0
$\begingroup$

A little context

Social networks such as Facebook, Twitter or Reddit can be represented as dynamic temporal communities. Temporal networks consist of snapshots, representing the state of the network within a given time-frame (for example: nodes = Reddit users, edges = interactions between users, time-frame = 15 minutes). The easiest method to employ community detection on temporal networks is to apply any static community detection algorithm (such as Louvain) on each snapshot and extract the static communities - then use them to extract various information.

The problem

I sampled a dataset consisting of all Reddit comments on the politics subreddit 2 days before and 3 days after Biden's inauguration. I chose a time-frame size of 15 minutes. The problem is any two snapshots in this dataset have a really small number of shared nodes, let's say less than 10% of each's snapshot's total number of nodes. Thus, my snapshots form separate, stand-alone static communities rather than a temporal network.

So, is this normal? Is there any way to sample a set of comments in order to maximize the shared nodes number?

Is there a recommended minimum number of shared nodes between snapshots in temporal community detection?

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.