Who are the the originators (first players) of tracks on the DJ community Mixcloud? Tracks played in common between DJs were investigated, first through the “following” relationship, then across all DJs. The network visualisation helped to suggest the DJs providing a large number of tracks to others, and also identified ome distinct communities of DJs, based on shared tracks.
Mixcloud provides a great web API that enables the retrieval of DJ profiles, following relationships, mixes played, and tracks within the mixes – all in JSON format. A portion of the network was retrieved by starting at the node for the DJ “dubmission”‘s profile, then iterating out based on that profiles’ followers (ie. All of the followers’ followers were processed). For each DJ, their “cloudcast” (mix) feed was parsed, then each mix was retrieved and the tracks played therein were logged along with the date played and the DJ who played them.
This resulted in a dataset of 1223 nodes with 593 edges and 1209 strongly connected components. Figure 1 shows an portion of the filtered network, with the biggest “providers”. Nodes are sized according to their weighted out-degree
Figure 1: Followers - DJs who follow each other with track influences
In the case where the following relationship constraint was removed, the same 1223 nodes had 135,626 edges between them, resulting in 76 strongly connected components. The community detection algorithm in Gephi was run and found four main communities. The highest track providers (highest out-degree) are shown in Figure 2. Nodes with high betweenness centrality in this case are likely to play an eclectic mix of genres – in this case the DJ beyondjazz has the highest betweenness in the dataset
Figure 2: All tracks - DJs with edges showing who played a track first (click for larger version)
The temporal sequence of track plays does not automatically indicate that one DJ is necessarily influencing the other – i.e. that the follower is copying the playlists from those they are following. But when there is a clear pattern of a large number tracks consistently played later than another, this is likely to be the case.
We can see from the diagrams that there are a few cases of one DJ clearly influencing the content played by another, though in the majority of cases DJs are more independent and more likely to be getting influences from other sources outside this community.
The large differences between the edge counts in the two cases suggests that the data on who played a track before you might be better harnessed, for instance to guide follower suggestions. It it likely that tracks-in-common is already used for follower suggestions, though possibly not with a temporal weighting. The temporal factor would be helpful in identifying DJs who have discovered tracks before you.