Analysis of User-Journeys on a Crowdsourcing Platform

sven_naef

Hello everyone,
I am researching the behaviour of Crowdsourcees on a large crowdsourcing platform. Therefore I got a lot of data and already analyzed quite a bit. For example, I used Prom to analyze my eventlogs and bring them to a process model (Trace, Eventlog and Activity).

As a next step I would like to analyze the impact of gamification of on the Crowdsourcees behaviour. There are eventlogs with badges and points. What i would like to see is whether the behavior of a Crowdsourcee changed after he received certain non-monetary gamification incentives. Does ProM allow such an analysis in a certain way? For example the frequency of eventlogs changed after certain incentives etc. OR in general, how do i bring in more attributes for each user like role (admin, moderator, user) in each case to later divide between certain types?

I hope this question is not out of context and any input is appreciated (papers, tools, etc). I already had a look at both MOOCs (coursera, futurelearn) and read quite many papers. Further research will include also the overview of crowdsourcee interacting with the crowdsourer etc. Maybe I have to consider using another tool? 

Best regards,

Sven

kfgz087

Hi Sven,

Thanks for your question. My colleague who may be able to help you already got the question. Please wait for a while.

Massimiliano

Dear Sven,

I think that the "Feature Prediction" plug-in will serve your purpose.
Unfortunately, the tool is not discussed in the coursera MOOC, nor in the process-mining book by Wil van der Aalst.

Through this plug-in, for example, you can correlate if the value of a certain attribute in the process has some influence on the activities that are performed. Or, also, you can correlate if the execution of certain activities in a process will trigger the execution of other activities. Assuming that your incentives are stored in the event log as an event type (i.e. an activity), you can correlate how the execution of that activity influences the number of executions of another activity (of course, if this correlate to 0 executions of another activity, this means that that latter activity is not executed). The correlation analysis builds on either decision-tree or regression-tree analysis (you can choose).

For further information, refers to the paper available at http://dx.doi.org/10.1016/j.is.2015.07.003 (if you have problem to access this paper, please contact me by email)

A screencast is available at

https://svn.win.tue.nl/repos/prom/Documentation/FeaturePrediction/screencast.avi. Unfortunately, it refers to an older version. Still, you won't find problem to relate it to the new version.

The latest stable version of the plug-in is available, e.g., in ProM Lite.

Regards,
Massimiliano de Leoni
Assistant Professor @ Eindhoven University of Technology

sven_naef

Dear Massimiliano

Thank you so much for the fast answer and the detailed description. I will work with your suggestions and report back about my progress.

Best regards,

Sven