The purpose of this notebook is to build a NLP model to make reading medical abstracts easier.

The paper we're replicating (the source of dataset that we'll be): is avalable here https://arxiv.org/abs/1612.05251

The goal of the dataset was to explore the ability for NLP models to classify sentences which appear in sequential order.

In other words, given the abstract of a RCT, what role does each sentence serve in the abstract?

Since we'll be replicating the paper above (PubMed 200k RCT), let's download the dataset they used.

We can do so from the authors GitHub: https://github.com/Franck-Dernoncourt/pubmed-rct

• Using TensorFlow Datasets to download and explore data.
• Creating preprocessing function for our data.
• Batching & preparing datasets for modelling (making our datasets run fast).
• Creating modelling callbacks.
• Building a Model Experiments.
  • Model 1: Conv1D with Token Embedding.
  • Model 2: Tfhub Pretrained feature extraction
  • Model 3: Conv1D with Character Embedding
  • Model 4: Pretrained token embedding + character embedding + Positional embedding.

Trained a NLP model with competitive performance to a research paper and in far less time ad with accuracy of 85.6%.