A Study on Measurement and Classification of Twitter Accounts

Twitter is a new web application playing dual roles of online social networking and microblogging. In this paper, we have studied the problem of automation by bots and cyborgs on Twitter. As a popular web application, Twitter has become a unique platform for information sharing with a large user base. However, its popularity and very open nature have made Twitter a very tempting target for exploitation by automated programs, i. e. , bots. The problem of bots on Twitter is further complicated by the key role that automation plays in everyday Twitter usage. Based on the data, we have identified features that can differentiate humans, bots, and cyborgs on Twitter. Using entropy measures, we have determined that humans have complex timing behavior, i. e. , high entropy, whereas bots and cyborgs are often given away by their regular or periodic timing, i. e. , low entropy. In examining the text of tweets, we have observed that a high proportion of bot tweets contain spam content. Lastly, we have discovered that certain account properties, like external URL ratio and tweeting device makeup, are very helpful on detecting automation.