The algorithm of chat frequency alone can identify potential cyber offenders
Researchers believe that with the help of AI, system administrators can better maintain network security and user rights.
Although the current AI cannot further predict the specific types of illegal events, it may be able to catch "invisible" offenders on the Internet and better protect our safety.
◎Sun Linyu
With the popularization of the Internet, network violations have also become a social issue that cannot be ignored.
The Internet eliminates the space-time distance between potential offenders and victims, making everyone have objective conditions to violate the law, and everyone is in danger of being harmed.
The "Characteristics and Trends of Cybercrime (2016.1—2018.12)" published by the China Judicial Big Data Research Institute pointed out that social platforms, especially QQ and WeChat, have become the main tools of virtual crimes. Commit crimes.
This process does not require physical contact, so it is very difficult to capture and brings many difficulties to law enforcement.
Recently, computer researchers from Tokushima University in Japan and Cyber Agent, an agent of a large Japanese Internet company, published a paper on "Human Behavior Computing". They used machine learning to analyze the usage data of a social game under Cyber Agent. And without monitoring the chat content, based only on basic information such as the number of chats, chat partners, chat time, etc., potential network offenders can be more accurately identified, and the approximate time of the illegal behavior can be predicted.
The Theoretical Basis of "Suspect Tracking"
This is not a whimsical idea.
Although everyone only relies on network cables to communicate in the game, our online behavior also leaves a huge amount of data, which provides a wealth of materials for predicting network violations.
Researchers developed this algorithm based on two traditional criminological theories: daily activity theory and social contagion theory.
The theory of daily activities puts forward that many crimes do not happen randomly, and the offender and the victim often overlap in their daily activities.
For example, in real life, the thief would go to the target location to step on the spot before the theft, and observe the behavior of the target person; similarly, the criminals on the Internet need to get in touch with the "prey" in advance to gain trust.
Therefore, there may be "criminal warnings" hidden in the player's social activity data.
In addition, the theory of social contagion adds an important point: illegal tendencies or illegal behaviors can also be contagious.
The most common example is cyber violence.
Cyber violence often comes from the widespread spread of a certain kind of extreme emotion: under the coercion of the group, some people unknowingly lose their ability to make independent judgments, and unwittingly become online perpetrators.
Some studies have pointed out that after "witnessing" other people in the group's online harassment behavior, bystanders can easily attack the same victim or try to harass other people.
Such contagious behavior also provides important objects and time clues for predicting network violations.
Based on these two theories, the researchers chose a mobile game called Pigg Party.
It features social functions. After logging in to the account, users can dress up virtual rooms and personal images, and communicate with friends or strangers in private, group, and public chats.
The researchers used an algorithm that is good at extracting features from complex data-a multi-layer nonlinear model, and analyzed the chat data generated by 550,000 users within 6 months, including each user's chat frequency, chat time, and message recipients.了Analysis.
The heart that wants to do bad things can't escape the eyes of AI
Researchers have combined a variety of neural network models and algorithms to build artificial intelligence that predicts illegal events.
Performance test results show that AI can more accurately predict future offenders and victim accounts based on user data.
Enter the user's chat time, frequency, and object within two months, AI's prediction accuracy for illegal accounts in the next two months can reach 84.85%, and the accuracy of prediction for victim accounts is also close to 85%.
In addition to the ability to predict the risk of illegal or victimization of individual accounts, only by providing user activity data within a week, AI can basically accurately predict the time of illegal events in the online community in the next week, with respect to hours and hours. The prediction accuracy rate of the date is as high as 95.83% and 85.71%, and the results are consistent with the time given for predicting the victimization.
What’s more interesting is that after AI analyzes the data, the time when an illegal incident occurred is not necessarily in the time period when the illegal incident occurred in the past. It can be seen that it does not only master the fixed rules, but also the true nature of the offender’s words and deeds. "logic".
The AI capable of illegal prediction compresses the massive and scattered daily activity records of users into data that can be quantitatively analyzed, and extracts and understands the laws from it, ultimately forming a powerful predictive ability.
Researchers believe that with the help of AI, system administrators can better maintain network security and user rights.
Although the current AI cannot further predict the specific types of illegal events, it may be able to catch "invisible" offenders on the Internet and better protect our safety.
(According to "Global Science")