Wikipedia edit number prediction from the past edit record based on auto-supervised learning

摘要

This paper describes our approach to the Wikipedia Participation Challenge that seeks to predict the number of edits a Wikipedia editor will make in the next five months. Our approach takes a time series analysis approach in combination with supervised learning, which we call auto-supervised learning. The best model of our solutions achieved a 41% improvement over WMF's baseline predictive model. The result is low accuracy compared with related work but showed 9th place in ICDM contest.