Wikipedia is amazing. It is so comprehensive that many potential authors cannot find topics on which to contribute an article. But there is debate about the quality of Wikipedia articles, so measures of their quality are of great interest. The authors’ idea is to separate Wikipedia articles into types and use different quality measures for each type. In this paper, they concentrate on two categories of articles: stabilized and controversial. Stabilized articles have only minor changes over time, while controversial articles are subject to a range of opinions. They use two classifiers to separate 96 articles into categories, and used only those articles that were positively classified by at least one classifier.
To measure the quality of stabilized articles, they extract six features--length, citation density, internal link density, external link density, image count density, and section count density--from Wikipedia’s featured articles. For controversial articles, the four criteria are average number of reverts, revisions per registered author, revisions per anonymous author, and percentage of anonymous authors.
Interestingly, the authors evaluate their quality measure by comparing its results to the judgments of readers, a Wikipedia-like approach, rather than to the judgments of experts. Their evaluations of stabilized articles do correlate with those of their student readers, but their evaluations of controversial articles have a much lower correlation. In their introduction, the authors do mention other approaches to quality evaluation; it would have been interesting for them to compare their quality measure with other approaches in this same way to see if their approach is an improvement on previous work.
Student evaluators used one scale for their estimate and a second scale to indicate their confidence. There is no mention of how the final quality estimate was derived from these numbers.