"The researchers tracked different users and noted the submissions they made, as well as the tags used on those posts. Taking this data, they could see what tags were frequently used in correlation with one another. This created a “coocurrence network,” which assigns weight to tags based on how often the tag was used and how many different users applied it. With this information, it was possible to conduct a random walk (stepping randomly from one tag to another) and note how tags that occur together can form an otherwise undetectable semantic chain. These tags, based on their association with one another, allowed the researchers to follow along as one popular trend gradually replaced its predecessor. When comparing individual random walks with one another, researchers noted that tags that appear close together in a non-obvious semantic network were likely to be visited by the same user, and tags that were far apart were visited together less often. Although no individual user might be aware of following these obscure connections, they became obvious when the data was examined in bulk." - Steven Perez from Bookmarklet