---

Skimlit

Overview

Skimlit is a sequential sentence classification project aimed at analyzing and classifying medical abstracts from the PubMed dataset. By utilizing cutting-edge embeddings, including character, token, and positional encodings, Skimlit demonstrates high-performance models, particularly the Tribrid Pos Char Token Embed model. The models are evaluated on the PubMed 20k and PubMed 200k RCT datasets, highlighting significant advancements in text classification.

Model Performance

PubMed 20k Dataset

• Baseline Model:

  • Accuracy: 72.18% | Precision: 0.719 | Recall: 0.722 | F1 Score: 0.699

• Custom Token Embed Conv1D:

  • Accuracy: 78.67% | Precision: 0.783 | Recall: 0.787 | F1 Score: 0.784

• Pretrained Token Embed:

  • Accuracy: 78.67% | Precision: 0.783 | Recall: 0.787 | F1 Score: 0.784

• Custom Char Embed Conv1D:

  • Accuracy: 65.18% | Precision: 0.643 | Recall: 0.652 | F1 Score: 0.643

• Hybrid Char Token Embed:

  • Accuracy: 73.21% | Precision: 0.733 | Recall: 0.732 | F1 Score: 0.730

• Tribrid Pos Char Token Embed:

  • Accuracy: 83.42% | Precision: 0.834 | Recall: 0.834 | F1 Score: 0.833

PubMed 200k Dataset

• Best Model (Tribrid Pos Char Token Embed):
  • Accuracy: 87.51% | Precision: 0.876 | Recall: 0.875 | F1 Score: 0.874

Getting Started

1. Clone the Repository:

   git clone https://github.com/your-username/skimlit.git

2. Navigate to the Project Directory:

   cd skimlit

3. Set Up Environment: Install required Python packages:

   pip install -r requirements.txt

4. Train Models: Refer to steps in notebook.

5. Run Analysis: Review metrics and model results outlined in notebook.ipynb.

Contributing

Contributions are welcome! Please feel free to submit issues or pull requests.

Contact

For any questions or feedback, reach out via email.

---