Yiran Qiao, Jing Chen, Xiang Ao, Qiwei Zhong, Yang Liu, Qing He
The study introduces AC-MIL, a novel framework for assessing risks in live streaming by analyzing user behaviors and coordination patterns, significantly improving detection accuracy and interpretability over existing methods.
Live streaming platforms are popular for real-time interactions, but they face challenges with detecting malicious behaviors hidden among normal activities. This research presents a new approach called AC-MIL to assess risks in live streaming rooms, even when only general room-level data is available. By analyzing patterns in user actions over time and among groups, AC-MIL can predict potential risks more accurately and provide explanations for identified risky behaviors. This method has been tested using data from Douyin, a major live streaming platform, and has shown impressive improvements over previous techniques.
