People spend great amounts of time on social networks. As a result, this has made them the perfect hunting grounds for criminals, to acquire targets via fake accounts. Fraudsters employ bots in social networks and other electronic communication means, such as email, where scam frauds and other criminal activities are triggered by the fact that bots are imitating human behavior when contacting users at a large scale.
Detection of social botnets is an active area of technological research and development. Due to the relatively recent raising of this sub-field of cybersecurity and data treatment, and its close relation with human and social sciences, there does not exist a single standard procedure to deal with them, so different criteria can be followed. Several of them are usual choices and keep on being refined every day.
• AI: Machine learning algorithms, particularly deep learning approaches, are used to detect social botnets. These algorithms analyse large datasets of social media activity to identify patterns and anomalies associated with bot behaviour, ranging from different combinations of data to a direct crawling of the platforms. Examples of works can be found in Multilingual Language Models, which mathematically modelize the particularities of spoken languages (ChatGPT or Bard being the most prominent today, or Duolingo), and Contextual string embeddings, that represent the semantics (meanings of words and expressions) in operational numeric vectors.
• Detection systems can analyze the behaviour of accounts in social media rather than relying solely on account attributes. This includes examining posting frequency, content types, interactions, and the temporal aspect of activities.
• Network analysis techniques are frequently applied to study the connections between accounts in social botnets. Identifying clusters of interconnected accounts with similar behaviours can indicate the presence of a botnet. This aspect is very widespread, particularly in social media with massive amounts of posts, such as Twitter/X .
• Analysing the content of posts and messages for characteristics common to bot-generated content, such as excessive use of certain keywords, the absence of personal information, and low-quality language, is another approach. The linguistic patterns in their posts and messages can be a valuable cue for detecting malicious actors.
Most works are specialized to deal with one or a small subset of these aspects, destined to cover a particular solution. Sometimes, they can be blended, e.g. content analysis by machine learning.
However, while automated solutions dealing with these different topics exist and are always sought and researched, human monitoring keeps being necessary, due to the abundant raising and deletion of fake profiles, and especially variations in their patterns happening with frequency.
The increasing accuracy of technology and creativity of engineers and data scientists are, in spite of all, a hopeful sign of the success of detection of social media in the very near future.