The Social Network of Twitter Bots
It’s no secret that modern social media is cluttered with bots. While some may be helpful, many are malicious or harmful. For example, the many bots that post phishing links disguised as promotions in Youtube or Facebook comments. This paper, by researchers at the Center for Complex Networks and Systems Research in Indiana University and Information Sciences Institute in University of Southern California, examines bot social networks in one platform–Twitter. They have created a Twitter bot detection framework that is freely available online which can analyze an account and assign it a score on how “botlike” it is. The higher the score, the more likely the account is to be a bot. The framework does this by using data from the account such as analyzing friend networks (people who are mutually following each other) and the networks created from retweets, mentions, and hashtags.
In class, we discussed some characteristics of social networks and the graph theory associated with them. In this framework, retweet and mention networks are created by making users nodes. These user nodes have directed links between them, representing the flow of information. The link points towards the user retweeting a tweet or being mentioned by Twitter handle. The hashtag network uses undirected links instead of directed links. Tweets create hashtag nodes when two hashtags are in the same tweet. In all of these networks, the links are given a weighting depending on popularity/number of interactions on the tweet. From these networks, the researchers were able to conclude that “bots use different retweet and mention strategies when interacting with humans or other bots”, which helped in training their model to distinguish between human and bot.
These are only a few of the 1000+ features that go towards the ranking of how likely an account is of being a bot. After testing the framework on a large collection of Twitter accounts, the researchers have concluded that between 9-15% of accounts on the site are bots and not real users. Many of these bots network within each other–“bots tend to follow other bots and they are mostly followed by bots”. Although we will most likely never be rid of them, hopefully research like this and the use of analysis using social networks can aid in the detection and removal of malicious bots from social media sites.