Identifying Influential Research that does not stand out at the first glance

This idea rooted from something completely irrelevant, but here is the idea:

It is about identifying influential research, which does not seem influential at the first look.

It works on building a graph, based on references. It assumes a DAG structure, so looping references is assumed to be removed in the preprocessing.

What this analysis firstly does is it bases itself on a basic popular metric to identify seemingly influential research. Let’s call any influential research “Ri”(i is an index). The way we measure influentiality could be by any metric, and let’s call that metric M1. A simple M1 could be number of references that points to that research.

Second and the main step is checking if any seemingly-not influential research (by M1) is actually influential by another metric M2. I’ll describe what M2 is now.

Remember we assume DAG structure. For any Ri, we will check if any other past research Rx is critical or not. While deciding on that criticality, if Rx is somehow an ancestor of more than %50 of Ri’s first-degree-parents, then Rx will be considered critical for the existence of Ri. The metric M2 could be the number of popular articles that an Rx is considered critical for.

%50 threshold in M2 method could be changed / experimented with.

This analysis is probably made before, but it could be interesting to combine it with extra analysis such as “which past topics were M2 critical for today’s hot topics”.

Tldr: It would be interesting to find an article that has a low number of references pointing to it, but somehow that same article is a common ancestor for most of the direct-parents of many influential research.

Machine Learning Engineer at Primer AI | Sharing thoughts on computers and other geeky stuff