Skip to content

Latest commit

 

History

History
32 lines (21 loc) · 802 Bytes

README.md

File metadata and controls

32 lines (21 loc) · 802 Bytes
Raw

Using LSTM for Sentiment Analysis of Amazon Reviews

LSTM built using PyTorch to predict whether a certain review is positive or negative (sentiment).

🍟 Results

A noticeable plateau occurs at around Step 5000 at the first epoch. A decrease in alpha (learning rate) is recommended and an increase in the number of embedding dimensions and hidden dimensions.

Test Loss: 0.211
Test Accuracy: 91.692%

🍰 Training the Model

  • Download the dataset
  • Extract the dataset
  • Run the jupyter notebook

⚙️ Requirements

  • Python >= 3.7.0
  • Jupyter Notebook
  • NumPy
  • PyTorch
  • CUDA 11.0
  • Optional: tqdm

📜 Dataset

Amazon Reviews for Sentiment Analysis (Kaggle)