SkimLit: Categorizing Medical Research Paper Abstracts

Overview

SkimLit is a project designed to automatically categorize sections of medical research paper abstracts into five distinct categories:

• Background
• Methods
• Objective
• Results
• Conclusions

This is achieved using a multimodal model that incorporates both word and character level tokenization, as well as a model for line number.

Dataset

The dataset is sourced from PubMed 200k RCT

Features

• Multimodal Tokenization: Utilizes both word and character level tokenization to capture detailed textual features.
• Line Number Model: Incorporates the position of each line within the abstract to provide context.
• Category Prediction: Accurately predicts the category for each line in the abstract.

Model Details

The multimodal model combines:

• Word Level Tokenization: Splits the abstract into words and extracts features.
• Character Level Tokenization: Splits the abstract into characters and extracts features.
• Line Number Model: Incorporates the line number of each line within the abstract.

Training

The training process involves:

Training the model using a suitable machine learning framework Tensorflow .