covid19-research.network PREVIEW
FIND ARTICLES
Search by keywords in abstract
FILTER BY CATEGORIES
FILTER BY PUBLICATION TYPE

network mode

Displaying the latest:

1000 peer-reviewed articles 500 preprints

text mode

Displaying the latest:

1000 peer-reviewed articles 500 preprints

Welcome to the initial preview of the covid19-research.network, an interactive platform that parses and visualizes the latest COVID-19-related scientific findings in an easily accessible interface.

It provides the latest peer-reviewed articles from PubMed as well as preprints from medRxiv, bioRxiv and arXiv.

What do I see when I look at this network?

Every node is an article. Links between articles are drawn based on the semantic similarity of their abstracts. The relative positions of nodes in the network are given by a force-directed layout, which groups articles into clusters that correspond to the sub-topics in the COVID-19 literature. The nodes are colored according to categories found and hand-labelled using a LDA-based topic model.
Imprint | This project is funded by an EOSC COVID-19 Fast Track Grant.

How do I start?

Press the network button on the upper left to enter the network mode.

You will see a network representation of the latest 1500 articles
(1000 peer-reviewed publications and 500 preprints).


Click on a node and highlight it as well as all other articles that are similar to it.

Enter the text mode on the upper left to see a classical list view of the selected articles.

Apply the filters in the command palette on the left to find relevant articles for you.
Imprint | This project is funded by an EOSC COVID-19 Fast Track Grant.

Data acquisition

The latest COVID-19 related research articles are collected daily from PubMed (peer-reviewed articles and letters), bioRxiv, medRxiv and arXiv (preprints).

Categorization

After preprocessing, a LDA topic model is trained on the whole corpus. The resulting topics are labelled by hand.

Network creation

Every node in the network is an article. They are linked based on the cosine similarity of the tf-idf vectors of their abstracts.

Graph layout and partitioning

We visualize the resulting graph using a force-directed algorithm, where the relative positions of the nodes depend on their adjacency (nodes that are strongly connected are close to each other). In this context, dense groups of nodes are expected to correspond to underlying sub-categories in the literature.
Imprint | This project is funded by an EOSC COVID-19 Fast Track Grant.

covid19-research.network

I found a bug!

Note that the initial release works best on Firefox and Chrome on desktop devices.

Please report any bugs to pournaki[at]mis.mpg.de and don't hesitate to get in touch for suggestions and improvements.