In a previous post we have shown how discussion patterns in talk pages associated to Wikipedia articles can be studied to identify conflict, while in another one we have applied these measures to characterize Wikipedia articles related to the climate change controversy.
A complementary approach, based on analyzing the edit history, has been more often adopted in literature to quantify conflict. In particular, several researchers focused on reverts, corresponding to edits in which a user undoes the changes introduced by another user.
In this post we will illustrate a recent study (Yasseri et al., 2012), which introduces a measure based on reverts to detect edit wars and controversial articles. According to the authors, reverts can be associated with conflict, but also with vandalism; the aim of their metric is to account for reverts which indicate conflict and not vandalism.
For each revert action, the total number of edits done to the article by each of the two users involved is measured, and the minimum of the two is taken as a weight for this action. The rational behind this choice is that if at least one of the two users has performed only a few edits, the action is more likely to correspond to vandalism, and will have a much lower weight. Instead when two users, who have both worked more on the page, revert one another’s changes, it is much more likely that a real conflict is going on. The measure is refined by multiplying it by the number of editors who reverted mutually, and by eliminating the topmost mutually reverting editors.
Studying the evolution of this metric over time, the authors distinguish three categories of articles, described below. For each category, a picture shows the evolution of the controversy measure (M) versus the corresponding number of edits (n) done to an example article for that category.
Consensus: growth of the controversy measure tends to start slowly, then it accelerates until it starts decreasing, and at the end consensus is reached.
Sequence of temporal consensus: multiple cycles of war and consensus like the one described in the previous case follow each other; the cycles can be initiated by endogenous or exogenous factors.
Never-ending wars: no consensus, either permanent or temporary, is ever reached, and the article content is object of continuous disputes among editors.
Finally, according to these three categories authors classify the articles having an overall value of M > 1000 (i.e. the around 12k most disputed articles, according to this measure). As it can be observed in the figure below, articles with the highest values of controversy are all characterized by never-ending wars, while for lower values of M it is increasingly more likely that consensus is reached; the pattern of multiple temporary consensus is only observed for a minority of articles.