The word 'bot' is splashed around all over the internet. This 6-part series attempts to establish what bots actually are — and how you might recognise the misleading ones. This is Part 3, where I give you clear pointers on how to do just that.
Here are some of the ways you can detect a bot:
1) The account's vocabulary is limited, and any text it produces can appear poorly constructed, imprecise, or nonsensical.
2) The account tweets & retweets constantly, but no-one speaks back to it. Conversely, the account might have little to no original tweets of its own, yet is consistently being mentioned by other accounts. The same goes for Facebook comments and shares; these dynamics could signal a network of bots.
3) The account's reposting & retweeting patterns seem either too systematic, or too erratic. Think bursts of activity, then abrupt bouts of silence.
4) The account gives an abnormally high number of likes. This is because a great deal of bot accounts are programmed to instantly follow back accounts which 'like' their content. It's another way of rapidly building a network which, though artificial, seems grassroots to the untrained eye.
5) The account reacts impossibly rapidly; its replies to mentions or retweets appear within a split second, faster than would be humanly possible.
6) The account's activity also follows non-human cycles:
- Tweets at bizarre hours of the night;
- Shows very little variety in the topics it engages with;
- Displays selective focus, for instance, staying dormant outside of political events.
7) The accounts' profile picture is either the platform's default image or a graphical image. The profile picture could also clearly not relate to an individual - a meme, an advertisement, a model, a slogan or tagline.
8) The account has a significantly imbalanced follower-to-following ratio: bot accounts often follow far more people than there are people following them.
9) On sites like Reddit, a bot user might systematically downvote certain types of content.
10) The account has an enormous output. The number of messages the average human releases is far below what some bot accounts achieve.
Several studies have tried to find formulas for detecting bot accounts or bot activity. Some of the formulaic principles are understandable even if you're not a maths wizard. For example: let's filter accounts based on skewed follower-following ratio; erratic activity; and high volume of tweets. Thus, any account which flags up two or more of the above characteristics may be a bot.
Other mathematical approaches to bot detection are more complicated. Earlier this year, Professor Dr Stefan Stieglitz from the University of Duisburg-Essen headed an algorithmically complex study into bots. One of its important ﬁndings was that "bots generated a new follower with every second tweet." This "may justify and explain the purpose behind an extremely high bot activity". Another reason for programming bot accounts to post extremely frequently is that the risk of someone noticing such an unnatural rate is smaller than the benefit gained by swarming platforms. Indeed, there's evidence that how many bot accounts you deploy matters less than how much your bot accounts post.
The next instalment of this series is Part 4: Bot Power.