Implemented using rubert-tiny-toxicity