This project was inspired by the VisualizingTwitchCommunities and TwitchOverlap projects and attempts to create a similar network graph but instead of looking at the largest twitch streamers it looks at streamers in the NoPixel community.
The list of channels to look at was compiled by looking at HasRoot. The project only monitors streamers who have been active in the past 3 months. For each of those channels, if they are live, we query their list of chatters every 30m. At the end of the day all this data is aggregated into a daily file and at the end of the month these daily files are aggregated into a monthly file. At the end of the month we look at HasRoot activity data for the past month and filter the data to include just those channels.
The nodes file represents all of the channels with size representing the total number of chatters that have been in chat for the past month whenever the chat list was queried every 30m.
The edges file represents chat overlap between two streamers.
The communities file was generated after importing the edge and nodes files into gephi and running modularity with a resolution of 0.25. Modifying the resolution will change the size of communities generated. The community number represents the community a channel was placed in signifying relevant overlap between their communities.
The graph file is generated by importing the nodes and edges to gephi, running modularity, force atlas, and label adjust.
Nodes represent channels and edges represent shared chatters between the two channels. My understanding of how the force atlas algorithm works is the individual nodes repulse one another and the edges attract based on their weight. This means that channels with large overlap should be placed closer together. A node's position is determined by the push and pull of other nodes as well as its edges.
Nodes are colored based on the community they were placed in. I believe a channel can have high overlap with another channel but may not be in the same community because they may not share high overlap with the rest of that community. Some nodes may also appear far away from the rest of their community. This is because the overlap they have with members outside of their community is pulling them away.
I was hoping the graph would look more like the general twitch community overlap graphs with cleaner separation between communities. Using the sample data provided here and running force atlas I was able to see better separation between communities but was unable to reproduce that kind of separation with this dataset. I theorize there are multiple possible contributing factors:
- Unlike the other two projects, this project's current data collection process does not attempt to filter out bots or chatters in multiple chats at once but instead attempts to remove low weighted edges at the very end
- The other graphs use much larger streamers with much larger chats. The average NoPixel streamer is fairly small. So filtering out low weighted edges too harshly may remove all of a node's edges.
- I theorize that the average NoPixel chatter is more likely watching more streams at the same time than a chatter in one of the top twitch streams so that may skew things.
- Because this is essentially looking at data of a subcommunity of twitch, there is bound to be far more overlap than when looking at the top twitch streamers in general which causes there to be far more edges which may be preventing the community clusters from spreading out.