25 billion actions a day or 65,000 actions a second! That is the volume of actions generated by Facebook’s 800 million users. And Facebook this week, released some information about its massive Security infrastructure called the Facebook Immune System or FIS that scans all of these actions for any kind of suspicious activities.
As New Scientist explains,
It protects against scams by harnessing artificially intelligent software to detect suspicious patterns of behaviour. The system is overseen by a team of 30 people, but it can learn in real time and is able to take action without checking with a human supervisor.
The system was developed over a three year period and the numbers released by Facebook shows that it has been pretty effective. The number of users affected by spam has been reduced to less than 1%. Even though that 1% accounts for about 8 million users, with a little bit of caution from the end user while using Facebook, that number can be reduced even further.
Microsoft Research has put forward a PDF detailing the principles of FIS. According to it, the main components of FIS are
â€¢ Classifier services: Classifier services are networked interfaces to an abstract classifier interface. That abstraction is implemented by a number of different machine-learning algorithms, using standard object-oriented methods. Implemented algorithms include random forests, SVMs, logistic regression, and a version of boosting, among other algorithms. Classifier services are always online and are designed never to be restarted.
â€¢ Feature Extraction Language (FXL): FXL is the dynamically executed language for expressing features and rules. It is a Turing-complete, statically-typed functional language. Feature expressions are checked then loaded into classifier services and feature tailers1 online, without service restart.
â€¢ Dynamic model loading: Models are built on features and those features are either basic or derived via an FXL expression. Like features, models are loaded online into classifier services, without service or tailer restart. As well, many of classifier implementations support online training.
â€¢ Policy Engine: Policies organize classification and features to express business logic, policy, and also holdouts for evaluating classifier performance. Policies are Boolean-valued FXL expressions that trigger responses. Policies execute on top of machine-learned classification and feature data providers. Responses are system actions. There are numerous responses.
Some examples are blocking an action, requiring an authentication challenge, and disabling an account.
â€¢ Feature Loops (Floops): Classification generates all kinds of information and associations during feature extraction. The floops take this data, aggregate it, and make it available to the classifiers as features. The floops also incorporate user feedback, data from crawlers2, and query data from the data warehouse.
Although FIS has come a long way in tackling spam, it should be noted that FIS is still vulnerable to tactics that are new to it, such as, socialbots. A socialbot works by sending friend requests to random people. The profile data of people who accept this friend request is used for identity theft, phishing attacks etc.
So, it is always up to the end user to remain cautious of these types of attacks in order to protect their personal information.
You can find some of the common tips to protect your Facebook account here.