Subgraph Classification, Clustering and Centrality for a Degree Asymmetric Twitter Based Graph Case Study: Suicidality

TL;DR

This study analyzes Twitter graphs related to suicidality, revealing asymmetric degree distributions, low clustering, and sentiment negativity, providing insights into social media communication patterns on sensitive topics.

Contribution

It introduces an analysis of degree asymmetric Twitter graphs on suicidality topics, highlighting topological features and misinformation aspects using NodeXL.

Findings

High in-degree to out-degree ratio of 4:25 with power law distribution

Low global clustering coefficient of 0.038 and density of 0.00034

Presence of bridging vertices indicating sparse community structure

Abstract

We present some initial results from a case study in social media data harvesting and visualization utilizing the tools and analytical features of NodeXL applied to a degree asymmetric vertex graph set. We consider twitter graphs harvested for topics related to suicidal ideation, suicide attempts, self-harm and bullycide. While the twitter-sphere only captures a small and age biased sample of communications it is a readily available public database for a wealth of rich topics yielding a large sample set. All these topics gave rise to highly asymmetric vertex degree graphs and all shared the same general topological features. We find a strong preference for in degree vertex information transfer with a 4:25 out degree to in degree vertex ratio with a power law distribution. Overall there is a low global clustering coefficient average of 0.038 and a graph clustering density of 0.00034 for Clauset-Newman-Moore grouping with a maximum geodesic distance of 6. Eigenvector centrality does not give any large central impact vertices and betweenness centrality shows many bridging vertices indicating a sparse community structure. Parts of speech sentiment scores show a strong asymmetry of predominant negative scores for almost all word and word pairs with salience greater than one. We used an Hoaxy analysis to check for deliberate misinformation on these topics by a Twitter-Bot.