Wikipedia edit number prediction based on temporal dynamics only
|Wikipedia edit number prediction based on temporal dynamics only|
|Citation:||missing journal missing volume : missing pages. 2011 October|
|Web:||Bing Google Yahoo! — Google PDF|
|Article:||BASE Google Scholar PubMed|
|Restricted:||DTU Digital Library|
|Extract:||Talairach coordinates from linked PDF: CSV-formated wiki-formated|
The system came in third in the Wikipedia Participation Challenge.
It was an entry in the Wikipedia Participation Challenge and used data from this competition.
They train a supervized learning model with data from the edit time series with the target also extracted from the edit time series, - what they called "self-supervized learning". They use no other feature than the edit history.
They use a "log(1+a)" link function and a square loss function.
They use Python scikit-learn and OpenCV and tried OLS, SVM, KNN, ANN, GBT with default parameters Gradient Boosted Tree outperformed the other methods and they then explored (hyper)parameters of the GBT model.