This NLP app automatically classifies each sentence into a relevant heading (e.g. objective, methods, results, etc..), so you can quickly skim through abstracts and find the information you need. This project is inspired by the 2017 paper PubMed 200k RCT: a Dataset for Sequential Sentence Classification in Medical Abstracts.
to see the model experiments: All the notebooks are available Notebooks
The table below summarizes some of the model experiments tried :
| Model | Accuracy |
|---|---|
| NaiveBiase Model "Baseline" | 72% |
| Tribrid model "Universal Sentence Encoder, character embeddings, positional embeddings" | 83% |
| Bert Tribrid model "PUBMED BERT, character embeddings, positional embeddings" | 88% |
Without using Bert Pubmed embeddings
With Bert Pubmed embeddings
- go to Notebooks and download Running_in_colab Notebook
- upload it to Colab
- Run the cells
- After running the cell
! streamlit run app.py & npx localtunnel --port 8501you will see output like this :
- copy the External IP from the External URl in this image it was : 34.125.226.87 you will have a different one when running the code
- click on the URL next to your url is: you will be directed to this page :
- paste the External IP into tunnel password then click on Click to submit button so that the Streamlit app will be initialized to be tested out.
