CN114727119A - Live broadcast and microphone connection control method and device and storage medium - Google Patents

Live broadcast and microphone connection control method and device and storage medium Download PDF

Info

Publication number
CN114727119A
CN114727119A CN202011521771.7A CN202011521771A CN114727119A CN 114727119 A CN114727119 A CN 114727119A CN 202011521771 A CN202011521771 A CN 202011521771A CN 114727119 A CN114727119 A CN 114727119A
Authority
CN
China
Prior art keywords
live broadcast
account
sample
data
live
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Granted
Application number
CN202011521771.7A
Other languages
Chinese (zh)
Other versions
CN114727119B (en
Inventor
申世伟
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Beijing Dajia Internet Information Technology Co Ltd
Original Assignee
Beijing Dajia Internet Information Technology Co Ltd
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Beijing Dajia Internet Information Technology Co Ltd filed Critical Beijing Dajia Internet Information Technology Co Ltd
Priority to CN202011521771.7A priority Critical patent/CN114727119B/en
Publication of CN114727119A publication Critical patent/CN114727119A/en
Application granted granted Critical
Publication of CN114727119B publication Critical patent/CN114727119B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04N—PICTORIAL COMMUNICATION, e.g. TELEVISION
    • H04N21/00—Selective content distribution, e.g. interactive television or video on demand [VOD]
    • H04N21/20—Servers specifically adapted for the distribution of content, e.g. VOD servers; Operations thereof
    • H04N21/21—Server components or server architectures
    • H04N21/218—Source of audio or video content, e.g. local disk arrays
    • H04N21/2187—Live feed
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
    • G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04N—PICTORIAL COMMUNICATION, e.g. TELEVISION
    • H04N21/00—Selective content distribution, e.g. interactive television or video on demand [VOD]
    • H04N21/20—Servers specifically adapted for the distribution of content, e.g. VOD servers; Operations thereof
    • H04N21/23—Processing of content or additional data; Elementary server operations; Server middleware
    • H04N21/239—Interfacing the upstream path of the transmission network, e.g. prioritizing client content requests
    • H04N21/2393—Interfacing the upstream path of the transmission network, e.g. prioritizing client content requests involving handling client requests
    • H04N21/2396—Interfacing the upstream path of the transmission network, e.g. prioritizing client content requests involving handling client requests characterized by admission policies
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04N—PICTORIAL COMMUNICATION, e.g. TELEVISION
    • H04N21/00—Selective content distribution, e.g. interactive television or video on demand [VOD]
    • H04N21/20—Servers specifically adapted for the distribution of content, e.g. VOD servers; Operations thereof
    • H04N21/25—Management operations performed by the server for facilitating the content distribution or administrating data related to end-users or client devices, e.g. end-user or client device authentication, learning user preferences for recommending movies
    • H04N21/254—Management at additional data server, e.g. shopping server, rights management server
    • H04N21/2541—Rights Management
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04N—PICTORIAL COMMUNICATION, e.g. TELEVISION
    • H04N21/00—Selective content distribution, e.g. interactive television or video on demand [VOD]
    • H04N21/40—Client devices specifically adapted for the reception of or interaction with content, e.g. set-top-box [STB]; Operations thereof
    • H04N21/45—Management operations performed by the client for facilitating the reception of or the interaction with the content or administrating data related to the end-user or to the client device itself, e.g. learning user preferences for recommending movies, resolving scheduling conflicts
    • H04N21/462—Content or additional data management, e.g. creating a master electronic program guide from data received from the Internet and a Head-end, controlling the complexity of a video stream by scaling the resolution or bit-rate based on the client capabilities
    • H04N21/4627—Rights management associated to the content

Landscapes

  • Engineering & Computer Science (AREA)
  • Databases & Information Systems (AREA)
  • Multimedia (AREA)
  • Signal Processing (AREA)
  • Computer Security & Cryptography (AREA)
  • Data Mining & Analysis (AREA)
  • Theoretical Computer Science (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Artificial Intelligence (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Evolutionary Biology (AREA)
  • Evolutionary Computation (AREA)
  • Physics & Mathematics (AREA)
  • General Engineering & Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Two-Way Televisions, Distribution Of Moving Picture Or The Like (AREA)

Abstract

The disclosure relates to a live broadcast wheat connection control method, a live broadcast wheat connection control device and a storage medium, and belongs to the technical field of live broadcast. The method comprises the following steps: receiving a live broadcasting room connecting request initiated by a target account, and acquiring live broadcasting data corresponding to the target account; determining candidate accounts from other accounts that are initiating live-air microphone connecting requests; acquiring live broadcast data corresponding to the candidate account; inputting live broadcast data corresponding to the target account and live broadcast data corresponding to the candidate account into a prediction model, and performing abnormal prediction on wheat connection between the target account and the candidate account to obtain abnormal prediction data of the wheat connection between the target account and the candidate account; and controlling the connection between the target account and the candidate account based on the abnormal prediction data. According to the method, the direct broadcasting data are analyzed to obtain accurate prediction data, so that a suitable wheat connecting object is matched for a wheat connecting requester based on the prediction data, wheat connecting with a high probability of violation is identified and avoided, and the incidence rate of bad wheat connecting is reduced.

Description

Live broadcast and microphone connection control method and device and storage medium
Technical Field
The present disclosure relates to the field of live broadcast technologies, and in particular, to a live broadcast and live broadcast wheat connection control method, apparatus, and storage medium.
Background
With the continuous upgrade of the mobile internet technology, the online live broadcast industry develops rapidly, and the influence of user groups, anchor groups and the industry is continuously expanded. At present, the habits of users in the online live broadcast industry are basically developed, each large live broadcast platform focuses on differentiated contents, high-quality special columns are created, and the platforms are searched for original self-made contents.
In the related live broadcasting technology, a main broadcasting PK (Player king) activity often occurs, and the main broadcasting end and the audience end can simultaneously play two paths of live broadcasting streams. The anchor PK can enrich the interest of the live broadcast content and is popular with users. However, the parties involved in the PK are matched randomly by the system, and some non-documentary PK phenomena may occur when the PK is hosted, such as perusal of both parties, which may adversely affect the viewer of the live broadcast.
Disclosure of Invention
The disclosure provides a live broadcast and microphone connection control method, a live broadcast and microphone connection control device and a storage medium. Whether the link between the target account and other accounts is established or not is determined by analyzing the live broadcast data corresponding to the target account and the candidate account, and the technical problem that link pairing is not appropriate due to random distribution of link objects in the related technology can be solved. The technical scheme of the disclosure is as follows:
according to a first aspect of the embodiments of the present disclosure, a live broadcast and live broadcast control method is provided, including:
receiving a live broadcast room microphone connecting request initiated by a target account, and acquiring live broadcast data corresponding to the target account;
determining candidate accounts from other accounts that are initiating live-air microphone connecting requests;
acquiring live broadcast data corresponding to the candidate account;
inputting live broadcast data corresponding to the target account and live broadcast data corresponding to the candidate account into a prediction model, and performing abnormal prediction on wheat connection between the target account and the candidate account to obtain abnormal prediction data of the wheat connection between the target account and the candidate account;
and controlling the connection between the target account and the candidate account based on the abnormal prediction data.
In an exemplary embodiment, the controlling the liaison between the target account and the candidate account based on the anomaly prediction data includes:
under the condition that the abnormal prediction data of the wheat connection between the target account and the candidate account is smaller than a preset threshold value, taking the candidate account as a wheat connection object of the target account;
and establishing the wheat connecting between the target account and the wheat connecting object.
In an exemplary embodiment, the method further comprises:
and under the condition that the abnormal prediction data of the link between the target account and the candidate account does not meet the preset condition, removing the client where the candidate account corresponding to the abnormal prediction data is located from other clients, and returning to execute the step of determining the candidate client from other clients initiating the live broadcast room link request until the abnormal prediction data of the link between the target account and the candidate account meets the preset condition, wherein the candidate account corresponding to the abnormal prediction data meeting the preset condition is used as the link object of the target account.
In an exemplary embodiment, the inputting live broadcast data corresponding to the target account and live broadcast data corresponding to the candidate account into a prediction model, performing abnormal prediction on wheat connecting between the target account and the candidate account, and obtaining abnormal prediction data of wheat connecting between the target account and the candidate account includes:
inputting live broadcast data corresponding to the target account and live broadcast data corresponding to the candidate account into a prediction model, respectively extracting features of the live broadcast data corresponding to the target account and the live broadcast data corresponding to the candidate account, merging the extracted live broadcast features of the target account and the live broadcast features of the candidate account, and performing probability prediction according to the merged feature vectors to obtain abnormal prediction data of the connection between the target account and the candidate account.
In an exemplary embodiment, the obtaining live data corresponding to the target account includes:
extracting a preset number of live broadcast images from the live broadcast segments corresponding to the target account to serve as first live broadcast images; performing feature extraction on the first live image to obtain a corresponding first image feature vector; carrying out face recognition on the first live image to obtain corresponding first image character information; carrying out anomaly detection on the first live image to obtain corresponding first image anomaly information; generating live broadcast image information of the target account according to the first image feature vector, the first image character information and the first image abnormal information;
acquiring historical abnormal information and account identity information of the target account, and determining the account information of the target account according to the historical abnormal information and the account identity information of the target account;
acquiring the viewer number and the live broadcast time period of the live broadcast room corresponding to the target account at the current moment, and determining the live broadcast room information of the target account according to the viewer number and the live broadcast time period of the live broadcast room corresponding to the target account at the current moment;
and determining the live broadcast data of the target account according to the live broadcast image information of the target account, the account information of the target account and the live broadcast room information of the target account.
In an exemplary embodiment, the obtaining live data corresponding to the candidate account includes:
extracting a preset number of live broadcast images from the live broadcast segments corresponding to the candidate accounts to serve as second live broadcast images; performing feature extraction on the second live broadcast image to obtain a corresponding second image feature vector, performing face recognition on the second live broadcast image to obtain corresponding second image character information, and performing anomaly detection on the second live broadcast image to obtain corresponding second image anomaly information; generating live image information of the candidate account according to the second image feature vector, second image character information and second image abnormal information;
acquiring historical abnormal information and account identity information of the candidate account, and determining the account information of the candidate account according to the historical abnormal information and the account identity information of the candidate account;
acquiring the viewer number and the live broadcast time period of the live broadcast room corresponding to the candidate account at the current moment, and determining the live broadcast room information of the candidate account according to the viewer number and the live broadcast time period of the live broadcast room corresponding to the candidate account at the current moment;
and determining live broadcast data of the candidate account according to the live broadcast image information of the candidate account, the account information of the candidate account and the live broadcast room information of the candidate account.
In an exemplary embodiment, the predictive model is determined by:
acquiring a training sample set, wherein the training sample set comprises at least one sample data with a label attribute of a positive sample and at least one sample data with a label attribute of a negative sample;
inputting each sample data in the training sample set into an initial deep learning model to obtain sample abnormality prediction data of each sample data in the training sample set;
determining the prediction attribute of each sample data according to the comparison result between the sample abnormity prediction data of each sample data and a preset threshold value;
and training the initial deep learning model based on the loss between the prediction attribute of each sample data and the corresponding label attribute to obtain the prediction model.
In an exemplary embodiment, the obtaining the set of training samples includes:
obtaining historical live broadcast data, wherein the historical live broadcast data comprises live broadcast segments without abnormal behaviors and live broadcast segments with abnormal behaviors;
acquiring sample live broadcast data corresponding to each live broadcast segment;
taking sample live broadcast data corresponding to the live broadcast segment with abnormal behavior as sample data with a positive label attribute; and taking sample live broadcast data corresponding to the live broadcast segment without abnormal behaviors as sample data with the label attribute of a negative sample.
In an exemplary embodiment, the obtaining sample live broadcast data corresponding to each live broadcast segment includes:
extracting a preset number of live broadcast images from the live broadcast connected with the TV as sample live broadcast images; performing feature extraction on the sample live image to obtain a corresponding sample image feature vector, performing face recognition on the sample live image to obtain sample image character information, and performing anomaly detection on the sample live image to obtain sample image anomaly prediction data; generating sample live broadcast image information according to the sample image feature vector, the sample image character information and the sample image abnormal prediction data;
determining a sample account corresponding to the live microphone connecting segment, acquiring historical abnormal information and account identity information of the sample account, and determining sample account information according to the historical abnormal information and the account identity information of the sample account;
acquiring audience number and live broadcast time period of a live broadcast room corresponding to the live broadcast segment, and determining sample live broadcast room information according to the audience number and the live broadcast time period of the live broadcast room corresponding to the live broadcast segment;
and determining the sample live broadcast data according to the sample live broadcast image information, the sample account information and the sample live broadcast room information.
According to a second aspect of the embodiments of the present disclosure, there is provided a live broadcast microphone connecting control apparatus, including:
the system comprises a first live broadcast data acquisition unit, a second live broadcast data acquisition unit and a third live broadcast data acquisition unit, wherein the first live broadcast data acquisition unit is configured to receive a live broadcast room microphone connecting request initiated by a target account and acquire live broadcast data corresponding to the target account;
the account candidate determining unit is configured to determine a candidate account from other accounts which initiate live broadcast room microphone connecting requests;
the second live broadcast data acquisition unit is configured to acquire live broadcast data corresponding to the candidate account;
an abnormal prediction data acquisition unit, configured to input live data corresponding to the target account and live data corresponding to the candidate account into a prediction model, perform abnormal prediction on wheat connecting between the target account and the candidate account, and obtain abnormal prediction data of wheat connecting between the target account and the candidate account;
a microphone connecting control unit configured to control microphone connecting between the target account and the candidate account based on the abnormality prediction data.
In an exemplary embodiment, the microphone connecting control unit includes:
the first wheat connecting control module is configured to execute the candidate account as a wheat connecting object of the target account and establish the wheat connecting between the target account and the wheat connecting object under the condition that the abnormal prediction data of the wheat connecting between the target account and the candidate account is smaller than a preset threshold value.
In an exemplary embodiment, the microphone connecting control unit further includes:
and the second wheat connecting control module is configured to remove the client where the candidate account corresponding to the abnormal prediction data is located from other clients under the condition that the abnormal prediction data of the wheat connecting between the target account and the candidate account meets a preset condition, and return to execute the step of determining the candidate client from other clients initiating the live broadcast room wheat connecting request, and when the abnormal prediction data of the wheat connecting between the target account and the candidate account meets the preset condition, the candidate account corresponding to the abnormal prediction data meeting the preset condition is used as a wheat connecting object of the target account.
In an exemplary embodiment, the abnormal prediction data obtaining unit is further configured to input live broadcast data corresponding to the target account and live broadcast data corresponding to the candidate account into a prediction model, perform feature extraction on the live broadcast data corresponding to the target account and the live broadcast data corresponding to the candidate account respectively, merge the extracted live broadcast features of the target account and the live broadcast features of the candidate account, perform probability prediction according to a feature vector obtained by the merging, and obtain abnormal prediction data of connections between the target account and the candidate account.
In an exemplary embodiment, the first broadcast data acquiring unit includes:
the first live image information acquisition module is configured to extract a preset number of live images from the live segments corresponding to the target account to serve as first live images; performing feature extraction on the first live image to obtain a corresponding first image feature vector; carrying out face recognition on the first live image to obtain corresponding first image character information; carrying out anomaly detection on the first live image to obtain corresponding first image anomaly information; generating live broadcast image information of the target account according to the first image feature vector, the first image character information and the first image abnormal information;
the first account information acquisition module is configured to acquire historical abnormal information and account identity information of the target account, and determine the account information of the target account according to the historical abnormal information and the account identity information of the target account;
a first live broadcasting room information acquisition module, configured to acquire a viewer and a live broadcasting time period of a live broadcasting room at a current moment corresponding to the target account, and determine live broadcasting room information of the target account according to the viewer and the live broadcasting time period of the live broadcasting room at the current moment corresponding to the target account;
the first live broadcast data determining module is configured to determine live broadcast data of the target account according to the live broadcast image information of the target account, the account information of the target account and the live broadcast room information of the target account.
In an exemplary embodiment, the second live data acquisition unit includes:
the second live broadcast image information acquisition module is configured to extract a preset number of live broadcast images from the live broadcast segments corresponding to the candidate accounts as second live broadcast images; performing feature extraction on the second live broadcast image to obtain a corresponding second image feature vector, performing face recognition on the second live broadcast image to obtain corresponding second image character information, and performing anomaly detection on the second live broadcast image to obtain corresponding second image anomaly information; generating live image information of the candidate account according to the second image feature vector, second image character information and second image abnormal information;
the second account information acquisition module is configured to acquire historical abnormal information and account identity information of the candidate account and determine the account information of the candidate account according to the historical abnormal information and the account identity information of the candidate account;
a second live broadcast room information acquisition module configured to acquire the viewer number and the live broadcast time period of the live broadcast room corresponding to the candidate account at the current time, and determine live broadcast room information of the candidate account according to the viewer number and the live broadcast time period of the live broadcast room corresponding to the candidate account at the current time;
the second live broadcast data determining module is configured to determine live broadcast data of the candidate account according to the live broadcast image information of the candidate account, the account information of the candidate account and the live broadcast room information of the candidate account.
In an exemplary embodiment, the live microphone connecting control apparatus further includes:
the system comprises a sample data acquisition unit, a training sample set and a data processing unit, wherein the sample data acquisition unit is configured to acquire the training sample set, and the training sample set comprises at least one sample data with a label attribute of a positive sample and at least one sample data with a label attribute of a negative sample;
the model training unit is configured to input each sample data in the training sample set into an initial deep learning model to obtain sample abnormality prediction data of each sample data in the training sample set, determine a prediction attribute of each sample data according to a comparison result between the sample abnormality prediction data of each sample data and a preset threshold, and train the initial deep learning model based on a loss between the prediction attribute of each sample data and a corresponding label attribute to obtain the prediction model.
In an exemplary embodiment, the sample data acquiring unit includes:
the historical data acquisition module is configured to acquire historical live broadcast data with live broadcast connected with wheat, and the historical live broadcast data with live broadcast connected with wheat does not have abnormal behaviors and comprises live broadcast connected with wheat segments with abnormal behaviors;
the live sample data acquisition module is configured to acquire live sample data corresponding to each live microphone segment; taking sample live broadcast data corresponding to the live broadcast segment with abnormal behavior as sample data with a positive label attribute; and taking sample live broadcast data corresponding to the live broadcast segment without abnormal behaviors as sample data with the label attribute of a negative sample.
In an exemplary embodiment, the sample live data acquisition module includes:
the live broadcast image information acquisition sub-module is configured to extract live broadcast images of a preset number from the live broadcast segment as sample live broadcast images; performing feature extraction on the sample live image to obtain a corresponding sample image feature vector, performing face recognition on the sample live image to obtain sample image character information, and performing anomaly detection on the sample live image to obtain sample image anomaly prediction data; generating sample live broadcast image information according to the sample image feature vector, the sample image character information and the sample image abnormal prediction data;
the sample account information acquisition sub-module is configured to determine a sample account corresponding to the live telecast link segment, acquire historical abnormal information and account identity information of the sample account, and determine sample account information according to the historical abnormal information and the account identity information of the sample account;
the sample live broadcast room information acquisition sub-module is configured to acquire the audience number and the live broadcast time period of a live broadcast room corresponding to the live broadcast connected segment, and determine sample live broadcast room information according to the audience number and the live broadcast time period of the live broadcast room corresponding to the live broadcast connected segment;
and the sample live broadcast data determining sub-module is configured to determine the sample live broadcast data according to the sample live broadcast image information, the sample account information and the sample live broadcast room information.
According to a third aspect of the embodiments of the present disclosure, there is provided an electronic apparatus, including:
a processor;
a memory for storing the processor-executable instructions;
wherein the processor is configured to execute the instructions to implement the live microphone connecting control method according to the first aspect.
According to a fourth aspect of embodiments of the present disclosure, there is provided a storage medium, where instructions, when executed by a processor of an electronic device, enable the electronic device to perform the live microphone connecting control method according to the first aspect.
The technical scheme provided by the embodiment of the disclosure at least brings the following beneficial effects:
according to the live broadcast microphone connecting control method, the device and the storage medium provided by the embodiment of the disclosure, live broadcast data corresponding to a target account is acquired by receiving a live broadcast microphone connecting request initiated by a client where the target account is located; determining candidate clients from other clients initiating live broadcast room microphone connecting requests, and acquiring live broadcast data corresponding to candidate accounts of the login candidate clients; inputting live broadcast data corresponding to the target account and live broadcast data corresponding to the candidate account into a prediction model, and performing abnormal prediction on the link between the target account and the candidate account to obtain abnormal prediction data corresponding to the link between the target account and the candidate account; and controlling the connection between the target account and the candidate account based on the abnormal prediction data. According to the method, in the process of determining the microphone connecting object for the target account, the probability of abnormality after microphone connecting between the target account and the candidate account is predicted by analyzing the live broadcast data of the target account and the live broadcast data of the candidate account, abnormality prediction is performed based on the live broadcast data capable of indicating user behaviors, and accurate abnormality prediction data can be obtained; the larger the abnormal prediction data is, the higher the probability of abnormal contents such as violation and the like in live broadcast of the target account and the candidate account, so that the connection between the two accounts with lower abnormal probability after the connection can be allowed, a processing mode without establishing the connection can be given to the two accounts with higher abnormal content probability after the connection, and compared with a mode of randomly distributing a connection object for a connection requester, the method determines whether to establish the connection between the accounts by analyzing the live broadcast data corresponding to the accounts, can match the connection requester with a proper connection object, identifies and avoids the connection with violation at high probability, reduces the incidence rate of bad connection, and creates a comfortable and healthy live broadcast and watching environment for a user.
In the embodiment of the disclosure, a training sample set is obtained, wherein the training sample set comprises at least one sample data with a label attribute of a positive sample and at least one sample data with a label attribute of a negative sample; inputting all sample data in the training sample set into an initial deep learning model to obtain sample abnormality prediction data of all sample data in the training sample set; determining the prediction attribute of each sample data according to the comparison result between the sample abnormity prediction data of each sample data and a preset threshold value; and training the initial deep learning model based on the loss between the prediction attribute of each sample data and the corresponding label attribute to obtain the prediction model. The initial deep learning model is trained by using the positive and negative sample data, so that the initial deep learning model can automatically learn the characteristics of the positive sample data and the characteristics of the negative sample data, and the accuracy of predicting the abnormal prediction data corresponding to the input data is improved.
It is to be understood that both the foregoing general description and the following detailed description are exemplary and explanatory only and are not restrictive of the disclosure.
Drawings
The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the present disclosure and, together with the description, serve to explain the principles of the disclosure and are not to be construed as limiting the disclosure.
Fig. 1 is a schematic diagram of an implementation environment of a live microphone connection control method according to an exemplary embodiment.
Fig. 2 is a schematic interface diagram illustrating a live connection to a microphone according to an exemplary embodiment.
Fig. 3 is a flowchart illustrating a live microphone connection control method according to an exemplary embodiment.
FIG. 4 is a flow diagram illustrating a method of training a predictive model in accordance with an exemplary embodiment.
FIG. 5 is a schematic diagram illustrating an XGB model in accordance with an exemplary embodiment.
FIG. 6 is an application diagram illustrating an XGB model in accordance with an exemplary embodiment.
Fig. 7 is a flowchart illustrating a live telecast control method based on a predictive model according to an exemplary embodiment.
Fig. 8 is a schematic structural diagram illustrating a live microphone connecting control apparatus according to an exemplary embodiment.
Fig. 9 is a schematic structural diagram of another live microphone connecting control device according to an exemplary embodiment.
Fig. 10 is a block diagram illustrating a terminal according to an example embodiment.
FIG. 11 is a block diagram illustrating a server in accordance with an example embodiment.
Detailed Description
In order to make the technical solutions of the present disclosure better understood by those of ordinary skill in the art, the technical solutions in the embodiments of the present disclosure will be clearly and completely described below with reference to the accompanying drawings.
It should be noted that the terms "first," "second," and the like in the description and claims of the present disclosure and in the above-described drawings are used for distinguishing between similar elements and not necessarily for describing a particular sequential or chronological order. It is to be understood that the data so used is interchangeable under appropriate circumstances such that the embodiments of the disclosure described herein are capable of operation in sequences other than those illustrated or otherwise described herein. The implementations described in the exemplary embodiments below are not intended to represent all implementations consistent with the present disclosure. Rather, they are merely examples of apparatus and methods consistent with certain aspects of the present disclosure, as detailed in the appended claims.
At present, a network live broadcast platform becomes a brand-new social media, videos recorded by a main broadcast can be synchronized to client sides of audiences in real time, and the audiences can make comments at the client sides and interact with the main broadcast in a manner of sending gifts to the main broadcast. In order to improve the interaction degree between the main broadcast and the audience, a live broadcast mode with the live broadcast is also provided at present. The microphones of two anchor may be connected and the two recorded video contents of the two anchors may be placed in the same live room.
Fig. 1 is a schematic diagram of an implementation environment of a live microphone connection control method according to an exemplary embodiment. Please refer to fig. 1, which includes: the system comprises a live broadcast server 102, a live broadcast microphone connecting control device 103 and a client 101, wherein the client 101 is in communication connection with the live broadcast server 102, and the live broadcast microphone connecting control device 103 is in communication with the live broadcast server 102. Wherein, the client 101 includes at least one viewer client (e.g. viewer client 1013 shown in the figure) and at least two anchor clients (e.g. anchor 1 client 1011 and anchor 2 client 1012 shown in the figure) participating in the live PK, and the live microphone connecting control device 103 may exist as a separate server or be built in the live server 102.
And the anchor client participating in the live PK is configured to obtain live data and send the live data to the live server 102, where the live data includes video data.
In a specific implementation, the client 101 may include: the physical devices may also include software running in the physical device, such as an application having a video key frame extraction function, and the like, which is not limited in the embodiment of the present disclosure. The Client 101 may be communicatively coupled to the live Server 102 based on a Browser/Server mode (Browser/Server, B/S) or a Client/Server mode (Client/Server, C/S).
The live broadcast server 102 is configured to respond to a live broadcast room microphone connecting request initiated by a client where the target account is located, and transmit request data to the live broadcast microphone connecting control device 103, where the live broadcast microphone connecting control device 103 is configured to receive the live broadcast room microphone connecting request initiated by the client where the target account is located, and obtain live broadcast data corresponding to the target account; determining candidate clients from other clients initiating live broadcast room microphone connecting requests; acquiring live broadcast data corresponding to a candidate account of a login candidate client; inputting live broadcast data corresponding to the target account and live broadcast data corresponding to the candidate account into a prediction model, and performing abnormal prediction on wheat connection between the target account and the candidate account to obtain abnormal prediction data of the wheat connection between the target account and the candidate account; and controlling the connection between the target account and the candidate account based on the abnormal prediction data.
The client where the target account and the candidate account are located may be an anchor client, and under the condition that a call connection is allowed between the anchor clients, the live broadcast server 102 integrates live broadcast data from the anchor client, and sends the integrated data to a viewer client viewing the live broadcast PK and the anchor client participating in the live broadcast PK. The client displays the live PK image, and as shown in fig. 2, the live PK image displays account information of two anchor clients participating in live PK, video data of the two anchor clients, speech data of each spectator client, and game data such as game images and game scores when participating in the live PK anchor for game competition.
The live server 102 may comprise a server operating independently, or a distributed server, or a server cluster consisting of multiple servers.
It should be noted that the number of the anchor client and the number of the audience clients may be changed according to actual requirements, the anchor client may also become an audience client when watching the live broadcast of other anchor, and the audience client may also become an anchor client when participating in the live broadcast or live broadcast, which is not limited in this disclosure.
Fig. 3 is a flowchart illustrating a live microphone connection control method according to an exemplary embodiment. As shown in fig. 3, the live broadcast microphone connecting control method disclosed by the present disclosure matches live broadcast data corresponding to users who need to perform microphone connecting PK, controls microphone connecting among the users according to a matching result, and adopts a prediction model to perform data matching, so that prediction efficiency can be improved, and accuracy of the prediction result can be improved. The live broadcast wheat connecting control method provided by the embodiment of the disclosure comprises two parts, namely a model training part and a model application part. The model training part is mainly used for training an initial deep learning model based on training samples (sample data with positive label attributes and sample data with negative label attributes) to obtain a prediction model; the model application part mainly inputs live broadcast data corresponding to the target account and live broadcast data corresponding to the candidate account into the prediction model, outputs abnormal prediction data of wheat connection between the target account and the candidate account, and then controls the wheat connection between the target account and the candidate account based on the abnormal prediction data.
FIG. 4 is a flow diagram illustrating a method of training a predictive model in accordance with an exemplary embodiment. As shown in fig. 4, the training process of the prediction model includes the following steps.
S401, a training sample set is obtained, wherein the training sample set comprises at least one sample data with a label attribute of a positive sample and at least one sample data with a label attribute of a negative sample.
In one possible implementation, the training sample set may be obtained by:
s4011, historical live broadcast data are obtained, and the historical live broadcast data comprise live broadcast segments without abnormal behaviors and live broadcast segments with abnormal behaviors.
In specific implementation, all live broadcast segments which are violated in the past can be found from live broadcast PK data stored in a live broadcast platform through modes such as image detection, voice detection and the like, meanwhile, an equal number of live broadcast segments which are violated are extracted from the number of live broadcast PKs which are not violated, and user IDs (namely accounts) of users participating in the PKs in live broadcast are recorded. Furthermore, manual reinspection can be carried out on the found live broadcast fragments with wheat so as to ensure the accuracy of the data.
Illustratively, when a picture and voice of both cursors appear in a certain time period in a complete live PK data, live broadcast content corresponding to the time period is taken as a live broadcast segment with abnormal behaviors.
And S4013, obtaining sample live broadcast data corresponding to each live broadcast segment.
In one possible implementation manner, sample live broadcast data corresponding to the live broadcast segment can be acquired by the following method. The method comprises the following steps:
(1) extracting a preset number of live broadcast images from live broadcast segments of continuous wheat as sample live broadcast images; performing feature extraction on a sample live image to obtain a corresponding sample image feature vector, performing face recognition on the sample live image to obtain sample image character information, and performing anomaly detection on the sample live image to obtain sample image anomaly prediction data; and generating sample live image information according to the sample image feature vector, the sample image character information and the sample image abnormal prediction data.
And randomly sampling X screenshots as sample live broadcast images aiming at the live broadcast fragments of the positive and negative samples.
The sample live broadcast Image is subjected to feature extraction by using a feature extraction model, for example, the sample live broadcast Image is input into a deep learning classification model (a model which can be obtained on a public network) such as resnet50 and inclusion v3 which is pre-trained by using an Image Net data set, output features (generally, feature vectors of 1024 dimensions or 2048 dimensions) of a fully-connected layer which is a second-to-last layer of the deep learning classification model are obtained, and the output features are used as sample Image feature vectors corresponding to the sample live broadcast Image. Wherein the Image Net dataset is a large-scale labeled Image dataset organized according to the Word Net architecture. Resnet50 is a typical Network of Residual networks (Residual networks) widely used in the field of object classification and the like and as part of the classical neural Network of the computer vision task backbone. Resnet50 contains 50 conv2d operations, and in the data processing process, the convolution operation is firstly carried out on the input, then 4 Residual blocks (Residual blocks) are contained, and finally the full connection operation is carried out so as to carry out the classification task. The inclusion v3 network is a very deep convolutional network developed by Google, and is a common network for picture recognition, and is not described herein.
When the live image of the sample is subjected to face recognition, public face recognition service can be called, such as a public face recognition model, the face recognition results of the X screenshots are obtained, the face recognition results comprise the features of the number of faces of each screenshot, the average age of the faces, the male proportion and the like which indicate the identity of a person, and the character information of the sample image is determined based on the face recognition results.
The sample image abnormality prediction data may be obtained by detecting, using a detection model, for example, a sample live image input to the detection model having an abnormal behavior recognition capability such as pornography, smoking, naked upper body, or the like, and obtaining sample image abnormality prediction data output from the detection model.
(2) Determining a sample account corresponding to the live microphone connecting segment, acquiring historical abnormal information and account identity information of the sample account, and determining the sample account information according to the historical abnormal information and the account identity information of the sample account.
Determining the anchor of the PK participating in the live broadcast room aiming at the live broadcast fragments of the live broadcast of the positive and negative samples, taking the anchor of the PK participating as a sample user, and acquiring respective historical abnormal information and account identity information of each sample user participating in the PK according to the user ID of the sample user, wherein the historical abnormal information comprises the broadcasting times, violation times and violation proportion of the previous [ a1, a2 and a3] days (such as the previous 7 days, the previous 30 days and the previous 90 days), and the account identity information comprises the user characteristics of the age, the gender, the city and the like of the anchor. The historical abnormal information can be obtained by inquiring in the violation database according to the user ID, and the account identity information can be obtained by inquiring in the user database according to the user ID.
(3) And acquiring the audience number and the live broadcast time period of the live broadcast room corresponding to the live broadcast segment, and determining sample live broadcast room information according to the audience number and the live broadcast time period of the live broadcast room corresponding to the live broadcast segment.
For each live broadcast segment, determining the PK live broadcast time corresponding to the live broadcast segment and the main broadcast of the PK participating in the live broadcast room, acquiring the view mode of the main broadcast live broadcast room in the PK live broadcast time, and determining the live broadcast time period corresponding to the PK live broadcast time, wherein each hour can be set as a time period so as to determine the live broadcast time period corresponding to the PK live broadcast time, and illustratively, if the PK live broadcast time corresponding to the live broadcast segment is 22:15-22:45, the corresponding live broadcast time period is 22:00-23: 00.
(4) And determining sample live broadcast data according to the sample live broadcast image information, the sample account information and the sample live broadcast room information.
For any live broadcast segment with wheat, the sample live broadcast data can be composed of sample live broadcast image information, sample account information and sample live broadcast room information, the sample live broadcast data embodies the account information of the anchor of the PK, the historical behavior and the current behavior of the anchor in the process of live broadcast of the PK, and the like, so that the effective information in the live broadcast PK scene can be comprehensively covered, the model can learn various characteristics in the PK scene, and the prediction accuracy is improved.
S4015, taking sample live broadcast data corresponding to the live broadcast segment with abnormal behavior as sample data with a positive sample as a label attribute; and taking sample live broadcast data corresponding to the live broadcast segment without abnormal behaviors as sample data with the label attribute of a negative sample.
The method and the device for model training adopt positive sample data and negative sample data to conduct model training, the training samples comprise the sample data and corresponding label attributes, and the label attributes are used for identifying the positive and negative sample attributes of the sample data. Specifically, the number of the sample data with the positive sample attribute is equal to that of the sample data with the negative sample attribute, so that model learning samples are enriched, and the generalization capability of the model is improved.
S403, inputting all sample data in the training sample set into the initial deep learning model to obtain sample abnormality prediction data of all sample data in the training sample set; determining the prediction attribute of each sample data according to the comparison result between the sample abnormity prediction data of each sample data and a preset threshold value; and training the initial deep learning model based on the loss between the prediction attribute of each sample data and the corresponding label attribute to obtain the prediction model.
The initial deep learning model of the disclosed embodiments may include a feature extraction network and a probabilistic prediction network, where an output of the feature extraction network is an input of the probabilistic prediction network. In the process of training the initial deep learning model by using sample data, inputting a training sample into the initial deep learning model, respectively extracting characteristics of sample live broadcast image information, sample account information and sample live broadcast room information in the training sample by using a characteristic extraction network to obtain corresponding sample live broadcast image characteristics, sample account characteristics and sample live broadcast room characteristics, further combining the sample live broadcast image characteristics, sample account characteristics and sample live broadcast room characteristics into a characteristic vector, inputting the combined characteristic vector into a probability prediction network, outputting sample abnormity prediction data of the sample data corresponding to the characteristic vector, comparing the sample abnormity prediction data with a preset threshold value, determining the prediction attribute of the sample data as a positive sample when the sample abnormity prediction data is greater than or equal to the preset threshold value, and determining the sample data as a negative sample when the sample abnormity prediction data is less than the preset threshold value, determining the prediction attribute of the sample data as a negative sample, calculating the loss between the prediction attribute of the sample data and the label attribute of the sample data, adjusting the parameters of the initial deep learning model based on the loss, and stopping adjusting the parameters of the initial deep learning model until a preset training stopping condition is met to obtain the prediction model. The training stopping condition may be that the loss value reaches a preset value or the training frequency reaches a preset frequency, or the training of the model is stopped when the loss value is not significantly reduced relative to the loss value obtained last time.
In one possible implementation, a Convolutional Neural Network (CNN) may be used as the feature extraction network to obtain the merged feature vector.
In a possible implementation manner, a Gradient Boosting Decision Tree (GBDT) can be selected as a probability prediction network, the GBDT is an addition model based on a boosting enhancement strategy, greedy learning is performed by using a forward distribution algorithm during training, and a CART Tree is learned every iteration to fit a residual error between a prediction result of a previous t-1 Tree and a true value of a training sample. Xgb (extreme vector boosting) is an industrial implementation of GBDT, which reduces the loss function by fitting pseudo-residuals with increasing new trees. The XGB model generates a plurality of regression trees based on the characteristics, each regression tree learns corresponding residual errors, and the sum of the residual errors is the predicted value of the data sample.
As shown in fig. 5, the tree structure of the XGB model includes nodes N1 to N5, where N1 is a root node, N2 is an intermediate node, N3, N4, and N5 are leaf nodes, and N1 and N2 are split nodes. In using the data for model prediction, the actual predicted path is shown in FIG. 5, i.e., from the root node N1 to the intermediate node N2, and then from the intermediate node to the leaf node N5. The XGB uses first-order and second-order partial derivatives, the second-order derivative is beneficial to quicker and more accurate gradient reduction, the function is obtained by using the Taylor expansion to serve as the second-order derivative form of the independent variable, the leaf splitting optimization calculation can be carried out only depending on the value of input data under the condition that the specific form of the loss function is not selected, and the selection of the loss function is essentially separated from the optimization/parameter selection of a model algorithm. This decoupling increases the applicability of the XGB, making it possible to choose the penalty function as desired, either for classification or regression.
FIG. 6 shows an example schematic of an XGB model. The XGB model shown in fig. 2 includes two trees (decision trees), tree 1 and tree 2. In other embodiments, the XGB model may include more trees. For each decision tree, assuming there are T leaf nodes, the decision tree model can be written as: f (x) ═ Wq(x) Wherein, q: rm→ 1, 2.... T. > is the mapping from the feature input x to the leaf node number, which is essentially the branching structure of the tree. w is formed by RTIs a leaf node weight vector, the leaf node weight vector is (w)1，w2，...，wT). Assuming that the example x falls on the leaf node j of the decision tree, the predicted output value is the weight value w of the leaf node jj。
As shown in fig. 6, tree 1 has 2 split nodes and 3 leaf nodes, and tree 2 has 1 split node and 2 leaf nodes. In Tree 1, the splitting characteristic of the first split node is { age }<15} and the splitting characteristic of the second split node is { is large? }. In tree 2, the splitting characteristic of the split node is { use computer hierarchy }. The leaf node weight vector of tree 1 is (W)11，W12，W13) I.e., (+2, +0.1, -1). The leaf node weight vector of Tree 2 is (W)21，W22) I.e., (+0.9, -0.9). Assuming instance x falls on leaf node 1 of tree 1, the prediction value is +2. If the leaf node 1 of the tree 1 and the leaf node 1 of the tree 2 are fallen at the same time, the predicted value is +2+0.9 to +2.9, as shown in fig. 6.
Specifically, the XGB is used as the probability prediction network, and related parameters of the model, including the learning rate, the maximum tree depth, the parameter regularization system and the optimization objective function, may be preset.
In a possible implementation mode, the learning rate of the XGB model can be set to be 0.03, the maximum tree depth is 6, the parameter regularization coefficient is 2, and the optimization objective function is classified cross entropy loss; model training is performed using the open source XGB tool until the penalty on the test set no longer decreases, at which point the trained XGB model M is obtained. And meanwhile, a threshold value R is determined according to the accuracy and recall condition of the model M, and the live PK is easy to violate if the threshold value R is larger than the threshold value R, and PK matching is not recommended.
In the embodiment of the disclosure, a training sample set is obtained, wherein the training sample set comprises at least one sample data with a label attribute of a positive sample and at least one sample data with a label attribute of a negative sample; inputting all sample data in the training sample set into an initial deep learning model to obtain sample abnormality prediction data of all sample data in the training sample set; determining the prediction attribute of each sample data according to the comparison result between the sample abnormity prediction data of each sample data and a preset threshold value; and training the initial deep learning model based on the loss between the prediction attribute of each sample data and the corresponding label attribute to obtain the prediction model. The initial deep learning model is trained by using the positive and negative sample data, so that the initial deep learning model can automatically learn the characteristics of the positive sample data and the characteristics of the negative sample data, and the accuracy of predicting the abnormal prediction data corresponding to the input data is improved.
The above method for training the initial deep learning model to obtain the prediction model may be performed on a terminal or a server.
Fig. 7 is a flowchart illustrating a live telecast control method based on a predictive model according to an exemplary embodiment. Referring to fig. 7, the live microphone connection control method is applied to a server and includes the following steps.
And S701, receiving a live broadcast room microphone connecting request initiated by a target account, and acquiring live broadcast data corresponding to the target account.
With the development of live broadcast technology, a mode of connecting at least two anchor broadcasters to each other to make pictures and sounds live in vermicelli in both rooms has appeared. In order to improve the interest of live broadcast to a greater extent, a competitive PK mechanism is introduced to allow two anchor broadcasters to play a short game on the basis of anchor wheat connection, which is also called anchor wheat connection PK.
In a scenario of a microphone PK, after an anchor broadcast corresponding to a first client is played, a user may want to connect with another user PK, for example, an interaction request may be sent to another user, and when a user or an anchor broadcast corresponding to a second client accepts an invitation, a server establishes a microphone connection for the first client and the second client, and issues an instruction to the first client and the second client to simultaneously start interaction. The anchor can select a specific user to initiate an interaction request, or can broadcast the interaction request to all online users, and the server randomly matches a user to perform a microphone connecting PK with the user. The method is characterized in that the connection of the wheat is an interaction mode between a main broadcast and a spectator, and between the main broadcast and the main broadcast, the identities of the main broadcast and the spectator are also converted into an initiator and a participant in the process of connecting the wheat, when the initiator initiates a wheat connection request to the participant, the participant accepts the wheat connection, a connection is established between the client sides of the initiator and the participant, and a live broadcast picture is provided by the two client sides together. In general, the live broadcast picture can be displayed in a picture-in-picture mode in which the live broadcast picture of the initiator is a large window and the main broadcast picture of the participant is a small window. Of course, the display mode can be adjusted by the initiator or the participant at will. In some examples, the continuous wheat can also be multi-person continuous wheat. In the related technology, in the live broadcast process, when a PK initiator user initiates a maskless PK to a PK recipient user, as long as the PK recipient user accepts an invitation, a connection between the PK initiator and the PK recipient is automatically established, and an illegal picture or language such as an abuse may occur in the maskless PK process, which may cause adverse effects on viewers in the live broadcast, and may bring negative effects such as reduction of the attention people to the anchor broadcast.
The embodiment of the disclosure provides a live broadcast microphone connecting control method, which is used for matching a target account with other accounts which also initiate a live broadcast microphone connecting request one by one under the condition that the target account initiates the live broadcast microphone connecting request until a candidate account which meets a preset condition with the target account microphone connecting PK is found, and then establishing the microphone connecting PK between the candidate account and the target account. The preset condition is configured that the abnormal prediction data of the connected microphones PK between the target account and the candidate account is smaller than a preset threshold value. This makes it possible to avoid the link PK with a high possibility of violation.
In the embodiment of the present disclosure, the live broadcast data of the target account includes live broadcast image information of the target account, account information of the target account, and live broadcast room information of the target account. Acquiring live broadcast data corresponding to the target account, which may specifically include:
11) extracting a preset number of live broadcast images from live broadcast segments corresponding to a target account to serve as first live broadcast images; performing feature extraction on the first live image to obtain a corresponding first image feature vector; carrying out face recognition on the first live image to obtain corresponding first image character information; anomaly detection is carried out on the first live image to obtain corresponding first image anomaly information; and generating live image information of the target account according to the first image feature vector, the first image character information and the first image abnormal information.
The method for acquiring the live image information of the target account specifically comprises the following steps:
randomly sampling X screenshots from a live broadcast segment corresponding to the target account to serve as a first live broadcast image.
The first live Image is subjected to feature extraction by using a feature extraction model, for example, the first live Image is input into a deep learning classification model (a model which can be acquired on a public network) such as Resnet50 and IncepotionV 3 which is pre-trained by using an Image Net data set, output features of a penultimate layer (namely, a fully connected layer) of the deep learning classification model are acquired, and the output features are used as a first Image feature vector corresponding to the first live Image.
When the Face recognition is performed on the first live-action image, a public Face recognition service may be specifically called, for example, a public Face recognition model Open Face is used to obtain a Face recognition result of the first live-action image, which includes features indicating human identities, such as the number of faces of each image, the average age of the faces, and the male duty ratio, and determine first image character information corresponding to the first live-action image based on the Face recognition result.
The first live-air image abnormality prediction data corresponding to the first live-air image is detected using a detection model, for example, the first live-air image is input to a detection model having an abnormal behavior recognition capability such as pornography, smoking, naked upper body, or the like, and the first image abnormality prediction data corresponding to the first live-air image output by the detection model is obtained.
The method and the device for extracting the live broadcast image feature, the character information and the image violation information from the live broadcast image extracted from the live broadcast segment corresponding to the target account are easy to implement and enrich the live broadcast image information of the target account.
12) Acquiring historical abnormal information and account identity information of the target account, and determining the account information of the target account according to the historical abnormal information and the account identity information of the target account.
And acquiring the user ID of the target account, inquiring in the violation database according to the user ID to obtain the historical abnormal information of the target account, and inquiring in the user database to obtain the account identity information of the target account. The violation database records a user ID of each user using the live application and corresponding violation information, where the violation information includes the number of times of broadcast, the number of violations, a violation proportion, and the like in a preset time period before the current time, and the preset time period may be, for example, 7 days, 30 days, or 90 days before the current time. The user database includes the user ID of each user and the corresponding account identity information, and the account identity information includes the user characteristics of the anchor, such as age, gender, city and the like.
13) And obtaining the viewer number and the live broadcast time period of the live broadcast room corresponding to the target account at the current moment, and determining the live broadcast room information of the target account according to the viewer number and the live broadcast time period of the live broadcast room corresponding to the target account at the current moment.
Specifically, each hour may be set as a time period, and the live broadcast time period corresponding to the target account is a time period to which the current time belongs, for example, when the current time is 12:30, the corresponding live broadcast time period is 12:00-13: 00. For live broadcast, some bad live broadcast contents are easy to concentrate on certain time periods, for example, the probability that live broadcast with naked pornography occurs in late night is higher, the live broadcast time periods are distinguished, so that the abnormal prediction directions of different time periods can be learned by a model, and the model prediction accuracy is improved.
14) And determining the live broadcast data of the target account according to the live broadcast image information of the target account, the account information of the target account and the live broadcast room information of the target account.
S703, determining candidate accounts from other accounts initiating the live broadcast room microphone connecting request.
Specifically, under the condition that a live broadcast room microphone connecting request initiated by a target account is received, the user is randomly selected from other accounts which are currently initiating the live broadcast room microphone connecting request to serve as candidate accounts. And matching the target account with the candidate account, and predicting whether the target account and the candidate account are suitable for performing live broadcast and microphone connection PK.
S705, acquiring live broadcast data corresponding to the candidate account.
In one possible implementation manner, the obtaining of live data corresponding to the candidate account may include the following steps:
21) extracting a preset number of live broadcast images from live broadcast segments corresponding to the candidate accounts to serve as second live broadcast images; extracting the characteristics of the second live broadcast image to obtain a corresponding second image characteristic vector, carrying out face recognition on the second live broadcast image to obtain corresponding second image character information, and carrying out abnormity detection on the second live broadcast image to obtain corresponding second image abnormity information; and generating live image information of the candidate account according to the second image feature vector, the second image character information and the second image abnormal information.
Specifically, the obtaining of live image information of the candidate account may include the following steps:
randomly sampling X screenshots from live segments corresponding to the candidate accounts to serve as second live images.
The second live Image is subjected to feature extraction by using a feature extraction model, for example, the second live Image is input into a deep learning classification model such as Resnet50 and inclusion v3 pre-trained by using an Image Net data set, the output feature of the last layer (i.e., the full connection layer) of the deep learning classification model is obtained, and the output feature is used as a second Image feature vector corresponding to the second live Image.
When the Face recognition is performed on the second live broadcast image, a public Face recognition service may be specifically called, for example, a public Face recognition model Open Face is used to obtain a Face recognition result of the second live broadcast image, where the Face recognition result includes features indicating the identity of a person, such as the number of faces of each image, the average age of the faces, and the male duty ratio, and the second image character information corresponding to the second live broadcast image is determined based on the Face recognition result.
And detecting second image abnormality information corresponding to the second live image by using the detection model, for example, inputting the second live image into the detection model with abnormal behavior recognition capability such as pornography, smoking, naked upper body and the like, and obtaining second image abnormality information corresponding to the second live image output by the detection model.
The method and the device for extracting the live broadcast image feature, the character information and the image violation information from the live broadcast image extracted from the live broadcast segment corresponding to the candidate account use the disclosed model, the feature extraction method is easy to realize, and the live broadcast image information of the candidate account is enriched.
22) Acquiring historical abnormal information and account identity information of the candidate account, and determining the account information of the candidate account according to the historical abnormal information and the account identity information of the candidate account.
And acquiring a user ID of the candidate account, inquiring in the violation database according to the user ID to obtain historical abnormal information of the candidate account, and inquiring in the user database to obtain account identity information of the candidate account. The violation database records a user ID of each user using the live application and corresponding violation information, where the violation information includes the number of times of broadcast, the number of violations, a violation proportion, and the like in a preset time period before the current time, and the preset time period may be, for example, 7 days, 30 days, or 90 days before the current time. The user database includes the user ID of each user and the corresponding account identity information, and the account identity information includes the user characteristics of the anchor, such as age, gender, city and the like.
23) And acquiring the viewer number and the live broadcast time period of the live broadcast room corresponding to the candidate account at the current moment, and determining the live broadcast room information of the candidate account according to the viewer number and the live broadcast time period of the live broadcast room corresponding to the candidate account at the current moment.
Specifically, the live broadcast time period corresponding to the candidate account is a time period to which the current time belongs, and each hour may be set to be a time period, for example, when the current time is 12:30, the corresponding live broadcast time period is 12:00-13: 00. For live broadcast, some bad live broadcast contents are easy to concentrate on certain time periods, for example, the probability that live broadcast with naked pornography occurs in late night is higher, the live broadcast time periods are distinguished, so that the abnormal prediction directions of different time periods can be learned by a model, and the model prediction accuracy is improved.
24) And determining live broadcast data of the candidate account according to the live broadcast image information of the candidate account, the account information of the candidate account and the live broadcast room information of the candidate account.
And S707, inputting the live broadcast data corresponding to the target account and the live broadcast data corresponding to the candidate account into the prediction model, and performing abnormal prediction on the wheat connection between the target account and the candidate account to obtain abnormal prediction data of the wheat connection between the target account and the candidate account.
The prediction model is trained in advance, and specifically, the prediction model can be obtained by training with a prediction model training method corresponding to fig. 4. For a detailed description of the model training process, please refer to fig. 4 and the corresponding description of the embodiment, which are not repeated herein.
Determining abnormal prediction data of the link between the target account and the candidate account based on the prediction model, specifically including: inputting live broadcast data corresponding to the target account and live broadcast data corresponding to the candidate account into a prediction model, respectively extracting characteristics of the live broadcast data corresponding to the target account and the live broadcast data corresponding to the candidate account, merging the extracted live broadcast characteristics of the target account and the candidate account, and performing probability prediction according to the merged characteristic vectors to obtain abnormal prediction data of the connecting between the target account and the candidate account.
And S709, controlling the link between the target account and the candidate account based on the abnormal prediction data.
According to the embodiment of the disclosure, according to abnormal prediction data of inter-microphone between the target account and the candidate account, inter-microphone behaviors which are easy to violate during inter-microphone PK are filtered, and the natural occurrence rate of inter-microphone PK violations is reduced. The specific implementation process comprises the following steps:
(1) comparing the abnormal prediction data of the link between the target account and the candidate account with a preset threshold, wherein the abnormal prediction data can be abnormal prediction probability, if the abnormal prediction data is smaller than the preset threshold, the abnormal prediction data of the link between the target account and the candidate account is determined to meet the preset condition, and if the abnormal prediction data is larger than or equal to the preset threshold, the abnormal prediction data of the link between the target account and the candidate account is determined not to meet the preset condition. The preset threshold may be determined according to the accuracy and recall condition of the prediction model, for example, the live broadcast segments corresponding to the output values of the prediction model are subjected to manual review, each output value and a corresponding manual review result are recorded, and the preset threshold is determined according to the output value distribution corresponding to the live broadcast segments with the violation of the review result. For example, in 1000 pieces of data, the output value corresponding to more than 98% of illegal live broadcast segments is greater than or equal to 75%, and 75% can be used as a preset threshold.
(2) Under the condition that abnormal prediction data of the wheat connection between the target account and the candidate account meet preset conditions, taking the candidate account as a wheat connection object of the target account; and establishing the connecting wheat between the target account and the connecting wheat object.
(3) Under the condition that the abnormal prediction data of the link between the target account and the candidate account is larger than or equal to a preset threshold value, the client side where the candidate account corresponding to the abnormal prediction data is located is removed from other client sides, the step of determining the candidate client side from the other client sides which are initiating the live broadcast room link request is returned to be executed, and the candidate account corresponding to the abnormal prediction data meeting the preset condition is used as the link object of the target account when the abnormal prediction data of the link between the target account and the candidate account meets the preset condition.
If the candidate account and the target account are determined to be in contact with the microphone PK easily to be violated through the steps, a candidate client can be re-determined from other clients currently initiating a live broadcast room microphone connecting request, the newly determined candidate client cannot be the client which is judged not to suggest microphone connecting PK, and whether the candidate account corresponding to the newly determined candidate client can be connected with the target account or not is judged by adopting the method in the figure 7 until a candidate account suitable for being connected with the target account is found.
The live broadcast microphone connecting control method provided by the embodiment of the disclosure obtains live broadcast data corresponding to a target account by receiving a live broadcast microphone connecting request initiated by the target account; determining a candidate account from other accounts which are initiating live broadcast room microphone connecting requests; acquiring live broadcast data corresponding to the candidate account; inputting live broadcast data corresponding to the target account and live broadcast data corresponding to the candidate account into a prediction model, and performing abnormal prediction on the link between the target account and the candidate account to obtain abnormal prediction data corresponding to the link between the target account and the candidate account; and controlling the connection between the target account and the candidate account based on the abnormal prediction data. According to the method, abnormal prediction is carried out on the wheat connecting behaviors between the target account and the candidate account in the live broadcast process, and the wheat connecting behaviors with a high probability of violation are identified and avoided based on the abnormal prediction data obtained through prediction, so that the incidence rate of bad wheat connecting is reduced, and a comfortable and healthy live broadcast and watching environment is created for a user.
Fig. 8 is a schematic structural diagram illustrating a live microphone connecting control apparatus according to an exemplary embodiment. Referring to fig. 8, the live broadcast wheat connection control apparatus 800 includes:
a first live broadcast data obtaining unit 810, configured to receive a live broadcast room microphone connecting request initiated by a target account, and obtain live broadcast data corresponding to the target account;
a candidate account determination unit 820 configured to determine a candidate account from other accounts that are initiating a live webmicrophone request;
a second live data obtaining unit 830 configured to obtain live data corresponding to the candidate account;
the abnormal prediction data acquisition unit 840 is configured to input live broadcast data corresponding to the target account and live broadcast data corresponding to the candidate account into the prediction model, perform abnormal prediction on wheat connecting between the target account and the candidate account, and obtain abnormal prediction data of wheat connecting between the target account and the candidate account;
a microphone connecting control unit 850 configured to control microphone connecting between the target account and the candidate account based on the anomaly prediction data.
Referring to fig. 9, in one possible implementation, the microphone connecting control unit 850 may include:
the first connecting control module 851 is configured to execute, when the abnormal prediction data of the connecting between the target account and the candidate account is smaller than a preset threshold, the candidate account as a connecting object of the target account and establish the connecting between the target account and the connecting object;
the second wheat connecting control module 852 is configured to, in a case that the abnormal prediction data of the wheat connecting between the target account and the candidate account is greater than or equal to the preset threshold, execute removing the candidate account corresponding to the abnormal prediction data from other accounts, and return to execute the step of determining the candidate account from other accounts which are initiating the live broadcast wheat connecting request until the abnormal prediction data of the wheat connecting between the target account and the candidate account is less than the preset threshold, and use the candidate account corresponding to the abnormal prediction data less than the preset threshold as the wheat connecting object of the target account.
In a possible implementation manner, the abnormal prediction data obtaining unit 840 is further configured to input live data corresponding to the target account and live data corresponding to the candidate account into the prediction model, perform feature extraction on the live data corresponding to the target account and the live data corresponding to the candidate account, merge the extracted live features of the target account and the live features of the candidate account, perform probability prediction according to a feature vector obtained by the merging, and obtain abnormal prediction data of the links between the target account and the candidate account.
In one possible implementation, the first direct broadcast data obtaining unit 810 may include:
a first live image information obtaining module 811 configured to extract a preset number of live images from a live clip corresponding to the target account as first live images; performing feature extraction on the first live image to obtain a corresponding first image feature vector; carrying out face recognition on the first live image to obtain corresponding first image character information; anomaly detection is carried out on the first live image to obtain corresponding first image anomaly information; generating live broadcast image information of a target account according to the first image feature vector, the first image character information and the first image abnormal information;
a first account information obtaining module 812 configured to obtain historical exception information and account identity information of the target account, and determine account information of the target account according to the historical exception information and the account identity information of the target account;
a first live broadcast room information obtaining module 813 configured to obtain the viewer number and the live broadcast time period of the live broadcast room corresponding to the target account at the current time, and determine live broadcast room information of the target account according to the viewer number and the live broadcast time period of the live broadcast room corresponding to the target account at the current time;
the first live broadcast data determining module 814 is configured to determine live broadcast data of the target account according to the live broadcast image information of the target account, the account information of the target account, and the live broadcast room information of the target account.
In one possible implementation, the second live data obtaining unit 830 may include:
a second live image information obtaining module 831 configured to extract a preset number of live images from live segments corresponding to the candidate accounts as second live images; performing feature extraction on the second live broadcast image to obtain a corresponding second image feature vector, performing face recognition on the second live broadcast image to obtain corresponding second image character information, and performing anomaly detection on the second live broadcast image to obtain corresponding second image anomaly information; generating live image information of the candidate account according to the second image feature vector, the second image character information and the second image abnormal information;
the second account information acquisition module 832 is configured to acquire historical abnormal information and account identity information of the candidate account and determine the account information of the candidate account according to the historical abnormal information and the account identity information of the candidate account;
a second live broadcasting room information obtaining module 833 configured to obtain the viewer number and the live broadcasting time period of the live broadcasting room at the current moment corresponding to the candidate account, and determine live broadcasting room information of the candidate account according to the viewer number and the live broadcasting time period of the live broadcasting room at the current moment corresponding to the candidate account;
a second live data determining module 834 configured to determine live data of the candidate account according to the live image information of the candidate account, the account information of the candidate account, and the live room information of the candidate account.
In one possible implementation manner, the live broadcast and microphone connection control apparatus 800 further includes:
the sample data acquisition unit 860 is configured to acquire a training sample set, where the training sample set includes at least one sample data with a positive label attribute and at least one sample data with a negative label attribute;
the model training unit 870 is configured to input each sample data in the training sample set into the initial deep learning model to obtain sample abnormality prediction data of each sample data in the training sample set, determine a prediction attribute of each sample data according to a comparison result between the sample abnormality prediction data of each sample data and a preset threshold, and train the initial deep learning model based on a loss between the prediction attribute of each sample data and a corresponding label attribute to obtain the prediction model.
The sample data acquiring unit 860 may include:
a historical data acquisition module 861, configured to acquire historical live microphone connecting data, where the historical live microphone connecting data includes live microphone connecting segments without abnormal behaviors and live microphone connecting segments with abnormal behaviors;
a sample live broadcast data acquisition module 862 configured to acquire sample live broadcast data corresponding to each live broadcast segment; taking sample live broadcast data corresponding to the live broadcast segment with abnormal behavior as sample data with a positive label attribute; and taking sample live broadcast data corresponding to the live broadcast segment without abnormal behaviors as sample data with the label attribute of a negative sample.
The sample live data obtaining module 862 may include:
a sample live broadcast image information obtaining sub-module 8621 configured to extract a preset number of live broadcast images from the live broadcast segment as sample live broadcast images; performing feature extraction on a sample live image to obtain a corresponding sample image feature vector, performing face recognition on the sample live image to obtain sample image character information, and performing anomaly detection on the sample live image to obtain sample image anomaly prediction data; generating sample live broadcast image information according to the sample image feature vector, the sample image character information and the sample image abnormal prediction data;
the sample account information obtaining sub-module 8622 is configured to determine a sample account corresponding to the live microphone connecting segment, obtain historical abnormal information and account identity information of the sample account, and determine sample account information according to the historical abnormal information and the account identity information of the sample account;
a sample live broadcast room information obtaining submodule 8623 configured to obtain the audience number and the live broadcast time period of a live broadcast room corresponding to the live broadcast connected with the microphone, and determine sample live broadcast room information according to the audience number and the live broadcast time period of the live broadcast room corresponding to the live broadcast connected with the microphone;
a sample live data determining sub-module 8624 configured to determine sample live data according to the sample live image information, the sample account information, and the sample live room information.
The live broadcast microphone connecting control device provided by the embodiment of the disclosure obtains live broadcast data corresponding to a target account by receiving a live broadcast microphone connecting request initiated by the target account; determining a candidate account from other accounts which are initiating live broadcast room microphone connecting requests; acquiring live broadcast data corresponding to the candidate account; inputting live broadcast data corresponding to the target account and live broadcast data corresponding to the candidate account into a prediction model, and performing abnormal prediction on the link between the target account and the candidate account to obtain abnormal prediction data corresponding to the link between the target account and the candidate account; and controlling the connection between the target account and the candidate account based on the abnormal prediction data. According to the method, the abnormal prediction is carried out on the wheat connecting behaviors between the target account and the candidate account in the live broadcast process, the larger the predicted abnormal prediction data is, the higher the probability of abnormal contents such as violation and the like in the live broadcast of the target account and the candidate account is, therefore, the wheat connecting between the two accounts with lower abnormal probability after the wheat connecting can be allowed, and the processing mode without the establishment of the wheat connecting can be given to the two accounts with higher abnormal content probability after the wheat connecting, so that the wheat connecting with violation probability can be identified and avoided, the incidence rate of bad wheat connecting is reduced, and comfortable and healthy live broadcast and watching environments are created for users.
With regard to the apparatus in the above-described embodiment, the specific manner in which each module performs the operation has been described in detail in the embodiment related to the method, and will not be elaborated here.
Fig. 10 is a block diagram illustrating a terminal according to an example embodiment. The terminal is used for executing the steps corresponding to the method for training the prediction model, and may be a portable mobile terminal, such as: a smart phone, a tablet computer, an MP3 player (Moving Picture Experts Group Audio Layer III, motion video Experts compression standard Audio Layer 3), an MP4 player (Moving Picture Experts Group Audio Layer IV, motion video Experts compression standard Audio Layer 4), a notebook computer, or a desktop computer. A terminal may also be referred to by other names such as user equipment, portable terminal, laptop terminal, desktop terminal, etc.
Referring to fig. 10, the terminal may include one or more of the following components: processing component 902, memory 904, power component 906, multimedia component 908, audio component 910, interface to input/output (I/O) 99, sensor component 914, and communication component 916.
The processing component 902 generally controls the overall operation of the terminal, such as operations associated with display, telephone calls, data communications, camera operations, and recording operations. Processing component 902 may include one or more processors 920 to execute instructions to perform all or a portion of the steps of the methods described above. Further, processing component 902 can include one or more modules that facilitate interaction between processing component 902 and other components. For example, the processing component 902 can include a multimedia module to facilitate interaction between the multimedia component 908 and the processing component 902.
The memory 904 is configured to store various types of data to support operations at the terminal. Examples of such data include instructions for any application or method operating on the terminal, contact data, phonebook data, messages, images, videos, and so forth. The memory 904 may be implemented by any type or combination of volatile or non-volatile storage devices, such as Static Random Access Memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic memory, flash memory, magnetic or optical disks.
The power supply component 906 provides power to the various components of the terminal. The power components 906 may include a power management system, one or more power supplies, and other components associated with generating, managing, and distributing power for the terminal.
The multimedia components 908 include a screen that provides an output interface between the terminal and the user. In some embodiments, the screen may include a Liquid Crystal Display (LCD) and a Touch Panel (TP). If the screen includes a touch panel, the screen may be implemented as a touch screen to receive an input signal from a user. The touch panel includes one or more touch sensors to sense touch, slide, and gestures on the touch panel. The touch sensor may not only sense the boundary of a touch or slide action, but also detect the duration and pressure associated with the touch or slide operation. In some embodiments, the multimedia component 908 includes a front facing camera and/or a rear facing camera. The front camera and/or the rear camera may receive external multimedia data when the terminal is in an operation mode, such as a photographing mode or a video mode. Each front camera and rear camera may be a fixed optical lens system or have a focal length and optical zoom capability.
The audio component 910 is configured to output and/or input audio signals. For example, the audio component 910 includes a Microphone (MIC) configured to receive external audio signals when the terminal is in an operating mode, such as a call mode, a recording mode, and a voice recognition mode. The received audio signal may further be stored in the memory 904 or transmitted via the communication component 916. In some embodiments, audio component 910 also includes a speaker for outputting audio signals.
The I/O interface 99 provides an interface between the processing component 902 and peripheral interface modules, which may be keyboards, click wheels, buttons, etc. These buttons may include, but are not limited to: a home button, a volume button, a start button, and a lock button.
The sensor component 914 includes one or more sensors for providing various aspects of state assessment for the terminal. For example, the sensor component 914 may detect an open/closed state of the terminal, the relative positioning of components, such as a display and keypad of the terminal, the sensor component 914 may also detect a change in position of the terminal or a component of the terminal, the presence or absence of user contact with the terminal, orientation or acceleration/deceleration of the terminal, and a change in temperature of the terminal. The sensor assembly 914 may include a proximity sensor configured to detect the presence of a nearby object in the absence of any physical contact. The sensor assembly 914 may also include a light sensor, such as a CMOS or CCD image sensor, for use in imaging applications. In some embodiments, the sensor assembly 914 may also include an acceleration sensor, a gyroscope sensor, a magnetic sensor, a pressure sensor, or a temperature sensor.
The communication component 916 is configured to facilitate communications between the terminal and other devices in a wired or wireless manner. The terminal may access a wireless network based on a communication standard, such as WiFi, an operator network (such as 2G, 3G, 4G, or 9G), or a combination thereof. In an exemplary embodiment, the communication component 916 receives a broadcast signal or broadcast associated information from an external broadcast management system via a broadcast channel. In an exemplary embodiment, the communication component 916 further includes a Near Field Communication (NFC) module to facilitate short-range communications. For example, the NFC module may be implemented based on Radio Frequency Identification (RFID) technology, infrared data association (IrDA) technology, Ultra Wideband (UWB) technology, Bluetooth (BT) technology, and other technologies.
In an exemplary embodiment, the terminal may be implemented by one or more Application Specific Integrated Circuits (ASICs), Digital Signal Processors (DSPs), Digital Signal Processing Devices (DSPDs), Programmable Logic Devices (PLDs), Field Programmable Gate Arrays (FPGAs), controllers, micro-controllers, microprocessors or other electronic components for performing the above-described methods.
In an exemplary embodiment, a storage medium comprising instructions, such as the memory 904 comprising instructions, executable by the processor 920 of the terminal to perform the above-described method is also provided. Alternatively, the storage medium may be a non-transitory computer readable storage medium, for example, the non-transitory computer readable storage medium may be a ROM, a Random Access Memory (RAM), a CD-ROM, a magnetic tape, a floppy disk, an optical data storage device, and the like.
In an exemplary embodiment, a computer program product is also provided, which comprises readable program code executable by the processor 920 of the terminal to perform the above-described method. Alternatively, the program code may be stored in a storage medium of the terminal, which may be a non-transitory computer-readable storage medium, for example, a ROM, a Random Access Memory (RAM), a CD-ROM, a magnetic tape, a floppy disk, an optical data storage device, and the like.
Fig. 11 is a block diagram illustrating a server according to an exemplary embodiment, where the server may be configured to execute steps corresponding to the above-described live microphone connection control method. Referring to fig. 11, server 1000 includes a processing component 1010 that further includes one or more processors and memory resources, represented by memory 1020, for storing instructions, such as applications, that are executable by processing component 1010. The application programs stored in memory 1020 may include one or more modules that each correspond to a set of instructions. Further, the processing component 1010 is configured to execute instructions to perform the content presentation methods described above.
The server 1000 may also include a power pack 1030 configured to perform power management of the server 1000, a wired or wireless network interface 1050 configured to connect the server 1000 to a network, and an input/output (I/O) interface 1040. The server 1000 may operate based on an operating system stored in memory 1020, such as Windows Server, Mac OS XTM, UnixTM, LinuxTM, FreeBSDTM, or the like.
In an exemplary embodiment, a computer program product or computer program is also provided, the computer program product or computer program comprising computer instructions stored in a computer readable storage medium. The processor of the computer device reads the computer instructions from the computer-readable storage medium, and executes the computer instructions, so that the computer device executes the live broadcast connection control method provided in the various optional implementation manners.
In an exemplary embodiment, an electronic device is further provided, where the electronic device includes a processor and a memory, and the memory is used to store processor executable instructions, where the processor is configured to execute the executable instructions to implement a live microphone connection control method corresponding to fig. 7.
The memory may be used to store software programs and modules, and the processor may execute various functional applications and data processing by operating the software programs and modules stored in the memory. The memory can mainly comprise a program storage area and a data storage area, wherein the program storage area can store an operating system, application programs needed by functions and the like; the storage data area may store data created according to use of the apparatus, and the like. Further, the memory may include high speed random access memory, and may also include non-volatile memory, such as at least one magnetic disk storage device, flash memory device, or other volatile solid state storage device. Accordingly, the memory may also include a memory controller to provide the processor access to the memory.
Other embodiments of the disclosure will be apparent to those skilled in the art from consideration of the specification and practice of the disclosure disclosed herein. This application is intended to cover any variations, uses, or adaptations of the disclosure following, in general, the principles of the disclosure and including such departures from the present disclosure as come within known or customary practice within the art to which the disclosure pertains. It is intended that the specification and examples be considered as exemplary only, with a true scope and spirit of the disclosure being indicated by the following claims.
It will be understood that the present disclosure is not limited to the precise arrangements described above and shown in the drawings and that various modifications and changes may be made without departing from the scope thereof. The scope of the present disclosure is limited only by the appended claims.

Claims (10)

1. A live broadcast wheat connection control method is characterized by comprising the following steps:
receiving a live broadcast room microphone connecting request initiated by a client where a target account is located, and acquiring live broadcast data corresponding to the target account;
determining candidate clients from other clients initiating live broadcast room microphone connecting requests;
acquiring live broadcast data corresponding to a candidate account for logging in the candidate client;
inputting live broadcast data corresponding to the target account and live broadcast data corresponding to the candidate account into a prediction model, and performing abnormal prediction on wheat connection between the target account and the candidate account to obtain abnormal prediction data of the wheat connection between the target account and the candidate account;
controlling the liaison between the target account and the candidate account based on the abnormal prediction data.
2. The method of claim 1, wherein the controlling the liaison between the target account and the candidate account based on the anomaly prediction data comprises:
under the condition that the abnormal prediction data of the link between the target account and the candidate account meet the preset conditions, taking the candidate account as a link object of the target account;
and establishing the wheat connecting between the target account and the wheat connecting object.
3. The method of claim 2, further comprising:
and under the condition that the abnormal prediction data of the link between the target account and the candidate account does not meet the preset condition, removing the client where the candidate account corresponding to the abnormal prediction data is located from other clients, and returning to execute the step of determining the candidate client from other clients initiating the live broadcast room link request until the abnormal prediction data of the link between the target account and the candidate account meets the preset condition, wherein the candidate account corresponding to the abnormal prediction data meeting the preset condition is used as the link object of the target account.
4. The method of claim 1, wherein the inputting live data corresponding to the target account and live data corresponding to the candidate account into a prediction model, performing abnormal prediction on wheat connecting between the target account and the candidate account, and obtaining abnormal prediction data of wheat connecting between the target account and the candidate account comprises:
inputting live broadcast data corresponding to the target account and live broadcast data corresponding to the candidate account into a prediction model, respectively extracting features of the live broadcast data corresponding to the target account and the live broadcast data corresponding to the candidate account, merging the extracted live broadcast features of the target account and the live broadcast features of the candidate account, performing abnormal prediction according to the feature vectors obtained by merging, and obtaining abnormal prediction data of the connection between the target account and the candidate account.
5. The method of claim 1, wherein the predictive model is determined by:
acquiring a training sample set, wherein the training sample set comprises at least one sample data with a label attribute of a positive sample and at least one sample data with a label attribute of a negative sample;
inputting each sample data in the training sample set into an initial deep learning model to obtain sample abnormality prediction data of each sample data in the training sample set;
determining the prediction attribute of each sample data according to the comparison result between the sample abnormity prediction data of each sample data and a preset threshold value;
and training the initial deep learning model based on the loss between the prediction attribute of each sample data and the corresponding label attribute to obtain the prediction model.
6. The method of claim 5, wherein the obtaining a set of training samples comprises:
obtaining historical live broadcast data, wherein the historical live broadcast data comprises live broadcast segments without abnormal behaviors and live broadcast segments with abnormal behaviors;
acquiring sample live broadcast data corresponding to each live broadcast segment;
taking sample live broadcast data corresponding to the live broadcast segment with abnormal behavior as sample data with a positive label attribute; and taking sample live broadcast data corresponding to the live broadcast segment without abnormal behaviors as sample data with the label attribute of a negative sample.
7. The method as claimed in claim 6, wherein the obtaining of sample live broadcast data corresponding to each live broadcast segment includes:
extracting a preset number of live broadcast images from live broadcast segments of continuous wheat as sample live broadcast images; performing feature extraction on the sample live image to obtain a corresponding sample image feature vector, performing face recognition on the sample live image to obtain sample image character information, and performing anomaly detection on the sample live image to obtain sample image anomaly prediction data; generating sample live broadcast image information according to the sample image feature vector, the sample image character information and the sample image abnormal prediction data;
determining a sample account corresponding to the live microphone connecting segment, acquiring historical abnormal information and account identity information of the sample account, and determining sample account information according to the historical abnormal information and the account identity information of the sample account;
acquiring audience number and live broadcast time period of a live broadcast room corresponding to the continuous live broadcast segment, and determining sample live broadcast room information according to the audience number and the live broadcast time period of the live broadcast room corresponding to the continuous live broadcast segment;
and determining the sample live broadcast data according to the sample live broadcast image information, the sample account information and the sample live broadcast room information.
8. A live broadcast wheat connecting control device is characterized by comprising:
the system comprises a first live broadcast data acquisition unit, a second live broadcast data acquisition unit and a third live broadcast data acquisition unit, wherein the first live broadcast data acquisition unit is configured to receive a live broadcast room microphone connecting request initiated by a target account and acquire live broadcast data corresponding to the target account;
a candidate account determination unit configured to determine a candidate account from other accounts that are initiating live broadcast room microphone connecting requests;
the second live broadcast data acquisition unit is configured to acquire live broadcast data corresponding to the candidate account;
an abnormal prediction data acquisition unit, configured to input live data corresponding to the target account and live data corresponding to the candidate account into a prediction model, perform abnormal prediction on wheat connecting between the target account and the candidate account, and obtain abnormal prediction data of wheat connecting between the target account and the candidate account;
a microphone connecting control unit configured to control microphone connecting between the target account and the candidate account based on the abnormality prediction data.
9. An electronic device, comprising:
a processor;
a memory for storing the processor-executable instructions;
wherein the processor is configured to execute the instructions to implement the live microphone connection control method of any one of claims 1 to 7.
10. A storage medium in which instructions, when executed by a processor of an electronic device, enable the electronic device to perform a live microphone attachment control method as claimed in any one of claims 1 to 7.
CN202011521771.7A 2020-12-21 2020-12-21 Live broadcast continuous wheat control method, device and storage medium Active CN114727119B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202011521771.7A CN114727119B (en) 2020-12-21 2020-12-21 Live broadcast continuous wheat control method, device and storage medium

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202011521771.7A CN114727119B (en) 2020-12-21 2020-12-21 Live broadcast continuous wheat control method, device and storage medium

Publications (2)

Publication Number Publication Date
CN114727119A true CN114727119A (en) 2022-07-08
CN114727119B CN114727119B (en) 2023-10-10

Family

ID=82229971

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202011521771.7A Active CN114727119B (en) 2020-12-21 2020-12-21 Live broadcast continuous wheat control method, device and storage medium

Country Status (1)

Country Link
CN (1) CN114727119B (en)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN116451139A (en) * 2023-06-16 2023-07-18 杭州新航互动科技有限公司 Live broadcast data rapid analysis method based on artificial intelligence

Citations (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20130110761A1 (en) * 2011-10-31 2013-05-02 Krishnamurthy Viswanathan System and method for ranking anomalies
CN109168051A (en) * 2018-09-11 2019-01-08 天津理工大学 A kind of network direct broadcasting platform supervision evidence-obtaining system based on blue-ray storage
CN109558990A (en) * 2018-12-15 2019-04-02 国网福建省电力有限公司 A kind of power distribution network based on steiner tree model is taken precautions against natural calamities bulk transmission grid planing method
CN111182314A (en) * 2018-11-12 2020-05-19 阿里巴巴集团控股有限公司 Live stream processing method and device and data processing method
CN111314721A (en) * 2020-02-11 2020-06-19 北京达佳互联信息技术有限公司 Method, device, equipment and medium for determining abnormal live broadcast
US20200288201A1 (en) * 2017-09-13 2020-09-10 Hong Kong Liveme Corporation Limited Fraud prevention method, operation detection method and apparatus, and electronic device
CN111800647A (en) * 2020-06-29 2020-10-20 广州市百果园信息技术有限公司 Live broadcast and live broadcast matching method, device, equipment and storage medium

Patent Citations (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20130110761A1 (en) * 2011-10-31 2013-05-02 Krishnamurthy Viswanathan System and method for ranking anomalies
US20200288201A1 (en) * 2017-09-13 2020-09-10 Hong Kong Liveme Corporation Limited Fraud prevention method, operation detection method and apparatus, and electronic device
CN109168051A (en) * 2018-09-11 2019-01-08 天津理工大学 A kind of network direct broadcasting platform supervision evidence-obtaining system based on blue-ray storage
CN111182314A (en) * 2018-11-12 2020-05-19 阿里巴巴集团控股有限公司 Live stream processing method and device and data processing method
CN109558990A (en) * 2018-12-15 2019-04-02 国网福建省电力有限公司 A kind of power distribution network based on steiner tree model is taken precautions against natural calamities bulk transmission grid planing method
CN111314721A (en) * 2020-02-11 2020-06-19 北京达佳互联信息技术有限公司 Method, device, equipment and medium for determining abnormal live broadcast
CN111800647A (en) * 2020-06-29 2020-10-20 广州市百果园信息技术有限公司 Live broadcast and live broadcast matching method, device, equipment and storage medium

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN116451139A (en) * 2023-06-16 2023-07-18 杭州新航互动科技有限公司 Live broadcast data rapid analysis method based on artificial intelligence
CN116451139B (en) * 2023-06-16 2023-09-01 杭州新航互动科技有限公司 Live broadcast data rapid analysis method based on artificial intelligence

Also Published As

Publication number Publication date
CN114727119B (en) 2023-10-10

Similar Documents

Publication Publication Date Title
US20210105535A1 (en) Control method of playing content and content playing apparatus performing the same
CN110519636B (en) Voice information playing method and device, computer equipment and storage medium
CN113709561B (en) Video editing method, device, equipment and storage medium
CN110557659B (en) Video recommendation method and device, server and storage medium
US20170065889A1 (en) Identifying And Extracting Video Game Highlights Based On Audio Analysis
US10541000B1 (en) User input-based video summarization
KR101895846B1 (en) Facilitating television based interaction with social networking tools
US10015385B2 (en) Enhancing video conferences
CN109783656B (en) Recommendation method and system of audio and video data, server and storage medium
US20150324099A1 (en) Connecting Current User Activities with Related Stored Media Collections
CN113766268B (en) Video processing method and device, electronic equipment and readable medium
WO2017210644A1 (en) System and method for a platform to identify and connect like-minded individuals based on interaction
CN112954390A (en) Video processing method, device, storage medium and equipment
CN114727119B (en) Live broadcast continuous wheat control method, device and storage medium
US20140012792A1 (en) Systems and methods for building a virtual social network
CN115994266A (en) Resource recommendation method, device, electronic equipment and storage medium
CN114302231B (en) Video processing method and device, electronic equipment and storage medium
CN109740557A (en) Method for checking object and device, electronic equipment and storage medium
US11778277B1 (en) Digital item processing for video streams
CN112165626B (en) Image processing method, resource acquisition method, related equipment and medium
CN110493609B (en) Live broadcast method, terminal and computer readable storage medium
CN114501132B (en) Resource processing method and device, electronic equipment and storage medium
CN111949808A (en) Multimedia content similarity determination method and device, electronic equipment and storage medium
JP6411274B2 (en) Timing correction system, method and program thereof
CN115022654B (en) Video editing method and device in live broadcast scene

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination
GR01 Patent grant
GR01 Patent grant