Using NLP techniques for answering questions about COVID-19 using Kaggle Dataset

This is a summary of what was done, the full report is at NLP_COVID_Presentation.

Introduction and Scope

• COVID-19 Open Research Dataset
  • ~13 GB
  • Over 135,000 scholarly articles
  • Including over 68,000 with full text
• First goal: What do we know about COVID-19 symptoms?
• Second goal: How can we cluster papers into coherent groups?

First goal

Words representation

• Chosen method: GloVe – Global Vectors for word representation
  1. Generate corpus using the provided dataset
  2. Create word vectors
  3. Measure cosine distance

Words clustering

• Main idea: Cluster words represented by their vector using k-means algorithm

Words cloud

Symptoms:

Organs:

Medications:

Second goal

1. Create feature vector for each paper using BOW model
2. Cluster vectors into coherent groups
3. Visualize clusters in a 2D plot

t-SNE with no labels:

t-SNE with k-means labels:

Interactive plot