What is the network of Game of Thrones on Twitter?

The Game of Thrones is such a popular TV show and has a massive impact on social media. There are about 8.2M followers on twitter. It might be interesting to all the fans in the world to know the followers’ network of @GameOfThones.

The data is collected using Twitter API. In this post, we only collect the top 50 friends of GameOfThones twitter account and the top 50 friends of them. A total of 2,500 accounts is collected as the nodes in the network. A bi-directional edge is formed when there is a mutual following between two accounts. We use the NetworkX Python package to construct and export the network as a graphhml file of 146KB. Then we use the NetworkX and Gephi to perform a series of network explorations.

After importing the graphhml file into Gephi software, We first adjust the layout of the graph via Fruchterman Reingold and Force Atlas. Then we show the lable (account name) of the nodes and scale it by the degree of the nodes. Meaning the nodes with larger degrees will have larger labels. Then we adjust the layout using “label adjust,” so the texts are not overlapping on each other. Then we color the edges by closeness centrality, and we have the final image shown below:

In this graph, we see that gameofthrones is connected to all the 50 accounts, since all those 50 accounts are the friends of gameofthrones. However, several accounts are closer to the center nodes, and they have the most shortest paths in the network, including TwitterTV, euphoriaHBO, MarissaBlanch, Zendaya, soup_city, BellaRamsey, etc.

We also produce the degree distribution of the network; the average degree is 2.132. The majority of nodes have less than 5 degrees, and only 50 nodes have more than 50 degrees.

They are The BellaRamsey, euphoriaHBO, TwitterTV, Zendaya, trvisXX, _richardmadden, MarenMorris, ThorBjornsson_ emiliaclarke, ATT, khloekardashian, TPAIN, daemonsanddust, CCXPoficial, VladimirFurdik, Comic_Con, HBODocs, JayREllis, omaze, sza, florencemachine, daisie, ava, shakeshack, thelumineers, RadioCity, nbcsnl, MountainDew, magicleap, johnvarvatos, roshanabbas, treecarving, jeffsotoart, allyson, AaronRodgers12, KChenoweth, MarissaBlanch, maweezyRN, DiscoverIreland, gameofthrones.

The Diameter of this network is four, which means from any node, within four edges, one can reach any other node in the network.

The node with the most shortest paths in the network is the TwitterTV, it means that TwitterTV can spread information to all the other nodes in the network most efficiently.

In Gephi, using Yifan Hu proportional layout to adjust the network, and apply betweenness centrality as the color to the edges. We can identify the bridges in the network: gameofthrones, khloekardashian, daisie, ava, roshanabbas, jeffsotoart. Meaning those Twitter accounts connect communities in the network.

Coloring the edges with the highest Eigenvector centrality, we have the account: gameofthrones, TwitterTV, euphoriaHBO, BellaRamsey, Zendaya, daemonsanddust, kristoferhivju, ATT, VladimirFurdik, emiliaclarke, MarenMorris, JayREllis, _richardmadden, MarissaBlanch, trvisXX, TPAIN, RadioCity, CCXPoficial, Comic_Con, magicleap. Meaning those accounts connect to very influential accounts, which makes them influential too.

Since we have many centrality measures, what are the consensus among all the measures?

We normalized the node degree and betweenness centrality to be between 0 and 1. Then the total score is calculated by:

total_score=degree_standardized+between_centraility_standardized+closeness_centrality+harmonic_closeness_centrality+eigen_centrality

The calculation is done in the excel sheet showing below:

We see that theluminieers shows up higher than discoverireland. Interestingly, the band has made songs inspired by Game of Thrones.

Coloring the edges with the Modularity Class, we can recognize communities in the network in the figure below:

Conclusion:

We use the NetworkX and Gephi to analyze and visualize the network of twitter account @GameOfThrones’s followers network (a sample of 2,500 accounts). We explore multiple centrality measures to find the influential accounts in the network besides the gameofthones account. We also calculate a total score to find the most important accounts across five different centrality measures. Then we find potential communities in the network.