This project is inspired by the paper Dernoncourt et al., 2016, which focuses on classifying sections of medical paper abstracts.
This model improves upon the original approach by incorporating: ✅ BERT embeddings for word representation. It also includes:
- ✅ Character-level embeddings for fine-grained text understanding.
- ✅ Additional position embeddings to capture abstract structure.
- ✅ BiLSTM layers to process sequential text information.
- ✅ Multi-input architecture, combining different feature representations.
The model classifies medical abstracts into 5 categories, improving upon the baseline architecture.
✔ BERT-based word embeddings (instead of traditional static embeddings).
- Clone the repository:
git clone https://github.com/yourusername/SkimLit_improved.git cd SkimLit_improved