If you are still interested: newer versions of DokuWiki (I think since spring last year) have index files where all links are stored which makes it relatively easy to extract the graph structure. The include plugin does also writes link information for the included pages which means you would even get a) the relation to the included pages and b) the links inside the included pages in the parent page. There is of course also the possibility to write other information into such index files so we could easily extract them later.
For my bachelor's thesis I've written a small C++ program which extracts content from these files and export them into the Metis format (it was the basis for the
10th DIMACS Implementation Challenge - Graph Partitioning and Graph Clustering). If you should be interested, my bachelor's thesis was about cut-clustering, it can be found
here (pdf). I have an implementation of this cut-clustering algorithm for this graph format and also a converter to a format which is supported by a modularity-based clustering algorithm (though I haven't implemented that algorithm myself and afaik it isn't freely available). I have also small programs which convert the output of these algorithms to GraphML so they are compatible with the
yEd graph editor so you can view the clustering in yEd. I could certainly provide these helper programs, my own implementation also shouldn't be a large problem (originally there were plans to publish it anyway) but I'm not sure about the modularity-based clustering algorithm. In general if you have any ideas what we could analyze please tell me, I'm interested in this topic and maybe I'll find some time this summer to implement anything that's still missing. Maybe it would also be interesting/helpful to port some of these algorithms e.g. to JavaScript or PHP so they could be used inside DokuWiki without external dependencies. The cut-clustering algorithm is relatively complex to implement (unless there is already a good graph algorithms library with some more fundamental algorithms e.g. for calculating minimum cuts) but I think the modularity-based algorithm is simple and fast enough to be implemented in PHP or JavaScript. There is already a JavaScript-based graph visualization library called
InfoVis which does already implement a graph layout algorithm. I don't know how well it works with larger graphs but it might be a start.