# US 12,010,358 B2 · Live streaming moderating method and apparatus, server, and storage medium > Agents: this data is available as JSON from the patentgrep API and MCP server — see https://patentgrep.com/llms.txt - Number: US 12,010,358 B2 - Document ID: US12010358B2 - Status: Granted - Kind: B2 - Application number: 18/259,646 - Filed: 2021-12-27 - Published: 2023-12-07 - Granted: 2024-06-11 - Earliest claimed: 2020-12-30 - Assignees: Bigo Technology Pte. Ltd. - Applicants: Bigo Technology Pte. Ltd. - Inventors: Yiyong Li; Zhun Sun; Qiushi Huang; Xue Jing; Wei Xiang - Web page: https://patentgrep.com/patent/US12010358B2 ## CPC - H04N 21/2187: Live feed - G06V 20/46: {Extracting features or characteristics from the video content, e.g. video fingerprints, representative shots or key frames} - G06V 40/20: Movements or behaviour, e.g. gesture recognition (recognition of facial expressions G06V40/16) - H04N 21/23418: {involving operations for analysing video streams, e.g. detecting features or characteristics (television picture signal circuitry for scene change detection H04N5/147; filtering for image enhancement G06T5/00; methods or arrangements for recognising scenes G06V20/00; arrangements characterised by components specially adapted for monitoring, identification or recognition of video in broadcast systems H04H60/59)} - H04N 21/4542: {Blocking scenes or portions of the received content, e.g. censoring scenes} - H04N: PICTORIAL COMMUNICATION, e.g. TELEVISION - G06V: IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING ## Abstract Provided is a method for moderating live streaming. The method includes: performing preliminary violation moderation on a current live streaming frame in a to-be-moderated live streaming room based on a high-accuracy moderation model and a high-recall moderation model that are cascaded with each other; and acquiring a target violation score of the current live streaming frame by inputting a first violation score of the current live streaming frame under the high-accuracy moderation model, a second violation score of the current live streaming frame under the high-recall moderation model, and a multi-dimensional behavior feature of the current live streaming frame in the to-be-moderated live streaming room to a pre-constructed behavior moderation model in response to the current live streaming frame passing the preliminary violation moderation. ## Claims ### Claim 1 A method for moderating live streaming, comprising: performing preliminary violation moderation on a current live streaming frame in a to-be-moderated live streaming room based on a high-accuracy moderation model and a high-recall moderation model that are cascaded with each other; and acquiring a target violation score of the current live streaming frame by inputting a first violation score of the current live streaming frame under the high-accuracy moderation model, a second violation score of the current live streaming frame under the high-recall moderation model, and a multi-dimensional behavior feature of the current live streaming frame in the to-be-moderated live streaming room to a pre-constructed behavior moderation model in response to the current live streaming frame passing the preliminary violation moderation; wherein performing the preliminary violation moderation on the current live streaming frame in the to-be-moderated live streaming room based on the high-accuracy moderation model and the high-recall moderation model that are cascaded with each other comprises: acquiring the first violation score of the current live streaming frame by inputting the current live streaming frame in the to-be-moderated live streaming room to a pre-constructed high-accuracy moderation model; and acquiring the second violation score of the current live streaming frame by continuously inputting the current live streaming frame to a pre-constructed high-recall moderation model in response to the first violation score failing to exceed a predetermined accuracy violation threshold; wherein the current live streaming frame is determined to pass the preliminary violation moderation in response to the first violation score of the current live streaming frame failing to exceed the predetermined accuracy violation threshold and the second violation score of the current live streaming frame failing to exceed a predetermined recall violation threshold. ### Claim 2 (depends on 1) The method according to claim 1, wherein acquiring the target violation score of the current live streaming frame by inputting the first violation score of the current live streaming frame under the high-accuracy moderation model, the second violation score of the current live streaming frame under the high-recall moderation model, and the multi-dimensional behavior feature of the current live streaming frame in the to-be-moderated live streaming room to the pre-constructed behavior moderation model comprises: acquiring a live streaming behavior feature of the current live streaming frame by combining the first violation score of the current live streaming frame under the high-accuracy moderation model, the second violation score of the current live streaming frame under the high-recall moderation model, and the multi-dimensional behavior feature of the current live streaming frame in the to-be-moderated live streaming room; and acquiring the target violation score of the current live streaming frame by inputting the live streaming behavior feature to the behavior moderation model. ### Claim 3 (depends on 1) The method according to claim 1, wherein the behavior moderation model is constructed by: extracting a corresponding history live streaming frame from each history live streaming video in a history live streaming video set, and determining a multi-dimensional history behavior feature of each history live streaming frame in a live streaming room where the each history live streaming frame is located and a violation label of the history live streaming frame; acquiring a history violation score of the history live streaming frame by inputting a first history violation score of the history live streaming frame under the high-accuracy moderation model, a second history violation score of the history live streaming frame under the high-recall moderation model, and a multi-dimensional history behavior feature of the history live streaming frame in the live streaming room to the pre-constructed behavior moderation model for each history live streaming frame; and continuously optimizing the behavior moderation model to minimize a difference between the acquired history violation score and the violation label of each history live streaming frame. ### Claim 4 (depends on 1) The method according to claim 1, wherein upon acquiring the target violation score of the current live streaming frame by inputting the first violation score of the current live streaming frame under the high-accuracy moderation model, the second violation score of the current live streaming frame under the high-recall moderation model, and the multi-dimensional behavior feature of the current live streaming frame in the to-be-moderated live streaming room to the pre-constructed behavior moderation model, the method further comprises: pushing live streaming information of the to-be-moderated live streaming room to a manual moderation platform in response to a to-be-moderated violation score exceeding a predetermined behavior violation threshold; and collecting a new current live streaming frame from the to-be-moderated live streaming room upon an interval of a predetermined live streaming duration in response to the target violation score failing to exceed the predetermined behavior violation threshold, and continuously performing violation moderation on the new current live streaming frame. ### Claim 5 (depends on 4) The method according to claim 4, further comprising: monitoring live streaming interaction items in the to-be-moderated live streaming room in real time, pushing live streaming information of the to-be-moderated live streaming room to the manual moderation platform in response to the live streaming interaction items exceeding a predetermined interaction threshold, and stopping the violation moderation performed on the current live streaming frame in the to-be-moderated live streaming room. ### Claim 6 (depends on 1) The method according to claim 1, further comprising: pushing the live streaming information of the to-be-moderated live streaming room to a manual moderation platform in a case that the first violation score exceeds the predetermined accuracy violation threshold or the second violation score exceeds the predetermined recall violation threshold. ### Claim 7 (depends on 1) The method according to claim 1, wherein training samples of the high-accuracy moderation model comprise easy positive samples and easy negative samples in a training sample set, and training samples of the high-recall moderation model comprise hard positive samples and hard negative samples in the training sample set. ### Claim 8 (depends on 1) A server, comprising: one or more processors; and a memory, configured to store one or more programs; wherein the one or more processors, when loading and running the one or more programs, are caused to perform the method for moderating live streaming as defined in claim 1. ### Claim 9 (depends on 1) A non-transitory computer-readable storage medium, storing a computer program, wherein the computer program, when loaded and run by a processor, causes the processor to perform the method for moderating live streaming as defined in claim 1. ### Claim 10 (depends on 8) The server according to claim 8, wherein the one or more processors, when loading and running the one or more programs, are caused to: acquire a live streaming behavior feature of the current live streaming frame by combining the first violation score of the current live streaming frame under the high-accuracy moderation model, the second violation score of the current live streaming frame under the high-recall moderation model, and the multi-dimensional behavior feature of the current live streaming frame in the to-be-moderated live streaming room; and acquire the target violation score of the current live streaming frame by inputting the live streaming behavior feature to the behavior moderation model. ### Claim 11 (depends on 8) The server according to claim 8, wherein the behavior moderation model is constructed by: extracting a corresponding history live streaming frame from each history live streaming video in a history live streaming video set, and determining a multi-dimensional history behavior feature of each history live streaming frame in a live streaming room where the each history live streaming frame is located and a violation label of the history live streaming frame; acquiring a history violation score of the history live streaming frame by inputting a first history violation score of the history live streaming frame under the high-accuracy moderation model, a second history violation score of the history live streaming frame under the high-recall moderation model, and a multi-dimensional history behavior feature of the history live streaming frame in the live streaming room to the pre-constructed behavior moderation model for each history live streaming frame; and continuously optimizing the behavior moderation model to minimize a difference between the history violation score and the violation label of each history live streaming frame. ### Claim 12 (depends on 8) The server according to claim 8, wherein the one or more processors, when loading and running the one or more programs, are caused to: push live streaming information of the to-be-moderated live streaming room to a manual moderation platform in response to a to-be-moderated violation score exceeding a predetermined behavior violation threshold; and collect a new current live streaming frame from the to-be-moderated live streaming room upon an interval of a predetermined live streaming duration in response to the target violation score failing to exceed the predetermined behavior violation threshold, and continuously perform violation moderation on the new current live streaming frame. ### Claim 13 (depends on 12) The server according to claim 12, the one or more processors, when loading and running the one or more programs, are caused to: monitor live streaming interaction items in the to-be-moderated live streaming room in real time, pushing live streaming information of the to-be-moderated live streaming room to the manual moderation platform in response to the live streaming interaction items exceeding a predetermined interaction threshold, and stop the violation moderation performed on the current live streaming frame in the to-be-moderated live streaming room. ### Claim 14 (depends on 8) The server according to claim 8, the one or more processors, when loading and running the one or more programs, are caused to: push live streaming information of the to-be-moderated live streaming room to a manual moderation platform in a case that the first violation score exceeds the predetermined accuracy violation threshold or the second violation score exceeds the predetermined recall violation threshold. ### Claim 15 (depends on 8) The server according to claim 8, wherein training samples of the high-accuracy moderation model comprise easy positive samples and easy negative samples in a training sample set, and training samples of the high-recall moderation model comprise hard positive samples and hard negative samples in the training sample set.