Twitter is a dynamic world that works in a confined space of 140 characters but the impact it can have on an audience is huge.
Just like fake news, fake followers can be detrimental to the image of businesses and even to those individual users who rely heavily on social media marketing. Since Twitter places significant constraints on the type of communication that is possible, it becomes easier for bots to reconstruct the human behavior that is demonstrated in the limited Twitter dimension.
Hence, results that we obtain in the form of social media metrics can often get skewed.
So why is the detection of Social Bots important for social media users?
Simply put — a wrong target audience could adversely impact engagement rates and for paid advertisers, this could mean huge revenue losses (as misconstrued advertising may not yield the desired results).
An early bot detection coming from a social user not only wastes time and energy but also directs relevant messages to end users that are not real. Eventually, this impacts sales since no revenue is generated from bot accounts. By imitating humans, some social bots are designed to manipulate discussions and even alter the popularity of users. Such actions can not only infect the content but also spread misinformation in the form of trending tweets for social media giants like Twitter.
Twitter bots as followers could prove detrimental to a genuine account since Twitter doesn’t encourage promoting accounts with a help of spam techniques.
Simple Ways to identify a Bot
Quite often social bot profiles mention that the account is being operated by a bot. Under such circumstances, followers are very aware of who they are following. But many times, the social game can get tricky when there is no mention of the type of account of a user.
For individuals and small businesses that are using social media metrics for marketing and advertisement, this part could get tricky.
For starters, a social bot can get easily recognizable if on a specific tweet, somebody starts following and immediately replies within microseconds.When replies are lightning fast (where they haven’t read the tweet or the link posted in the tweet), chances are that the user is interacting with a social bot. But this alone may not be the only way to identify a social bot.
Coupled with a few other indicators, a Twitter or an Instagram user (human) can easily gauge at the chances of a follower turning out to be a social bot.
The oldest trick that is most common across most social media platforms is the game of follow/unfollow. Once a photo or a tweet is posted, chances of a social bot immediately following are significantly high. The following by a bot on Twitter could be based on the hashtag used in our tweets, the time of the tweet or a pertinent topic that a bot is tagged to.
A social bot will start following you and if you don’t follow back (or even if you do) they will most likely unfollow you within 24 hours.
Usually social bots have a huge number of followers but tend to follow few people (since they are auto tuned to follow and unfollow within a time frame).
Having said that, some bots are not built to be very smart and may also duplicate pictures that could be of either a celebrity or not really unique. Another pretty solid sign is if the tweet originates ‘from API’. A human user would tweet from the web or mobile phones but a user ‘from API’ would potentially mean that the tweets are automated.
If the same tweet with the same language appears on other profiles, then it is highly likely that the human user is dealing with a bot.
But most of these bot indicators suffer drawbacks.
For example, a huge celebrity that operates his/her own account could have a huge amount of followers on social media but may be following few people. Similarly, pictures that are not unique could be merely fan pages, operated by a group of people who are fans or an individual user.
This is where data analytics play a very crucial role.
Data Analytics and Social Bots
With the plethora of data available on social media, sifting a social bot from a genuine human user may require a series of actions.
Such websites allow a user to identify fake followers that are potentially social bots. Algorithms allow detection of social bots, by typically classifying them into different categories. BotOrNot’s classifier uses supervised learning method called the Random Forest. The extracted features are designed to train with both the identified bots and human accounts so as to build a heuristic decision tree model that nullifies bias.
BotorNot looks at six interesting features that it uses to identify a user and once identified allows a human user to block the fake followers by giving a score.
The higher the score is, the more is the likelihood of it being a social bot.
i. Network: This feature looks for information diffusion patterns and creates networks based on re-tweets, mentions, and hashtag co-occurrence. By extracting important statistical features like degree distribution, clustering coefficient, and centrality measures, identification through network classification becomes simpler.
ii. User: This classification relies on Twitter meta-data related to a specified account, and looks at language, geographic locations, and account creation time.
iii. Friends: The friend’s feature compiles descriptive statistics that include the median, moments, and distributions of their number of followers, following, posts etc.
iv. Temporal: By sequentially capturing patterns of generated content and consumption, like tweet rate and inter-tweet time distribution, temporal feature adds an additional filter.
v. Content: Based on linguistic signals, natural language processing, especially part-of-speech tagging is another metric that BotorNot considers during its evaluation.
vi. Sentiment uses Twitter specific sentiment analysis algorithms, including emotions.
It won’t be wrong to say that social bots exist and are very close to humans in their interactions.
But the bigger problem is their influence on ideas which could potentially impact big events since their automated tweets and twitter activities tend to skew algorithms including the trending content.
Now whether a ‘super smart’ bot will supersede us in the near future remains a mystery (and probably best left for another story) but for now — we should be more focused on separating breaking news by a human user from that which is auto-generated by a social bot.