Posted by stefan
on November 29, 2014
In this paper we try to identify characteristics of influential users on Twitter using attributes related to community structure, network position, text quality and user activity. [PDF]
We pursue a supervised approach and collect a dataset containing influential and non-influential users. Human annotators assigned the labels to each tweet and user.
You can find some of the code we used to extract the attributes here on my page.
In case you have any questions, contact me.
The code provided here [https://github.com/fensta/linegraphcreator] modifies the algorithm provided by Evans et al.  to transform a graph into a line graph and detect communities. It allows to apply the aforementioned algorithm also to large graphs that don’t fit into memory.
I have implemented it for directed graphs so far, but it is straightforward to adjust the code for the undirected case. I might add that code later on. I tested the code against the original implementation on several datasets successfully, but the code might still include some bugs.
If you have questions or suggestions for improvement, feel free to contact me.
I used this algorithm in my paper “A Framework for Validating the Merit of Properties that Predict the Influence of a Twitter user” [see here for the paper]
 – Evans, T. S., & Lambiotte, R. (2009). Line graphs, link partitions, and overlapping communities. Physical Review E, 80(1), 016105.