3
$\begingroup$

I am looking for a way to analyze typical user behavior with a very large number of events. The events have names, user id's and other non-vital attributes. It would be very helpful for seeing use patterns if we could make a graph, similar to the following mock-up: Flow Graph

There are obviously flaws with this graph, but I think the idea is clear. Essentially, given a particular event, what events are most likely to occur next and what are their percentages respectively. It seems like a very doable task, but using industry tools like Tableau, I have been unable to find a solution.

$\endgroup$
2
  • $\begingroup$ Can you assume that people's next action only depends on where they are now? $\endgroup$ Oct 21, 2015 at 20:19
  • $\begingroup$ Have a look at Disco: fluxicon.com/disco $\endgroup$
    – timbp
    Oct 21, 2015 at 22:12

2 Answers 2

2
$\begingroup$

What you are looking for is Process Mining. Process mining takes a event log (a sequence of events grouped by some ID) and derives a process that could produce that event log.

You can use this for simply discovering a process and learning about it (and then improving it), or to compare the discovered process with a documented process to look for variations (people not following procedure, or events that are not covered by the specified process).

There is free open-source software for processing mining. PRoM (http://www.promtools.org/doku.php) is the main one. More user friendly is the commercial software Disco by Fluxicon (https://fluxicon.com/disco/).

$\endgroup$
1
  • $\begingroup$ This looks like a great tool. Checkin on price, but my only current complaint is that it does not support SQL connections. $\endgroup$
    – David
    Oct 22, 2015 at 2:09
1
$\begingroup$

The combination of a network analysis software tool such as Ucinet exported to a visualization tool such as Pajek would probably work. Both are free to download...

https://sites.google.com/site/ucinetsoftware/downloads

http://mrvar.fdv.uni-lj.si/pajek/

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.