Word graph

‍

Word graph is a local graph, which has a selected word in center and top-N words around, most commonly used in articles with selected word. Each node is provided with at least one link leading to appropriate article tab with article-wide graph and ability to request access to this article.

‍

‍

To compute word graph, we search for most connected words for a given word. Here is the code implementation:

‍

‍

Edges conflicts

‍

In cases, when links in edges conflict with each other, like “[Alcohol] enhances [organism]” and “[Alcohol] is harmful for [organism]”, according to 2 different articles, we can mark it with different colors in order to make graph more representative and involve users into research process.

Here are 2 good approaches:

‍

VADER (NLP method)

‍

‍

Pros: instant speed

Cons: not accurate

‍

HuggingFace Transformers

‍

‍

Pros: accurate

Cons: not fast (5+ seconds), uses 200MB Transformers model