Extracting and Visualizing User Engagement on Wikipedia Talk Pages

As Wikipedia has grown in popularity, it is important to investigate its diverse user community and collaborative editorial base. Although all user data, from traffic to user edits, are available for download under a free and open license, it is difficult to work with this data due to its scale. In this paper, we demonstrate how consumer hardware can be used to create a local database of Wikipedia’s full edit history from their public XML data dumps. Using this database, we create and present the first visualizations of how editing on talk pages differs between user groups. Our visualizations demonstrate that low quality edits are primarily performed by IP users, rather than blocked users, and that overall engagement with talk pages has plateaued over the last 10 years across all user groups. Finally, we investigate the feasibility of classifying blocked users using this dataset as an example of future research directions. However, we demonstrate the difficulty of this task and find that additional data or a more advanced model would be needed to classify them, as our approach didn’t provide sufficient information to do this. We anticipate that our visualizations and data extraction process are of interest to the community and will provide researchers with the tools needed to use Wikipedia’s valuable data when resources are limited.