Live streaming has become a cornerstone of today's internet, enabling massive real-time social interactions. However, it faces severe risks arising from sparse, coordinated malicious behaviors among multiple participants, which are often concealed within normal activities and challenging to detect timely and accurately. In this work, we provide a pioneering study on risk assessment in live streaming rooms, characterized by weak supervision where only room-level labels are available. We formulate the task as a Multiple Instance Learning (MIL) problem, treating each room as a bag and defining structured user–timeslot capsules as instances. These capsules represent subsequences of user actions within specific time windows, encapsulating localized behavioral patterns.

Abstract Low‐rate Distributed Denial of Service (LDDoS) attacks, in which the attackers send packets to a victim at a sufficiently low rate to avoid being detected, are considered to be a subtype of DDoS attacks and a potential threat to Internet security. However, an overwhelming attack paradigm on the Internet has rarely been reported due to the harsh requirements for launching LDDoS attacks; therefore, most existing LDDoS attacks are constructed and evaluated through theoretical deduction and/or simulation tests. In this backdrop, the authors aim to figure out what the conditions for launching a successful LDDoS attack are, and how harmful an attack could be. They first analyse the characteristics of LDDoS attacks, and derive the conditions and parameters for initiating LDDoS attacks using a queuing model. Based on the analysis results, an LDDoS algorithm is presented. Then, an LDDoS validation prototype is built on a Network Function Virtualization network to validate the derived parameters and conditions. Finally, a series of experiments are conducted on the testbed, and the results show that a successful LDDoS attack could be achieved based on the derived algorithm; however, its attack effect only lasts for a short time compared with its DDoS counterparts.