EvidenceFlow is an open-source interactive web application built upon COVID-19 specific literature vetted by the WHO,
for tracking literature trends using alluvial diagrams, projection of influential entities,
and network analysis across different months. The dashboard assists the user to understand the current and upcoming trends in the literature.
The functionality of each tab has been mentioned below.
Description /
Guide
Alluvial Diagram
Alluvial Diagram
Select two or more months from the Add Month section, one by one. Once the months are selected, you can drag and change the
order of these months in the Node/Month list that appears below. Click on the Create Diagram button to visualize the alluvial diagram.
The alluvial diagram helps in tracking the trends in the literature between the selected months. It eases tracing the temporal dynamics
of literature across different time intervals. In the Module Explorer panel on the right, there are multiple features that can help in
better visualization, e.g., by painting all nodes of a selected module.
Multi-level Network
Select a specific month from the "Select Month" button. This illustrates the communities in the networks that are
formed across entities extracted from the literature. More information about this network is given in the paper.
Source-level Network
Select a specific month from the "Select Month" button. This illustrates the source networks that are formed across
diseases entities extracted from the literature. The link between two diseases suggests an association between two entities based
on the cosine similarity.
Embedding Projector
Once we click on the "Embedding Projector" tab, it illustrates the latent space of the low dimensional word embeddings
trained on the literature of COVID-19. The search option allows the user to query the nearest entity present across it. Isolate point
allows the user to isolate N nearest points present around it.
Emerging Trends
This tab demonstrates the forecasted trends for the upcoming months based on the Link Prediction of entities.
Overview
This demonstrates the architecture of our current study. We have also attached the link to the paper in that tab.
Extra Features
This tab cumulates two features, "Text Summarisation" and "Word Algebra".
Text summarisation allows text summary of keywords from the abstract of literature.
It highlights important points related to searched keywords from the extensive corpus of abstracts.
The "Word Algebra" facilitates the linking of dimensional space based on vector algebra.
It instantiates an intuition related to the vector space of the corpus for which the language model has been trained.