WSDM Cup 2017: Vandalism Detection

• Challenge webpage: http://www.wsdm-cup-2017.org/vandalism-detection.html
• The goal of the competition is to predict whether a wikidata edit will be rolled back or not
• Slides with the solution description: http://www.slideshare.net/AlexeyGrigorev/wsdm-cup-2017-vandalism-detection
• Video (in Russian): https://www.youtube.com/watch?v=cV1rqZG8mUY

Running the solution:

• download all the data, unpack, and put the files to the data folder,
• run 01_xml_to_csv.py for converting the wikidata dump files into a bunch of csv files
• run 02_join_data.py to join the data from the xml files with the meta information and labels
• 03_extract_features.py processes the data so it can be used for the model. This includes
  • specifying the training, validation and testing folds
  • processing the information about the users (including the meta information)
  • extracting useful features from the comments
• 04_train_svm.py creates two models:
  • vectorizer for creating a large one-hot-encoding matrix for all the string features
  • a linear SVM model with L1 penalty for performing the classification
• tira_client.py is used for running the model on http://tira.io/