Data detection method and apparatus, computer, readable storage medium, and program product
By acquiring and integrating various information features of public broadcasting service group accounts, and using machine learning models for prediction, the problem of low account detection accuracy in existing technologies has been solved, achieving higher detection accuracy and efficiency.
Patent Information
- Application Number
- CN202111250379.8
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2021-10-26
- Publication Date
- 2025-12-12
- Estimated Expiration
- 2041-10-26
AI Technical Summary
In existing technologies, keyword detection of text content from searched public broadcasting service group accounts is prone to misjudgment, resulting in low accuracy of account detection.
By acquiring the account information, entity information, and registration information to be detected, extracting their respective features, and fusing features based on the correlation, generating neighborhood correlation information and account fusion features, and using machine learning models for prediction, the account detection results are obtained.
It improves the accuracy and efficiency of data detection, reduces false positives, and ensures the accuracy of account detection.
Smart Images

Figure CN116028857B_ABST
Abstract
Description
TECHNICAL FIELD
[0001] The present application relates to the technical field of computers, and in particular to a data detection method and device, a computer, a readable storage medium, and a program product. BACKGROUND
[0002] For a search product, the quality of search results reflects the quality of the search product, and also affects the user experience of the search product, that is, the more normal results in the search results, the fewer abnormal results, the better the user experience, and the better the quality of the search product. For example, when searching for a public broadcasting business group account, if an abnormal public broadcasting business group account is found, the user may not be able to access the required service when accessing the abnormal public broadcasting business group account, resulting in waste of time and the like, thereby resulting in poor experience of the search product, and further resulting in poor operation effect of the search product. Therefore, it is necessary to screen the public broadcasting business group account searched by the search product. At present, the text content in the searched public broadcasting business group account is generally analyzed by text analysis, and the detection result of the public broadcasting business group account is determined based on the analyzed keywords. However, the keyword detection of the public broadcasting business group account often leads to misjudgment, resulting in low accuracy of account detection. SUMMARY
[0003] The embodiments of the present application provide a data detection method, device, computer, and readable storage medium, which can improve the accuracy and detection efficiency of data detection.
[0004] The embodiments of the present application provide a data detection method, device, computer, and readable storage medium, which can improve the accuracy and detection efficiency of data detection.
[0005] Obtaining to-be-detected account information, to-be-detected subject information, and to-be-detected registration information of a to-be-detected public broadcasting business group account;
[0006] Extracting to-be-detected account features of the to-be-detected account information, extracting to-be-detected subject features of the to-be-detected subject information, and generating random registration features of the to-be-detected registration information;
[0007] Obtaining a first association relationship between the to-be-detected account information and the to-be-detected subject information, and obtaining a second association relationship between the to-be-detected account information and the to-be-detected registration information;
[0008] Based on the first association relationship and the second association relationship, performing feature fusion on the to-be-detected subject features and the random registration features to obtain neighborhood association information;
[0009] The account feature to be detected is fused with the neighborhood association information to obtain account fusion features, the account fusion features are predicted to obtain an account detection result of the public broadcast service group account to be detected.
[0010] The embodiment of the application provides a data detection method, which comprises the following steps:
[0011] Obtaining a public broadcast service group account sample, obtaining account information samples, subject information samples and registration information samples of the public broadcast service group account sample, and obtaining account sample labels of the public broadcast service group account sample;
[0012] The account information samples, the subject information samples and the registration information samples are input into an initial account detection model, sample account features of the account information samples are extracted in a text detection network of the initial account detection model, sample subject features of the subject information samples are extracted in the text detection network, and sample registration features of the registration information samples are obtained by random initialization in a random initialization network of the initial account detection model;
[0013] Obtaining a first sample association relationship between the account information samples and the subject information samples, and obtaining a second sample association relationship between the account information samples and the registration information samples;
[0014] Based on the first sample association relationship and the second sample association relationship, the sample subject features and the sample registration features are fused to obtain sample neighborhood association information;
[0015] The sample account features are fused with the sample neighborhood association information to obtain sample fusion features, and the sample fusion features are predicted based on a classification network of the initial account detection model to obtain sample detection results of the public broadcast service group account sample;
[0016] The initial account detection model is adjusted based on the account sample labels and the sample detection results to obtain an account detection model.
[0017] The embodiment of the application provides a data detection device, which comprises the following steps:
[0018] An information acquisition module is configured to acquire account information to be detected, subject information to be detected and registration information to be detected of a public broadcast service group account to be detected;
[0019] An account feature generation module is configured to extract account features to be detected from the account information to be detected;
[0020] A subject feature generation module is configured to extract subject features to be detected from the subject information to be detected;
[0021] The registration feature generation module is used to generate random registration features for the registration information to be detected;
[0022] The relationship acquisition module is used to acquire the first association between the account information to be detected and the subject information to be detected, and to acquire the second association between the account information to be detected and the registration information to be detected.
[0023] The information generation module is used to perform feature fusion between the features of the subject to be detected and the randomly registered features based on the first association relationship and the second association relationship to obtain neighborhood association information.
[0024] The information fusion module is used to fuse the features of the account to be detected with the neighboring association information to obtain the account fusion features;
[0025] The account prediction module is used to predict account fusion features and obtain the account detection results for the public broadcast service group accounts to be detected.
[0026] The account feature generation module includes:
[0027] The first feature extraction unit is used to perform word segmentation on the account information to be detected, obtain m account word groups to be detected, and perform vector mapping and concatenation on the m account word groups to be detected to obtain the initial account features of the account information to be detected; m is a positive integer.
[0028] The first feature transformation unit is used to perform convolutional transformation on the initial account features to obtain the account features to be detected of the account information to be detected.
[0029] The main feature generation module includes:
[0030] The second feature extraction unit is used to perform word segmentation on the subject information to be detected, obtain m subject word groups to be detected, and perform vector mapping and concatenation on the m subject word groups to be detected to obtain the initial subject features of the subject information to be detected.
[0031] The second feature transformation unit is used to perform convolution transformation on the initial subject features to obtain the subject features to be detected.
[0032] The registration feature generation module includes:
[0033] The type acquisition unit is used to input the registration information to be detected into the random initialization network and identify the target registration type of the registration information to be detected in the random initialization network.
[0034] The feature mapping unit is used to obtain the registration mapping matrix corresponding to the target registration type and generate random registration features of the registration information to be detected based on the registration mapping matrix.
[0035] The information generation module comprises:
[0036] The first association determination unit is configured to perform feature fusion on the to-be-detected account feature and the to-be-detected principal feature based on the first association relationship, to obtain a first feature association degree corresponding to the first association relationship.
[0037] The second association determination unit is configured to perform feature fusion on the to-be-detected account feature and the random registration feature based on the second association relationship, to obtain a second feature association degree corresponding to the second association relationship.
[0038] The feature weighting unit is configured to perform weighted summation on the to-be-detected principal feature and the random registration feature by using the first feature association degree and the second feature association degree, to obtain the neighborhood association information.
[0039] The first association determination unit comprises:
[0040] The linear mapping sub-unit is configured to map the to-be-detected account feature into a to-be-detected account linear feature, to obtain a first information type corresponding to the to-be-detected principal information, and to map the to-be-detected principal feature into a to-be-detected principal vector based on a first node mapping matrix corresponding to the first information type.
[0041] The feature fusion sub-unit is configured to perform feature fusion on the to-be-detected account linear feature and the to-be-detected principal vector based on the first association relationship, to obtain the first feature association degree of the first association relationship.
[0042] The feature fusion sub-unit comprises:
[0043] The feature multiplication sub-unit is configured to perform multiplication processing on the to-be-detected account linear feature and the to-be-detected principal vector based on the first association relationship, to obtain a first initial association degree.
[0044] The importance acquisition sub-unit is configured to acquire a first tuple importance of a first association tuple composed of the to-be-detected account information, the to-be-detected principal information, and the first association relationship.
[0045] The association optimization sub-unit is configured to perform optimization processing on the first initial association degree based on the first tuple importance, to obtain the first feature association degree of the first association relationship.
[0046] The feature weighting unit comprises:
[0047] The association normalization sub-unit is configured to perform normalization processing on the first feature association degree and the second feature association degree, to obtain a first association weight corresponding to the first feature association degree and a second association weight corresponding to the second feature association degree.
[0048] The information weighting subunit is configured to perform weighted summation on the to-be-detected principal feature and the random registration feature based on the first correlation weight and the second correlation weight to obtain the neighborhood correlation information.
[0049] The feature weighting unit comprises:
[0050] The principal transmission subunit is configured to acquire a first correlation type of a first correlation relationship corresponding to the to-be-detected principal feature, acquire a first information weight matrix corresponding to the first correlation type, and perform correlation information fusion on the to-be-detected principal feature through the first information weight matrix to obtain a principal transmission feature corresponding to the to-be-detected principal feature.
[0051] The registration transmission subunit is configured to acquire a second correlation type of a second correlation relationship corresponding to the random registration feature, acquire a second information weight matrix corresponding to the second correlation type, and perform correlation information fusion on the random registration feature through the second information weight matrix to obtain a registration transmission feature corresponding to the random registration feature.
[0052] The transmission weighting subunit is configured to perform weighted summation on the principal transmission feature and the registration transmission feature through the first feature correlation degree and the second feature correlation degree to obtain the neighborhood correlation information.
[0053] The account prediction module comprises:
[0054] The probability prediction unit is configured to predict the account fusion feature to obtain a normal probability corresponding to an account normal result label and an abnormal probability corresponding to an account abnormal result label.
[0055] The normal determination unit is configured to determine that the account detection result of the to-be-detected public broadcast service group account is the account normal result if the normal probability is greater than the abnormal probability.
[0056] The abnormal determination unit is configured to determine that the account detection result of the to-be-detected public broadcast service group account is the account abnormal result if the normal probability is less than or equal to the abnormal probability.
[0057] The device further comprises:
[0058] The account output module is configured to output the to-be-detected public broadcast service group account if the account detection result of the to-be-detected public broadcast service group account is the account normal result.
[0059] The abnormal feedback module is configured to delete the to-be-detected public broadcast service group account if the account detection result of the to-be-detected public broadcast service group account is the account abnormal result, and send an account abnormal message to a service device associated with the to-be-detected public broadcast service group account based on the to-be-detected registration information.
[0060] The embodiment of the application provides a data detection device, which comprises:
[0061] A sample acquisition module is configured to acquire a public broadcast service group account sample, acquire account information samples, subject information samples and registration information samples of the public broadcast service group account sample, and acquire account sample labels of the public broadcast service group account sample.
[0062] A first feature extraction module is configured to input the account information samples, the subject information samples and the registration information samples into an initial account detection model, and extract sample account features of the account information samples in a text detection network of the initial account detection model.
[0063] A second feature extraction module is configured to extract sample subject features of the subject information samples in the text detection network.
[0064] A third feature extraction module is configured to perform random initialization on the registration information samples in a random initialization network of the initial account detection model, and obtain sample registration features of the registration information samples.
[0065] A sample relationship acquisition module is configured to acquire a first sample correlation between the account information samples and the subject information samples, and acquire a second sample correlation between the account information samples and the registration information samples.
[0066] A sample information acquisition module is configured to perform feature fusion on the sample subject features and the sample registration features based on the first sample correlation and the second sample correlation, and obtain sample neighborhood correlation information.
[0067] A sample information fusion module is configured to perform feature fusion on the sample account features and the sample neighborhood correlation information, and obtain sample fusion features.
[0068] A sample prediction module is configured to predict the sample fusion features based on a classification network of the initial account detection model, and obtain sample detection results of the public broadcast service group account sample.
[0069] A model training module is configured to perform parameter adjustment on the initial account detection model according to the account sample labels and the sample detection results, and obtain an account detection model.
[0070] The device further comprises:
[0071] A graph establishment module is configured to acquire account promotion data and account menus of the public broadcast service group account sample, acquire an account relationship heterogeneous graph associated with the public broadcast service group account sample according to the account promotion data and the account menus.
[0072] The model optimization module is configured to obtain adjacent account information associated with the registration information sample based on the account relationship heterogeneous graph, and to perform parameter optimization on the account detection model based on the adjacent account information, thereby obtaining an optimized account detection model.
[0073] The embodiment of the present application provides a computer device, including a processor, a memory, an input and output interface;
[0074] The processor is connected with the memory and the input and output interface respectively, wherein the input and output interface is configured to receive and output data, the memory is configured to store a computer program, and the processor is configured to call the computer program, so that the computer device including the processor executes the data detection method in the embodiment of the present application.
[0075] The embodiment of the present application provides a computer readable storage medium, and the computer readable storage medium stores a computer program, the computer program is suitable for being loaded and executed by a processor, so that a computer device having the processor executes the data detection method in the embodiment of the present application.
[0076] The embodiment of the present application provides a computer program product or a computer program, and the computer program product or the computer program includes 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 the processor executes the computer instructions, so that the computer device executes the method provided in various optional manners in the embodiment of the present application. In other words, the computer instructions are executed by the processor to realize the method provided in various optional manners in the embodiment of the present application.
[0077] The embodiment of the present application is implemented, and the following beneficial effects are obtained:
[0078] In the embodiment of the present application, the account information to be detected, the subject information to be detected and the registration information to be detected of the public broadcast service group account to be detected are acquired; the account feature to be detected of the account information to be detected is extracted, the subject feature to be detected of the subject information to be detected is extracted, and the random registration feature of the registration information to be detected is generated; the first association relationship between the account information to be detected and the subject information to be detected is acquired, and the second association relationship between the account information to be detected and the registration information to be detected is acquired; based on the first association relationship and the second association relationship, the subject feature to be detected and the random registration feature are fused to obtain neighborhood association information; the account feature to be detected and the neighborhood association information are fused to obtain account fusion features, and the account fusion features are predicted to obtain the account detection result of the public broadcast service group account to be detected. Through the association analysis of the public broadcast service group account to be detected, other features are added based on the basic information (i.e. the account information to be detected) of the public broadcast service group account to be detected, so that the account information to be detected can be enriched, and the association relationship between different nodes can be considered, and other information features can be added in the account information to be detected, thereby improving the accuracy of data detection. BRIEF DESCRIPTION OF DRAWINGS
[0079] In order to more clearly illustrate the technical solutions of the embodiments of the present application or the prior art, the drawings needed to be used in the embodiments or the prior art description will be briefly introduced. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor.
[0080] Figure 1 is a network interaction architecture diagram of data detection provided by an embodiment of the present application;
[0081] Figure 2 is a data detection scene schematic diagram provided by an embodiment of the present application;
[0082] Figure 3 is a method flowchart of data detection provided by an embodiment of the present application;
[0083] Figure 4 is a model architecture diagram provided by an embodiment of the present application;
[0084] Figure 5 is a node association graph schematic diagram provided by an embodiment of the present application;
[0085] Figure 6 is a data detection scene schematic diagram provided by an embodiment of the present application;
[0086] Figure 7is a model training method schematic diagram provided by an embodiment of the present application.
[0087] Figure 8 is a data detection device schematic diagram provided by an embodiment of the present application.
[0088] Figure 9 is another data detection device schematic diagram provided by an embodiment of the present application.
[0089] Figure 10 is a structural schematic diagram of a computer device provided by an embodiment of the present application. DETAILED DESCRIPTION
[0090] The technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only some of the embodiments of the present application, but not all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative work fall within the scope of protection of the present application.
[0091] In the embodiments of the present application, machine learning technology in the field of artificial intelligence can be involved. Artificial intelligence (AI) is the theory, method, technology and application system for using digital computers or computer-controlled machines to simulate, extend and expand human intelligence, perceive environment, acquire knowledge and use knowledge to obtain optimal results. In other words, artificial intelligence is a comprehensive technology of computer science, which aims to understand the essence of intelligence and produce a new intelligent machine that can react in a similar way to human intelligence. Artificial intelligence is the design principle and implementation method of various intelligent machines, so that machines have the functions of perception, reasoning and decision-making. Through artificial intelligence, the detection of the public broadcasting service group account to be detected is realized, and the account detection result detected in a similar way to human intelligence is obtained.
[0092] Artificial intelligence technology is a comprehensive discipline, involving a wide range of fields, including both hardware and software technologies. Artificial intelligence basic technologies generally include technologies such as sensors, special artificial intelligence chips, cloud computing, distributed storage, big data processing, operation / interaction systems, mechatronics, etc. Artificial intelligence software technologies mainly include computer vision technology, speech processing technology, natural language processing technology, and machine learning / deep learning, autonomous driving, intelligent transportation, etc.
[0093] Further, artificial intelligence can be realized based on machine learning (ML), so that the obtained model can simulate or realize human learning behavior as much as possible, and thus realize the automation and intelligentization of data detection.
[0094] In the embodiments of the present application, please refer to Figure 1 , Figure 1 is a network interaction architecture diagram for data detection provided by the embodiments of the present application. Wherein, the computer device 101 can obtain the related information of the to-be-detected public broadcast service group account from the terminal device (such as the terminal device 102a, the terminal device 102b or the terminal device 102c), and detect the to-be-detected public broadcast service group account to obtain the account detection result of the to-be-detected public broadcast service group account. Wherein, the computer device 101 can perform data interaction with the terminal device, and can also obtain the training sample for model training from each terminal device when training the model. Optionally, the computer device 101 can respond to the account detection request for the to-be-detected public broadcast service group account, execute the scheme in the present application, and predict the account detection result of the to-be-detected public broadcast service group account. Wherein, the account detection request for the to-be-detected public broadcast service group account can be sent by any one terminal device, or can be generated by the computer device 101. For example, the terminal device 102a sends the account detection request for the to-be-detected public broadcast service group account to the computer device 101, the computer device 101 can respond to the account detection request, predict the account detection result of the to-be-detected public broadcast service group account, and send the account detection result to the terminal device 102a, and the terminal device 102a can determine whether to output the to-be-detected public broadcast service group account based on the account detection result.
[0095] Please refer to Figure 2 , Figure 2 is a data detection scene schematic diagram provided by the embodiments of the present application. As Figure 2As shown, the computer device can acquire to-be-detected account information, to-be-detected subject information and to-be-detected registration information of the to-be-detected public broadcast service group account 201. The to-be-detected account feature of the to-be-detected account information is extracted, the to-be-detected subject feature of the to-be-detected subject information is extracted, and the random registration feature of the to-be-detected registration information is generated. The first association relationship between the to-be-detected account information and the to-be-detected subject information is acquired, which can also be regarded as the first association relationship between the to-be-detected account feature and the to-be-detected subject feature; the second association relationship between the to-be-detected account information and the to-be-detected registration information is acquired, which can also be regarded as the second association relationship between the to-be-detected account feature and the random registration feature. The computer device can increase the related feature of the first association relationship in the to-be-detected subject feature, increase the related feature of the second association relationship in the random registration feature, and perform feature fusion on the to-be-detected subject feature and the random registration feature based on the first association relationship and the second association relationship, to obtain neighborhood association information, which refers to information associated with the to-be-detected account feature for enriching the to-be-detected account feature. The computer device can perform feature fusion on the to-be-detected account feature and the neighborhood association information to obtain account fusion features, so as to enrich the self feature of the to-be-detected account feature, effectively fuse the self feature and the relationship feature of the to-be-detected account feature, improve the richness of the semantic information of the account fusion feature, predict the account fusion feature, and obtain the account detection result of the to-be-detected public broadcast service group account.
[0096] It can be understood that the computer device mentioned in the embodiments of the present application includes but is not limited to a terminal device or a server. In other words, the computer device can be a server or a terminal device, or a system composed of a server and a terminal device. The terminal device mentioned above can be an electronic device, including but not limited to a mobile phone, a tablet computer, a desktop computer, a notebook computer, a palm computer, a vehicle-mounted device, an augmented reality / virtual reality (AR / VR) device, a head-mounted display, a smart television, a wearable device, a smart speaker, a digital camera, a camera, and other mobile internet devices (MIDs) with network access capability, or terminal devices in scenes such as trains, ships, and flights. Figure 1 As shown in FIG. 1, the terminal device can be a notebook computer (as shown in terminal device 102a or terminal device 102b) or a mobile phone (as shown in object device 102c), etc. Figure 1Part of the equipment is only exemplified. Among them, the above-mentioned server can be a stand-alone physical server, can also be a server cluster or distributed system composed of multiple physical servers, can also be a cloud server providing cloud service, cloud database, cloud computing, cloud function, cloud storage, network service, cloud communication, middleware service, domain name service, security service, vehicle-road cooperation, content delivery network (Content Delivery Network, CDN), and basic cloud computing services such as big data and artificial intelligence platform.
[0097] Optionally, the data involved in the embodiments of the present application can be stored in a computer device, or the data can be stored based on cloud storage technology or a blockchain network, which is not limited herein.
[0098] Further, please refer to Figure 3 , Figure 3 is a data detection method flowchart provided by the embodiments of the present application. As Figure 3 indicated, an example of a to-be-detected public broadcast service group account is described, in other words, in the method embodiment described in Figure 3 , the data detection process includes the following steps:
[0099] Step S301, obtaining to-be-detected account information, to-be-detected subject information and to-be-detected registration information of the to-be-detected public broadcast service group account.
[0100] In the embodiment of the present application, the computer device can acquire to-be-detected account information of a to-be-detected public broadcast service group account, to-be-detected subject information and to-be-detected registration information. The to-be-detected account information refers to account-related information of the to-be-detected public broadcast service group account, including but not limited to account name, account introduction, account broadcast content and account menu information, etc. The to-be-detected subject information refers to valid information provided when registering the to-be-detected public broadcast service group account, which can be determined based on the account type of the to-be-detected public broadcast service group account. For example, if the account type of the to-be-detected public broadcast service group account is an object type, the to-be-detected subject information is the object identity information associated with the to-be-detected public broadcast service group account. The object type refers to an account type registered by one or more objects, which can be a user, etc. For example, if the object type is a personal type corresponding to a user. If the account type of the to-be-detected public broadcast service group account is an enterprise type, the to-be-detected subject information can refer to the business license information of the enterprise associated with the to-be-detected public broadcast service group account, etc. If the account type of the to-be-detected public broadcast service group account is a government, media or other organization type, etc., the to-be-detected subject information can refer to the organization code information associated with the to-be-detected public broadcast service group account, etc. In other words, the to-be-detected subject information can be considered as the object legal proof information of the account registration object registering the to-be-detected public broadcast service group account, which can be but not limited to a user, an enterprise, a government, a media or other organizations. The to-be-detected registration information can be used to represent the object description information of the account registration object registering the to-be-detected public broadcast service group account, including but not limited to the unit identification code (uin) of the account registration object, the identity information and the social communication information, etc. The uin can represent the identity verification code of the account registration object, etc.
[0101] In step S302, to-be-detected account features of the to-be-detected account information are extracted, to-be-detected subject features of the to-be-detected subject information are extracted, and random registration features of the to-be-detected registration information are generated.
[0102] In the embodiments of the present application, the computer device can perform word segmentation processing on the to-be-detected account information to obtain m to-be-detected account word groups, perform vector mapping splicing processing on the m to-be-detected account word groups to obtain initial account features of the to-be-detected account information, where m is a positive integer. Specifically, the computer device can perform vector mapping conversion on the m to-be-detected account word groups respectively to obtain to-be-detected account word group vectors respectively corresponding to each to-be-detected account word group, and splice the m to-be-detected account word group vectors into the initial account features of the to-be-detected account information. The computer device can use any text vector conversion method to perform vector mapping conversion on the m to-be-detected account word groups, such as a method of converting through a text vector conversion model, a method of converting through a text mapping dictionary, or one-hot encoding, etc. For example, the computer device maps each to-be-detected account word group into a k-dimensional to-be-detected account word group vector, splices the m to-be-detected account word group vectors to obtain the initial account features of the to-be-detected account information, and the initial account features can be an m*k-dimensional matrix, where k is a positive integer. Further, the computer device can perform convolution conversion on the initial account features to directly obtain to-be-detected account features of the to-be-detected account information, where the width of a first convolution kernel for performing convolution conversion on the initial account features can be the dimension of each to-be-detected account word group vector, that is, k. Optionally, the computer device can perform convolution conversion on the initial account features to obtain convolution account features, further perform pooling processing on the convolution account features, and then obtain the to-be-detected account features of the to-be-detected account information. Further optionally, when obtaining the convolution account features, the computer device can perform pooling processing on the convolution account features to obtain pooling account features, and further perform full connection processing on the pooling account features to obtain the to-be-detected account features of the to-be-detected account information.
[0103] The computer device can perform word segmentation processing on the to-be-detected principal information to obtain m to-be-detected principal word groups, perform vector mapping splicing processing on the m to-be-detected principal word groups to obtain an initial principal feature of the to-be-detected principal information. Specifically, the computer device can perform vector mapping conversion on the m to-be-detected principal word groups respectively to obtain a to-be-detected principal word group vector corresponding to each to-be-detected principal word group respectively, and splice the m to-be-detected principal word group vectors into the initial principal feature of the to-be-detected principal information. The computer device can use any text vector conversion method to perform vector mapping conversion on the m to-be-detected principal word groups, for details, refer to the related description of performing vector mapping conversion on the m to-be-detected account word groups. For example, the computer device can map each to-be-detected principal word group into a k-dimensional to-be-detected principal word group vector, and splice the m to-be-detected principal word group vectors to obtain the initial principal feature of the to-be-detected principal information. The initial principal feature can be an m*k-dimensional matrix, and k is a positive integer. The dimension of the to-be-detected principal word group vector is the same as that of the to-be-detected account word group vector, which can improve the convenience of feature fusion and the like, therefore, the dimensions of the two can also be different. Further, the computer device can perform convolution conversion on the initial principal feature to obtain a to-be-detected principal feature of the to-be-detected principal information, wherein the width of the second convolution kernel for performing convolution conversion on the initial principal feature can be the dimension of each to-be-detected principal word group vector. Optionally, the computer device can perform convolution conversion on the initial principal feature to obtain a convolution principal feature, further perform pooling processing on the convolution principal feature to obtain the to-be-detected principal feature of the to-be-detected principal information. Further optionally, when obtaining the convolution principal feature, the computer device can perform pooling processing on the convolution principal feature to obtain a pooling principal feature, and further perform full connection processing on the pooling principal feature to obtain the to-be-detected principal feature of the to-be-detected principal information.
[0104] The computer device can input the to-be-detected registration information into a random initialization network, identify a target registration type of the to-be-detected registration information in the random initialization network, obtain a registration mapping matrix corresponding to the target registration type, and generate a random registration feature of the to-be-detected registration information based on the registration mapping matrix. The target registration type is determined by the type of the to-be-detected registration information, and can be any one of n registration types, where n is a positive integer. For example, when the to-be-detected registration information is a uin of an account registration object, the target registration type is a uin type; when the to-be-detected registration information is identity information, the target registration type is an identity type; and when the to-be-detected registration information is social communication information, the target registration type is a social communication type. The to-be-detected registration information can include one or at least two registration information, and the registration type of each registration information can be different.
[0105] Optionally, the computer device can perform parallel and synchronous processing on the "extracting the to-be-detected account feature of the to-be-detected account information", the "extracting the to-be-detected subject feature of the to-be-detected subject information", and the "generating the random registration feature of the to-be-detected registration information", or can perform serial processing, which is not limited herein, and the execution order of the three can be adjusted according to actual needs.
[0106] Optionally, the computer device can process the to-be-detected account information, the to-be-detected subject information, and the to-be-detected registration information through the account detection model. For details, please refer to Figure 4 , Figure 4 is a model architecture diagram provided by an embodiment of the present application, as shown in Figure 4 , the computer device can input the to-be-detected account information 4011 into the text detection network for feature extraction to obtain the to-be-detected account feature 4021 of the to-be-detected account information 4011; input the to-be-detected subject information 4012 into the text detection network for feature extraction to obtain the to-be-detected subject feature 4022 of the to-be-detected subject information 4012. Input the to-be-detected registration information into the random initialization network to generate the random registration feature of the to-be-detected registration information. For example, it is assumed that the to-be-detected registration information includes uin registration information 4013, identity registration information 4014, and social registration information 4015, etc., the computer device inputs the uin registration information 4013 into the random initialization network to generate the random registration feature 4023 of the uin registration information 4013, which can be denoted as uin registration feature 4023; inputs the identity registration information 4014 into the random initialization network to generate the random registration feature 4024 of the identity registration information 4014, which can be denoted as identity registration feature 4024; and inputs the social registration information 4015 into the random initialization network to generate the random registration feature 4025 of the social registration information 4015, which can be denoted as social registration feature 4025.
[0107] In step S303, the first association relationship between the to-be-detected account information and the to-be-detected subject information is obtained, and the second association relationship between the to-be-detected account information and the to-be-detected registration information is obtained.
[0108] In the embodiment of the present application, the computer device can obtain the association relationship between the to-be-detected account information and the to-be-detected subject information and the to-be-detected registration information, respectively. Specifically, there can be at least (n+2) kinds of association relationships in the present application, including the first association relationship between the to-be-detected account information and the to-be-detected subject information, the second association relationship between the to-be-detected account information and the to-be-detected registration information of n kinds of registration types, and the association relationship between the to-be-detected public broadcast service group account and the adjacent service group account, etc. For example, please refer to Figure 5 , Figure 5 is a node association graph diagram provided by an embodiment of the present application, as shown inFigure 5 As shown, it is assumed that the to-be-detected registration information includes n kinds of registration information, including but not limited to social communication type social registration information 502, identity identification type identity registration information 504, and uin type uin registration information 505, etc. Among them, the to-be-detected account information 501 is associated with the n kinds of registration information, including the second association relationship between the to-be-detected account information 501 and the social registration information 502, the second association relationship between the to-be-detected account information 501 and the identity registration information 504, and the second association relationship between the to-be-detected account information 501 and the uin registration information 505, etc. Figure 5 For example, it is assumed that n is 3, there is a first association relationship ① between the to-be-detected account information 501 and the to-be-detected subject information 503; there is a second association relationship between the to-be-detected account information 501 and the n kinds of registration information, including the second association relationship ② between the to-be-detected account information 501 and the social registration information 502, the second association relationship ③ between the to-be-detected account information 501 and the identity registration information 504, and the second association relationship ④ between the to-be-detected account information 501 and the uin registration information 505, etc.; there is an association relationship (which can be denoted as a third association relationship) between the to-be-detected public broadcast service group account and the adjacent service group account. Among them, the third association relationship can refer to the association relationship between the public broadcast service group account registered by the uin registration information 505 and the uin registration information 505, that is, the adjacent service group account can refer to the account with the same registration information as the to-be-detected public broadcast service group account, for example, the uin registration information 505 registers the adjacent service group account in addition to the to-be-detected public broadcast service group account, and the uin registration information 505 can exist between the third association relationship of the adjacent account information of the adjacent service group account registered by it, such as the third association relationship ⑤ between the uin registration information 505 and the adjacent account information 506, the third association relationship ⑤ between the uin registration information 505 and the adjacent account information 507, and the third association relationship ⑤ between the uin registration information 505 and the adjacent account information 508, etc.
[0109] That is, in the case of Figure 5Five types of information nodes and five association relationships are exemplified in the figure. "O" represents an account type information node, such as the to-be-detected account information 501, the adjacent account information 506, the adjacent account information 507, the adjacent account information 508, and the like. "☆" is used to represent a subject type information node, such as the to-be-detected subject information 503. "◇" represents a social communication type information node, such as the social registration information 502. "△" is used to represent a uin type information node, such as the uin registration information 505. "□" is used to represent an identity identification type information node, such as the identity registration information 504, and the like. "①" refers to a first association relationship between the to-be-detected account information 501 and the to-be-detected subject information 503. "②" refers to a second association relationship between the to-be-detected account information 501 and the social registration information 502. "③" refers to a second association relationship between the to-be-detected account information 501 and the identity registration information 504. "④" refers to a second association relationship between the to-be-detected account information 501 and the uin registration information 505. "⑤" refers to a third association relationship between the uin registration information 505 and the adjacent account information, and the like. Optionally, the node association graph can be updated based on the need for information node update or association relationship update, and is not limited to the information nodes or association relationships shown in the figure. The information nodes in the node association graph refer to information to be collected, and the edges are used to represent the association relationships. Figure 5 The information nodes in the node association graph refer to information to be collected, and the edges are used to represent the association relationships.
[0110] Further, the computer device can determine the information node corresponding to the to-be-detected account information, the information node corresponding to the to-be-detected subject information, and the information node corresponding to the to-be-detected registration information based on the node association graph, obtain a first association relationship between the to-be-detected account information and the to-be-detected subject information from the node association graph, and obtain a second association relationship between the to-be-detected account information and the to-be-detected registration information. The node association graph can be a graph as shown in Figure 5 The node association graph can be a graph as shown in
[0111] Optionally, as shown in Figure 4 The computer device can obtain an association relationship 40a and introduce the association relationship 40a into the account detection model. The association relationship 40a includes a first association relationship between the to-be-detected account information 4011 and the to-be-detected subject information 4012, a second association relationship between the to-be-detected account information 4011 and the to-be-detected registration information, and the like. The to-be-detected registration information includes the identity registration information 4014. Figure 4 The to-be-detected registration information includes the identity registration information 4014.
[0112] In step S304, the to-be-detected subject feature and the random registration feature are fused based on the first association relationship and the second association relationship to obtain neighborhood association information.
[0113] In the embodiment of the present application, the computer device can perform feature fusion on the to-be-detected account feature and the to-be-detected principal feature based on the first association relationship to obtain a first feature correlation degree corresponding to the first association relationship; perform feature fusion on the to-be-detected account feature and the random registration feature based on the second association relationship to obtain a second feature correlation degree corresponding to the second association relationship; and perform weighted summation on the to-be-detected principal feature and the random registration feature through the first feature correlation degree and the second feature correlation degree to obtain the neighborhood association information.
[0114] In the process of performing feature fusion on the to-be-detected account feature and the to-be-detected principal feature based on the first association relationship to obtain the first feature correlation degree corresponding to the first association relationship, the computer device can map the to-be-detected account feature into a to-be-detected account linear feature, obtain a first information type corresponding to the to-be-detected principal information, map the to-be-detected principal feature into a to-be-detected principal vector based on a first node mapping matrix corresponding to the first information type, perform feature fusion on the to-be-detected account linear feature and the to-be-detected principal vector based on the first association relationship, and obtain the first feature correlation degree of the first association relationship. Specifically, the computer device can perform continuous multiplication processing on the to-be-detected account linear feature and the to-be-detected principal vector based on the first association relationship to obtain a first initial correlation degree, obtain a first tuple importance of a first association tuple composed of the to-be-detected account information, the to-be-detected principal information, and the first association relationship, and perform optimization processing on the first initial correlation degree based on the first tuple importance to obtain the first feature correlation degree of the first association relationship.
[0115] Similarly, in the process of performing feature fusion on the to-be-detected account feature and the random registration feature based on the second association relationship to obtain the second feature correlation degree corresponding to the second association relationship, the computer device can map the to-be-detected account feature into a to-be-detected account linear feature, obtain a second information type corresponding to the to-be-detected registration information, map the random registration feature into a random registration vector based on a second node mapping matrix corresponding to the second information type, perform feature fusion on the to-be-detected account linear feature and the random registration vector based on the second association relationship, and obtain the second feature correlation degree of the second association relationship. Specifically, the computer device can perform continuous multiplication processing on the to-be-detected account linear feature and the random registration vector based on the second association relationship to obtain a second initial correlation degree, obtain a second tuple importance of a second association tuple composed of the to-be-detected account information, the to-be-detected registration information, and the second association relationship, and perform optimization processing on the second initial correlation degree based on the second tuple importance to obtain the second feature correlation degree of the second association relationship.
[0116] Furthermore, the computer device can convert the first feature correlation degree into a first correlation weight and the second feature correlation degree into a second correlation weight. Based on the first and second correlation weights, a weighted sum is performed on the features of the subject to be detected and the randomly registered features to obtain neighborhood correlation information. Specifically, the first and second feature correlation degrees are normalized to obtain the first correlation weight corresponding to the first feature correlation degree and the second correlation weight corresponding to the second feature correlation degree; based on the first and second correlation weights, a weighted sum is performed on the features of the subject to be detected and the randomly registered features to obtain neighborhood correlation information.
[0117] Alternatively, the computer device can perform feature transfer on the features of the subject to be detected, obtaining the subject-transferred features corresponding to the subject features; perform feature transfer on the random registration features, obtaining the registration-transferred features corresponding to the random registration features; and then perform a weighted sum of the subject-transferred features and the registration-transferred features using the first feature correlation degree and the second feature correlation degree to obtain neighborhood association information. Specifically, the first association type of the first association relationship corresponding to the subject features to be detected is obtained, and the first information weight matrix corresponding to the first association type is obtained. The association information of the subject features to be detected is fused using the first information weight matrix to obtain the subject-transferred features corresponding to the subject features to be detected; the second association type of the second association relationship corresponding to the random registration features is obtained, and the second information weight matrix corresponding to the second association type is obtained. The association information of the random registration features is fused using the second information weight matrix to obtain the registration-transferred features corresponding to the random registration features; and the subject-transferred features and the registration-transferred features are weighted summed using the first feature correlation degree and the second feature correlation degree to obtain neighborhood association information.
[0118] Alternatively, the computer device can convert the first feature correlation degree into a first correlation weight, convert the second feature correlation degree into a second correlation weight, perform feature transfer on the subject feature to be detected to obtain the subject transfer feature corresponding to the subject feature to be detected, perform feature transfer on the random registration feature to obtain the registration transfer feature corresponding to the random registration feature, and perform weighted summation on the subject transfer feature and the registration transfer feature based on the first correlation weight and the second correlation weight to obtain the neighborhood correlation information.
[0119] Optional, such as Figure 4 As shown, in the relation weight determination network 40b, the computer device can perform feature fusion based on the first association relationship between the account feature 4021 to be detected and the subject feature 4022 to be detected, to obtain the first association weight 4041 corresponding to the first association relationship; and perform feature fusion based on the second association relationship between the account feature 4021 to be detected and the registration information to be detected, to obtain the second association weight corresponding to the second association relationship, as follows. Figure 4In the example, based on the second association relationship, the feature fusion is performed on the to-be-detected account feature 4021 and the identity registration feature 4024, to obtain a second association weight 4042 corresponding to the second association relationship between the identity registration information 4014 and the to-be-detected account information 4011. In the information transmission network, the feature transmission is performed on the to-be-detected principal feature 4022 to obtain a principal transmission feature 4043 corresponding to the to-be-detected principal feature 4022, and the feature transmission is performed on the random registration feature to obtain a registration transmission feature corresponding to the random registration feature. Specifically, the computer device inputs the to-be-detected account feature 4021 into a first mapping network in the relationship weight determination network 40b, performs linear mapping on the to-be-detected account feature 4021 to obtain a to-be-detected account linear feature 4031; obtains a first information type corresponding to the to-be-detected principal information 4012, and in the second mapping network, based on a first node mapping matrix corresponding to the first information type, maps the to-be-detected principal feature 4022 into a to-be-detected principal vector 4032; obtains a second information type corresponding to the to-be-detected registration information, and in the second mapping network, based on a second node mapping matrix corresponding to the second information type, maps the random registration feature into a random registration vector. Taking the identity registration information as an example, obtaining a second information type corresponding to the identity registration information 4014, in the second mapping network, based on a second node mapping matrix corresponding to the second information type, mapping the identity registration feature 4024 into an identity registration vector 4033. Wherein, the mapping process in the first mapping network can be referred to formula ①:
[0120] K(s)=K_Linear T(s) (H (l-1) [s]) ①
[0121] Wherein, the K_Linear is used to represent the first mapping network, s is used to represent the source information node, which refers to the feature corresponding to the information node input into the first mapping network, T() is used to represent the information node mapping function, T(s) refers to mapping the source information node into the linear feature, for example, the s is the to-be-detected account information, K_Linear T(s) refers to mapping the to-be-detected account feature 4021 of the to-be-detected account information into the to-be-detected account linear feature 4031 in the first mapping network. H (l-1) [s] refers to the feature of the source information node, for example, if the s is the to-be-detected account information, then H (l-1) [s] refers to the to-be-detected account feature 4021.
[0122] Wherein, the mapping process of the second mapping network can be referred to formula ②:
[0123] Q(t)=Q_Linear T(t) (H (l-1) [t]) ②
[0124] wherein the Q_Linear is used for representing the second mapping network, t is used for representing the target information node, and refers to an information node for adding a feature in the source information node, which can be considered as an information node corresponding to the feature input into the second mapping network. The T() is used for representing an information node mapping function, and T(t) refers to mapping the target information node into a linear feature. For example, the t is the to-be-detected subject information, and the Q_Linear T(t) refers to mapping the to-be-detected subject feature 4022 of the to-be-detected subject information into the to-be-detected subject vector 4032 in the second mapping network; the t is the to-be-detected registration information, and the Q_Linear T(t) refers to mapping the random registration feature of the to-be-detected registration information into the random registration vector in the second mapping network. (l-1) The [t] refers to a feature of the target information node. For example, if the t is the to-be-detected subject information, then the H (l-1) The [t] refers to the to-be-detected subject feature 4022. For example, if the t is the to-be-detected registration information, then the H (l-1) The [t] refers to the random registration feature.
[0125] Further, in the attention layer of the relationship weight determination network 40b, the linear feature of the source information node and the linear feature of the target information node are multiplied based on the association relationship to obtain an initial association degree, the tuple importance of the association tuple composed of the source information node, the target information node and the association relationship is obtained, and the initial association degree is optimized based on the tuple importance to obtain the feature association degree of the association relationship.
[0126] wherein the feature association degree can be obtained according to the formula ③:
[0127]
[0128] As shown in the formula ③, the ATT_head(s, e, t) refers to the feature association degree, the K(s) refers to the linear feature of the source information node, the Q(t) refers to the linear feature of the target information node, and the e refers to the association relationship between the source information node and the target information node. is used for representing the weight matrix of the association relationship in the attention layer. The <T(s), φ(e), T(t)> is used for representing the association tuple, the μ is used for representing the tuple importance, and the d is the dimension of the feature of the source information node, that is, the H (l-1)The dimension of [s]. As shown in formula ③, the source information node s can be the account information to be detected, and the target information node t can be the subject information to be detected or the registration information to be detected, etc. For example, if the source information node s is the account information to be detected and the target information node t is the subject information to be detected, then e represents the first association between the account information to be detected and the subject information to be detected. Based on formula ③, the linear feature 4031 of the account to be detected and the subject vector 4032 to be detected are multiplied together to obtain the first initial association degree, that is, at this time... Let μ represent the first initial correlation degree, and let e represent the importance of the first tuple in the first correlation tuple. Based on the importance of the first tuple, the first initial correlation degree is optimized to obtain the first feature correlation degree of the first correlation relationship. For example, if the source information node s is the account information to be detected and the target information node t is the registration information to be detected, then e represents the second correlation relationship between the account information to be detected and the registration information to be detected. Based on formula ③, the linear feature 4031 of the account to be detected is multiplied by the random registration vector to obtain the second initial correlation degree, i.e., at this point... Let μ represent the second initial correlation degree, and let μ represent the importance of the second tuple in the second correlation tuple. The second initial correlation degree is optimized based on the importance of the second tuple to obtain the second feature correlation degree of the second correlation relationship.
[0129] Furthermore, the correlation degrees of the first and second features are normalized to obtain the first correlation weight 4041 corresponding to the first feature correlation degree and the second correlation weight 4042 corresponding to the second feature correlation degree. The determination process of these correlation weights can be found in formula ④.
[0130] Attention(s,e,t)=Softmax(|| i∈[1,h] ATT_head i (s,e,t)) ④
[0131] In formula ④, h represents the number of feature correlations, and h is a positive integer. For example, if the registration information to be detected includes uin registration information, identity registration information, and social registration information, then h is 4, and ATT_head i (s,e,t) represents the i-th feature correlation among h feature correlations, where i is a positive integer less than or equal to h. Here, the h feature correlations include the first feature correlation of the first correlation, the second feature correlation of the second correlation between uin registration information and the account information to be detected, the second feature correlation of the second correlation between identity registration information and the account information to be detected, and the second feature correlation of the second correlation between social registration information and the account information to be detected, etc. Softmax is used to represent the normalization process.
[0132] Further, the computer device inputs the to-be-detected principal feature 4022 into an information mapping network of the information transmission network 4044, linearly maps the to-be-detected principal feature 4022 in the information mapping network, and obtains a principal information vector 4034 of the to-be-detected principal feature 4022; linearly maps the random registration feature, and obtains a registration information vector of the random registration feature, for example, linearly maps the identity registration feature 4024, and obtains an identity information vector 4035 of the identity registration feature 4024, and the like. Wherein, the mapping process in the information mapping network can be recorded as shown in formula 5:
[0133]
[0134] In formula 5, M_Linear is used to represent the information mapping network, t is used to represent the target information node, which refers to the information node for adding the feature in the source information node, t here can be the to-be-detected principal information or the to-be-detected registration information. Wherein, when t is the to-be-detected principal information, is used to represent the first information weight matrix corresponding to the first association relationship, e represents the first association relationship between the to-be-detected principal information and the to-be-detected account information; when t is the to-be-detected registration information, is used to represent the second information weight matrix corresponding to the second association relationship, e represents the second association relationship between the to-be-detected registration information and the to-be-detected account information.
[0135] In the information transmission layer, the first information weight matrix corresponding to the first association type is obtained, the principal information vector 4034 is subjected to association information fusion through the first information weight matrix, and the principal transmission feature 4043 corresponding to the principal information vector 4034 is obtained. In the information transmission layer, the second information weight matrix corresponding to the second association type of the second association relationship is obtained, the registration information vector is subjected to association information fusion through the second information weight matrix, and the registration transmission feature corresponding to the registration information vector is obtained; for example, the identity information vector 4035 is subjected to association information fusion based on the second information weight matrix corresponding to the second association relationship between the identity registration information 4014 and the to-be-detected account information 4011, and the registration transmission feature 4044 corresponding to the identity information vector 4035 is obtained.
[0136] Further, the computer device can weight and sum the principal transmission feature and the registration transmission feature based on the first association weight and the second association weight, and obtain the neighborhood association information. As shown in formula 6, Figure 4 weight the principal transmission feature 4043 based on the first association weight 4041, weight the registration transmission feature 4044 based on the second association weight 4042, and then sum the weighted features to obtain the neighborhood association information. Wherein, Figure 4 in formula 6, Used to represent the weighted processing procedure. This is used to represent the summation process. The weighted summation process can be seen in formula ⑥:
[0137]
[0138] In formula ⑥, This indicates neighborhood association information. Used to represent the summation process.
[0139] Step S305: The features of the account to be detected are fused with the neighborhood association information to obtain the account fusion features. The account fusion features are then predicted to obtain the account detection results of the public broadcast service group account to be detected.
[0140] In this embodiment of the application, the computer device can fuse the features of the account to be detected with the neighborhood association information to obtain the account fusion feature. The process of determining the account fusion feature can be referred to as formula ⑦:
[0141]
[0142] Optional, such as Figure 4 As shown, the computer device can input neighborhood association information into the third mapping network for mapping to obtain neighborhood mapping information, making the neighborhood mapping information match the distribution of the features of the account to be detected. Feature fusion is performed between the features of the account to be detected and the neighborhood mapping information to obtain account fusion feature 405. In the classification network, the account fusion feature 405 is predicted to obtain the account detection result 406 for the public broadcast service group account to be detected. In formula ⑦, H... (l-1) [s] represents the features of the account to be detected, A_Linear represents the third mapping network, and σ represents the neighborhood association information in the third mapping network. The mapping function that performs the mapping. H (l) [s] is used to represent account fusion characteristics.
[0143] Furthermore, the computer equipment can predict the account fusion features to obtain the normal probability corresponding to the normal result label of the account and the abnormal probability corresponding to the abnormal result label of the account. If the normal probability is greater than the abnormal probability, the account detection result of the public broadcasting service group account to be detected is determined to be a normal result. If the normal probability is less than or equal to the abnormal probability, the account detection result of the public broadcasting service group account to be detected is determined to be an abnormal result.
[0144] Further, if the account detection result of the to-be-detected public broadcast service group account is the account normal result, the to-be-detected public broadcast service group account is output; if the account detection result of the to-be-detected public broadcast service group account is the account abnormal result, the to-be-detected public broadcast service group account is deleted, and an account abnormal message is sent to a service device associated with the to-be-detected public broadcast service group account based on to-be-detected registration information.
[0145] In a possible application scenario, refer to Figure 6 , Figure 6 is a data detection scene diagram provided by an embodiment of the present application, as Figure 6 shown, the computer device can respond to the object search request in the function page 601, obtain p candidate public broadcast service group accounts associated with the target object requested by the object search request, including public broadcast service group account 1 to public broadcast service group account p, p is a positive integer. Wherein, without performing the present application, the computer device can output p candidate public broadcast service group accounts in the virtual display page 602, through the present application, p candidate public broadcast service group accounts are determined as to-be-detected public broadcast service group accounts in turn, the account detection result of each candidate public broadcast service group account is obtained, from p candidate public broadcast service group accounts, the abnormal group account 6021 with the account detection result of the account abnormal result is obtained, the abnormal group account 6021 belongs to p candidate public broadcast service group accounts, the computer device can delete the abnormal group account 6021 from p candidate public broadcast service group accounts, and output the normal group account in the search result display page 603, the normal group account refers to the candidate public broadcast service group account except the abnormal group account 6021 in p candidate public broadcast service group accounts.
[0146] In the embodiment of the present application, the computer device can obtain to-be-detected account information, to-be-detected subject information and to-be-detected registration information of a to-be-detected public broadcast service group account, extract to-be-detected account features of the to-be-detected account information, extract to-be-detected subject features of the to-be-detected subject information, and generate random registration features of the to-be-detected registration information; obtain a first association relationship between the to-be-detected account information and the to-be-detected subject information, and obtain a second association relationship between the to-be-detected account information and the to-be-detected registration information; based on the first association relationship and the second association relationship, perform feature fusion on the to-be-detected subject features and the random registration features to obtain neighborhood association information; perform feature fusion on the to-be-detected account features and the neighborhood association information to obtain account fusion features, predict the account fusion features, and obtain an account detection result of the to-be-detected public broadcast service group account. By performing association analysis on the to-be-detected public broadcast service group account, other features are added based on the basic information (i.e., the to-be-detected account information) of the to-be-detected public broadcast service group account in the association relationship, so that the to-be-detected account information can be enriched, and the association relationship between different nodes can be considered, and other information features are added in the to-be-detected account information, thereby improving the accuracy of data detection.
[0147] Further, please refer to Figure 7 , Figure 7 is a model training method provided by the embodiment of the present application. As shown in Figure 7 , the method comprises the following steps:
[0148] Step S701, obtaining a public broadcast service group account sample, obtaining account information sample, subject information sample and registration information sample of the public broadcast service group account sample, and obtaining account sample label of the public broadcast service group account sample.
[0149] In this embodiment, the computer device can acquire public broadcast service group account samples. Optionally, it can acquire at least two public broadcast service group accounts and the number of times each public broadcast service group account is exposed, and public broadcast service group accounts with an exposure count greater than or equal to a sample acquisition threshold are identified as public broadcast service group account samples. Alternatively, it can randomly acquire at least two public broadcast service group accounts and identify the acquired at least two public broadcast service group accounts as public broadcast service group account samples, etc., without limitation. The account sample label includes a normal sample label and a abnormal sample label, used to indicate whether the corresponding public broadcast service group account sample is normal. For example, if the account sample label of a public broadcast service group account sample is a normal sample label, then the public broadcast service group account sample is a normal account and can be considered a positive sample; if the account sample label of a public broadcast service group account sample is an abnormal sample label, then the public broadcast service group account sample is an abnormal account and can be considered a negative sample.
[0150] Step S702: Input the account information sample, the subject information sample, and the registration information sample into the initial account detection model.
[0151] In this embodiment of the application, the model architecture of the initial account detection model can be found in [reference needed]. Figure 4 As shown.
[0152] Step S703: Extract sample account features of account information samples from the text detection network of the initial account detection model.
[0153] In the embodiments of this application, this step can be referred to Figure 3 As shown in step S302, please refer to the relevant description of the characteristics of the account to be detected for extracting the information of the account to be detected.
[0154] Step S704: Extract the subject features of the subject information sample from the text detection network.
[0155] In the embodiments of this application, this step can be referred to Figure 3 As shown in step S302, please refer to the relevant description of the features of the subject to be detected for extracting the subject information.
[0156] Step S705: Randomly initialize the registration information samples in the random initialization network of the initial account detection model to obtain the sample registration features of the registration information samples.
[0157] In the embodiments of this application, this step can be referred to Figure 3 As shown in step S302, please refer to the relevant description of the random registration features for generating the registration information to be detected.
[0158] In step S706, a first sample correlation between the account information sample and the subject information sample is obtained, and a second sample correlation between the account information sample and the registration information sample is obtained.
[0159] In the embodiments of the present application, the process can refer to the specific description shown in step S303 in the foregoing method. Figure 3
[0160] In step S707, sample subject features and sample registration features are fused based on the first sample correlation and the second sample correlation, to obtain sample neighborhood correlation information.
[0161] In the embodiments of the present application, the process can refer to the specific description shown in step S304 in the foregoing method. Figure 3
[0162] In step S708, sample account features and sample neighborhood correlation information are fused to obtain sample fusion features, and the sample fusion features are predicted based on a classification network of an initial account detection model, to obtain a sample detection result of the public broadcast service group account sample.
[0163] In the embodiments of the present application, the process can refer to the specific description shown in step S305 in the foregoing method. Figure 3
[0164] In step S709, the initial account detection model is adjusted based on the account sample label and the sample detection result, to obtain an account detection model.
[0165] In the embodiments of the present application, the computer device can generate a model loss function based on the account sample label and the sample detection result, and adjust the initial account detection model based on the model loss function, to obtain the account detection model.
[0166] Optionally, the computer device can acquire account promotion data and an account menu of the public broadcast service group account sample, acquire an account relationship heterogeneous graph associated with the public broadcast service group account sample according to the account promotion data and the account menu, acquire adjacent account information associated with the registration information sample based on the account relationship heterogeneous graph, and perform parameter optimization on the account detection model based on the adjacent account information to obtain an optimized account detection model. Specifically, the adjacent account information is input into the account detection model as a target information node for parameter optimization, that is, a third sample association relationship between the account information sample and the adjacent account information is acquired, and feature fusion is performed on a sample main body feature and an adjacent account feature of the adjacent account information based on the third sample association relationship to obtain sample adjacent account association information. Feature fusion is performed on the sample adjacent account association information and the sample account feature to obtain adjacent fusion features, the adjacent fusion features are predicted to obtain an optimized detection result, and parameter optimization is performed on the account detection model based on the account sample label and the optimized detection result to obtain the optimized account detection model.
[0167] Optionally, in steps S702 to S709, the adjacent account information can be directly used as account information associated with the account information sample in a third sample association relationship, and the initial account detection model can be directly adjusted based on the account information sample, the main body information sample, the registration information sample, and the adjacent account information to generate the account detection model. At this time, the information node type of the adjacent account information is the same as the information node type of the account information sample, and it can be considered that the feature of the adjacent account information is linearly mapped based on the first mapping network. The process of the third sample association weight and the adjacent transmission feature corresponding to other adjacent account information can be referred to the process of acquiring the first association weight and the main body transmission feature in Figure 3 .
[0168] Through Figure 7 , the parameters of each network in the initial account detection model can be adjusted, so that the well-trained randomly initialized network can randomly initialize the features of the account itself when extracting the features of the to-be-detected registration information, and the association weight corresponding to the target information node is determined based on the same mapping network (the second mapping network) for feature mapping conversion, so that each parameter in the model can fully learn the commonality and specificity between different information nodes in the account relationship heterogeneous graph, thereby improving the accuracy of model prediction.
[0169] The computer device performing the method of Figure 3 and the computer device performing the method of Figure 7 may be the same device or different devices.
[0170] Further, please refer to Figure 8 ,Figure 8 is a schematic diagram of a data detection device provided by an embodiment of the present application. The data detection device can be a computer program (including program code, etc.) running in a computer device, for example, the data detection device can be an application software; the device can be used to execute corresponding steps in the method provided by the embodiments of the present application. As shown in the figure, the data detection device 800 can be used for the computer device in the corresponding embodiments, specifically, the device can include: an information acquisition module 11, an account feature generation module 12, a principal feature generation module 13, a registration feature generation module 14, a relationship acquisition module 15, an information generation module 16, an information fusion module 17, and an account prediction module 18. Figure 8 Figure 3 The information acquisition module 11 is configured to acquire account information to be detected, principal information to be detected, and registration information to be detected of a public broadcast service group account to be detected.
[0171] The information acquisition module 11 is configured to acquire account information to be detected, principal information to be detected, and registration information to be detected of a public broadcast service group account to be detected.
[0172] The account feature generation module 12 is configured to extract account features to be detected of the account information to be detected.
[0173] The principal feature generation module 13 is configured to extract principal features to be detected of the principal information to be detected.
[0174] The registration feature generation module 14 is configured to generate random registration features of the registration information to be detected.
[0175] The relationship acquisition module 15 is configured to acquire a first association relationship between the account information to be detected and the principal information to be detected, and acquire a second association relationship between the account information to be detected and the registration information to be detected.
[0176] The information generation module 16 is configured to perform feature fusion on the principal features to be detected and the random registration features based on the first association relationship and the second association relationship, to obtain neighborhood association information.
[0177] The information fusion module 17 is configured to perform feature fusion on the account features to be detected and the neighborhood association information, to obtain account fusion features.
[0178] The account prediction module 18 is configured to predict the account fusion features, to obtain an account detection result of the public broadcast service group account to be detected.
[0179] The account feature generation module 12 includes:
[0180] The first feature extraction unit 121 is configured to perform word segmentation processing on the account information to be detected, to obtain m account word groups to be detected, perform vector mapping splicing processing on the m account word groups to be detected, and obtain initial account features of the account information to be detected; m is a positive integer.
[0181] The first feature conversion unit 122 is configured to perform convolutional conversion on the initial account feature to obtain an account feature to be detected of the account information to be detected.
[0182] The subject feature generation module 13 comprises:
[0183] The second feature extraction unit 131 is configured to perform word segmentation processing on the subject information to be detected to obtain m subject word groups to be detected, and perform vector mapping and splicing processing on the m subject word groups to be detected to obtain an initial subject feature of the subject information to be detected.
[0184] The second feature conversion unit 132 is configured to perform convolutional conversion on the initial subject feature to obtain a subject feature to be detected of the subject information to be detected.
[0185] The registration feature generation module 14 comprises:
[0186] The type acquisition unit 141 is configured to input the registration information to be detected into a random initialization network, and identify a target registration type of the registration information to be detected in the random initialization network.
[0187] The feature mapping unit 142 is configured to acquire a registration mapping matrix corresponding to the target registration type, and generate a random registration feature of the registration information to be detected based on the registration mapping matrix.
[0188] The information generation module 16 comprises:
[0189] The first association determination unit 161 is configured to perform feature fusion on the account feature to be detected and the subject feature to be detected based on the first association relationship to obtain a first feature association degree corresponding to the first association relationship.
[0190] The second association determination unit 162 is configured to perform feature fusion on the account feature to be detected and the random registration feature based on the second association relationship to obtain a second feature association degree corresponding to the second association relationship.
[0191] The feature weighting unit 163 is configured to perform weighted summation on the subject feature to be detected and the random registration feature by using the first feature association degree and the second feature association degree to obtain the neighborhood association information.
[0192] The first association determination unit 161 comprises:
[0193] The linear mapping sub-unit 1611 is configured to map the account feature to be detected into an account linear feature to be detected, acquire a first information type corresponding to the subject information to be detected, and map the subject feature to be detected into a subject vector based on a first node mapping matrix corresponding to the first information type.
[0194] The feature fusion subunit 1612 is configured to perform feature fusion on the linear feature of the to-be-detected account and the to-be-detected principal vector based on the first correlation relationship, to obtain a first feature correlation degree of the first correlation relationship.
[0195] The feature fusion subunit 1612 includes:
[0196] The feature multiplication subunit 161a is configured to perform multiplication processing on the linear feature of the to-be-detected account and the to-be-detected principal vector based on the first correlation relationship, to obtain a first initial correlation degree.
[0197] The importance acquisition subunit 161b is configured to acquire a first tuple importance of a first correlation tuple composed of the to-be-detected account information, the to-be-detected principal information, and the first correlation relationship.
[0198] The correlation optimization subunit 161c is configured to perform optimization processing on the first initial correlation degree based on the first tuple importance, to obtain the first feature correlation degree of the first correlation relationship.
[0199] The feature weighting unit 163 includes:
[0200] The correlation normalization subunit 1631 is configured to perform normalization processing on the first feature correlation degree and the second feature correlation degree, to obtain a first correlation weight corresponding to the first feature correlation degree and a second correlation weight corresponding to the second feature correlation degree.
[0201] The information weighting subunit 1632 is configured to perform weighted summation on the principal feature and the random registration feature based on the first correlation weight and the second correlation weight, to obtain the neighborhood correlation information.
[0202] The feature weighting unit 163 includes:
[0203] The principal transmission subunit 1633 is configured to acquire a first correlation type of the first correlation relationship corresponding to the principal feature, acquire a first information weight matrix corresponding to the first correlation type, and perform correlation information fusion on the principal feature through the first information weight matrix, to obtain a principal transmission feature corresponding to the principal feature.
[0204] The registration transmission subunit 1634 is configured to acquire a second correlation type of the second correlation relationship corresponding to the random registration feature, acquire a second information weight matrix corresponding to the second correlation type, and perform correlation information fusion on the random registration feature through the second information weight matrix, to obtain a registration transmission feature corresponding to the random registration feature.
[0205] The transmission weighting subunit 1635 is configured to perform weighted summation on the principal transmission feature and the registration transmission feature through the first feature correlation degree and the second feature correlation degree, to obtain the neighborhood correlation information.
[0206] The account prediction module 18 comprises:
[0207] The probability prediction unit 181 is configured to predict the account fusion features to obtain a normal probability corresponding to a normal result label of the account and an abnormal probability corresponding to an abnormal result label of the account.
[0208] The normal determination unit 182 is configured to determine that the account detection result of the to-be-detected public broadcast service group account is the account normal result if the normal probability is greater than the abnormal probability.
[0209] The abnormal determination unit 183 is configured to determine that the account detection result of the to-be-detected public broadcast service group account is the account abnormal result if the normal probability is less than or equal to the abnormal probability.
[0210] The device 800 further comprises:
[0211] The account output module 19 is configured to output the to-be-detected public broadcast service group account if the account detection result of the to-be-detected public broadcast service group account is the account normal result.
[0212] The abnormal feedback module 20 is configured to delete the to-be-detected public broadcast service group account if the account detection result of the to-be-detected public broadcast service group account is the account abnormal result, and send an account abnormal message to a service device associated with the to-be-detected public broadcast service group account based on the to-be-detected registration information.
[0213] The data detection device provided by the embodiment of the present application can obtain to-be-detected account information, to-be-detected subject information and to-be-detected registration information of a to-be-detected public broadcast service group account, extract to-be-detected account features of the to-be-detected account information, extract to-be-detected subject features of the to-be-detected subject information, and generate random registration features of the to-be-detected registration information; obtain a first association relationship between the to-be-detected account information and the to-be-detected subject information, and obtain a second association relationship between the to-be-detected account information and the to-be-detected registration information; based on the first association relationship and the second association relationship, perform feature fusion on the to-be-detected subject features and the random registration features to obtain neighborhood association information; perform feature fusion on the to-be-detected account features and the neighborhood association information to obtain account fusion features, and predict the account fusion features to obtain an account detection result of the to-be-detected public broadcast service group account. Through the association analysis on the to-be-detected public broadcast service group account, other features are added based on the basic information (i.e., the to-be-detected account information) of the to-be-detected public broadcast service group account in terms of the association relationship, so that the to-be-detected account information can be enriched, and the association relationship between different nodes can be considered, and other information features are added in the to-be-detected account information, thereby improving the accuracy of data detection.
[0214] Further, please refer to Figure 9 , Figure 9 is another data detection device schematic diagram provided by the embodiment of the application. The data detection device can be a computer program (including program code, etc.) running in a computer device, for example, the data detection device can be an application software; the device can be used to execute the corresponding steps in the method provided by the embodiment of the application. As shown in Figure 9 , the data detection device 900 can be used in the computer device in the embodiment corresponding to Figure 7 , specifically, the device can include: a sample acquisition module 21, a first feature extraction module 22, a second feature extraction module 23, a third feature extraction module 24, a sample relationship acquisition module 25, a sample information acquisition module 26, a sample information fusion module 27, a sample prediction module 28 and a model training module 29.
[0215] The sample acquisition module 21 is configured to acquire a public broadcast service group account sample, acquire account information samples, principal information samples and registration information samples of the public broadcast service group account sample, and acquire account sample labels of the public broadcast service group account sample.
[0216] The first feature extraction module 22 is configured to input the account information samples, the principal information samples and the registration information samples into an initial account detection model, and extract sample account features of the account information samples in a text detection network of the initial account detection model.
[0217] The second feature extraction module 23 is configured to extract sample principal features of the principal information samples in the text detection network.
[0218] The third feature extraction module 24 is configured to perform random initialization on the registration information samples in a random initialization network of the initial account detection model, and obtain sample registration features of the registration information samples.
[0219] The sample relationship acquisition module 25 is configured to acquire a first sample association relationship between the account information samples and the principal information samples, and acquire a second sample association relationship between the account information samples and the registration information samples.
[0220] The sample information acquisition module 26 is configured to perform feature fusion on the sample principal features and the sample registration features based on the first sample association relationship and the second sample association relationship, and obtain sample neighborhood association information.
[0221] The sample information fusion module 27 is configured to perform feature fusion on the sample account features and the sample neighborhood association information, and obtain sample fusion features.
[0222] The sample prediction module 28 is configured to predict the sample fusion features based on the classification network of the initial account detection model, to obtain a sample detection result of the public broadcast service group account sample.
[0223] The model training module 29 is configured to adjust parameters of the initial account detection model according to the account sample label and the sample detection result, to obtain the account detection model.
[0224] The device 900 further includes:
[0225] The graph establishing module 30 is configured to obtain account promotion data and an account menu of the public broadcast service group account sample, and obtain an account relationship heterogeneous graph associated with the public broadcast service group account sample according to the account promotion data and the account menu.
[0226] The model optimization module 31 is configured to obtain adjacent account information associated with the registration information sample based on the account relationship heterogeneous graph, and perform parameter optimization on the account detection model based on the adjacent account information, to obtain an optimized account detection model.
[0227] Referring to Figure 10 , Figure 10 is a structural schematic diagram of a computer device provided by an embodiment of the present application. As shown in Figure 10 , the computer device in the embodiment of the present application can include one or more processors 1001, a memory 1002, and an input-output interface 1003. The processor 1001, the memory 1002, and the input-output interface 1003 are connected through a bus 1004. The memory 1002 is configured to store a computer program, the computer program including program instructions, the input-output interface 1003 is configured to receive data and output data, such as for data interaction between the computer device and a terminal device, or for data interaction between each network in the account detection model; and the processor 1001 is configured to execute the program instructions stored in the memory 1002.
[0228] When the processor 1001 is located in Figure 3 , the processor 1001 can perform the following operations:
[0229] Obtain to-be-detected account information, to-be-detected principal information, and to-be-detected registration information of a to-be-detected public broadcast service group account;
[0230] Extract to-be-detected account features of the to-be-detected account information, extract to-be-detected principal features of the to-be-detected principal information, and generate random registration features of the to-be-detected registration information;
[0231] Obtain a first association relationship between the to-be-detected account information and the to-be-detected principal information, and obtain a second association relationship between the to-be-detected account information and the to-be-detected registration information;
[0232] based on the first association relationship and the second association relationship, performing feature fusion on the to-be-detected principal feature and the random registration feature to obtain neighborhood association information;
[0233] performing feature fusion on the to-be-detected account feature and the neighborhood association information to obtain account fusion features, performing prediction on the account fusion features to obtain an account detection result of the to-be-detected public broadcast service group account.
[0234] The processor 1001 is located in the computer device shown in the figure, and can perform the following operations: Figure 7 The processor 1001 is located in the computer device shown in the figure, and can perform the following operations:
[0235] obtaining a public broadcast service group account sample, obtaining account information samples, principal information samples and registration information samples of the public broadcast service group account sample, and obtaining account sample labels of the public broadcast service group account sample;
[0236] inputting the account information samples, the principal information samples and the registration information samples into an initial account detection model, extracting sample account features of the account information samples in a text detection network of the initial account detection model, extracting sample principal features of the principal information samples in the text detection network, and performing random initialization on the registration information samples in a random initialization network of the initial account detection model to obtain sample registration features of the registration information samples;
[0237] obtaining a first sample association relationship between the account information samples and the principal information samples, and obtaining a second sample association relationship between the account information samples and the registration information samples;
[0238] based on the first sample association relationship and the second sample association relationship, performing feature fusion on the sample principal features and the sample registration features to obtain sample neighborhood association information;
[0239] performing feature fusion on the sample account features and the sample neighborhood association information to obtain sample fusion features, and performing prediction on the sample fusion features based on a classification network of the initial account detection model to obtain a sample detection result of the public broadcast service group account sample;
[0240] performing parameter adjustment on the initial account detection model according to the account sample labels and the sample detection result to obtain an account detection model.
[0241] In some possible implementation manners, the processor 1001 can be a central processing unit (CPU), and can also be other general-purpose processors, digital signal processors (DSPs), application specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs) or other programmable logic devices, discrete gates or transistor logic, discrete hardware components, or the like. The general-purpose processor can be a microprocessor or the processor can also be any conventional processor.
[0242] The memory 1002 can include a read-only memory and a random access memory, and provide instructions and data for the processor 1001 and the input output interface 1003. A part of the memory 1002 can further include a non-volatile random access memory. For example, the memory 1002 can further store device type information.
[0243] In specific implementation manners, the computer device can execute the implementation manners provided by each step in the Figure 3 or Figure 7 , and specific implementation manners can be referred to the implementation manners provided by each step in the Figure 3 or Figure 7 , which will not be described herein again.
[0244] Embodiments of the present application provide a computer device, including: a processor, an input output interface, a memory, obtaining a computer program in the memory through the processor, executing the implementation manners provided by each step in the Figure 3The data detection operation is performed on each step of the method shown in the method. The embodiment of the present application realizes obtaining the to-be-detected account information, the to-be-detected subject information and the to-be-detected registration information of the to-be-detected public broadcast service group account; extracting the to-be-detected account feature of the to-be-detected account information, extracting the to-be-detected subject feature of the to-be-detected subject information, and generating the random registration feature of the to-be-detected registration information; obtaining the first association relationship between the to-be-detected account information and the to-be-detected subject information, and obtaining the second association relationship between the to-be-detected account information and the to-be-detected registration information; based on the first association relationship and the second association relationship, the to-be-detected subject feature and the random registration feature are fused to obtain the neighborhood association information; the to-be-detected account feature and the neighborhood association information are fused to obtain the account fusion feature, and the account fusion feature is predicted to obtain the account detection result of the to-be-detected public broadcast service group account. Through the association analysis of the to-be-detected public broadcast service group account, other features are added based on the basic information (i.e. the to-be-detected account information) of the to-be-detected public broadcast service group account based on the association relationship, so that the to-be-detected account information can be enriched, and the association relationship between different nodes can be considered, and other information features are added in the to-be-detected account information, thereby improving the accuracy of data detection.
[0245] The embodiment of the present application also provides a computer readable storage medium, which stores a computer program suitable for being loaded and executed by the processor Figure 3 or Figure 7 The data detection method provided by each step in the method, and the implementation manner provided by each step in the method will be described in detail in the method. Figure 3 or Figure 7 The beneficial effects of the same method are not described again. For technical details not disclosed in the computer readable storage medium embodiment of the present application, please refer to the description of the method embodiment of the present application. As an example, the computer program can be deployed to be executed on one computer device, or on multiple computer devices located in one place, or on multiple computer devices distributed in multiple places and interconnected through a communication network.
[0246] The computer-readable storage medium can be the data detection device provided in any of the foregoing embodiments or the internal storage unit of the computer device, such as the hard disk or memory of the computer device. The computer-readable storage medium can also be an external storage device of the computer device, such as a plug-in hard disk, smart media card (SMC), secure digital (SD) card, flash card, etc., provided on the computer device. Furthermore, the computer-readable storage medium can include both internal storage units and external storage devices of the computer device. The computer-readable storage medium is used to store the computer program and other programs and data required by the computer device. The computer-readable storage medium can also be used to temporarily store data that has been output or will be output.
[0247] This application also provides a computer program product or computer program, which includes computer instructions stored in a computer-readable storage medium. A processor of a computer device reads the computer instructions from the computer-readable storage medium and executes the computer instructions, causing the computer device to perform... Figure 3 or Figure 7 The method provided among the various optional methods realizes the association parsing of the public broadcast service group account to be detected. Based on the basic information of the public broadcast service group account to be detected (i.e. the account information to be detected), other features are added based on the association relationship. This not only enriches the account information to be detected, but also takes into account the association relationship between different nodes, and adds other information features to the account information to be detected with emphasis, thereby improving the accuracy of data detection.
[0248] The terms "first," "second," etc., in the specification, claims, and drawings of this application are used to distinguish different objects, not to describe a specific order. Furthermore, the term "comprising," and any variations thereof, are intended to cover non-exclusive inclusion. For example, a process, method, apparatus, product, or device that includes a series of steps or units is not limited to the listed steps or modules, but may optionally include steps or modules not listed, or may optionally include other step units inherent to these processes, methods, apparatuses, products, or devices.
[0249] Those skilled in the art can understand that the units and algorithm steps of each example described in combination with the embodiments disclosed herein can be realized in electronic hardware, computer software or a combination of both. In order to clearly illustrate the interchangeability of hardware and software, the components and steps of each example have been described in general terms in the description. Whether the functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. A person skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of the present application.
[0250] The method and related apparatus provided by the embodiments of the present application are described with reference to the method flowchart and / or structural schematic diagram provided by the embodiments of the present application. Each flow and / or block in the method flowchart and / or structural schematic diagram, and the combination of the flows and / or blocks in the flowchart and / or block diagram can be realized by computer program instructions. These computer program instructions can be provided to a general-purpose computer, a special-purpose computer, an embedded processor or other programmable data processing devices to produce a machine, so that the instructions executed by the computer or other programmable data processing devices produce a device implemented in the flow Figure 1 The computer program instructions can also be stored in a computer readable memory capable of causing a computer or other programmable data processing devices to work in a specific manner, so that the instructions stored in the computer readable memory produce a manufactured product including instruction devices that implement the functions specified in the flow Figure 1 The computer program instructions can also be stored in a computer readable memory capable of causing a computer or other programmable data processing devices to work in a specific manner, so that the instructions stored in the computer readable memory produce a manufactured product including instruction devices that implement the functions specified in the flow Figure 1 The computer program instructions can also be stored in a computer readable memory capable of causing a computer or other programmable data processing devices to work in a specific manner, so that the instructions stored in the computer readable memory produce a manufactured product including instruction devices that implement the functions specified in the flow Figure 1 The computer program instructions can also be stored in a computer readable memory capable of causing a computer or other programmable data processing devices to work in a specific manner, so that the instructions stored in the computer readable memory produce a manufactured product including instruction devices that implement the functions specified in the flow Figure 1 The computer program instructions can also be stored in a computer readable memory capable of causing a computer or other programmable data processing devices to work in a specific manner, so that the instructions stored in the computer readable memory produce a manufactured product including instruction devices that implement the functions specified in the flow
[0251] The steps in the method embodiments of the present application can be adjusted, combined and deleted according to actual needs.
[0252] The modules in the apparatus embodiments of the present application can be combined, divided and deleted according to actual needs.
[0253] The above disclosure is only the preferred embodiments of the present application, and of course cannot limit the scope of the rights of the present application, so equivalent changes made in accordance with the claims of the present application are still within the scope of the present application.
Claims
1. A data detection method characterized by, The method comprises: obtaining to-be-detected account information, to-be-detected subject information and to-be-detected registration information of a to-be-detected public broadcast service group account; extracting to-be-detected account features of the to-be-detected account information and extracting to-be-detected subject features of the to-be-detected subject information; inputting the to-be-detected registration information into a randomly initialized network, identifying a target registration type of the to-be-detected registration information in the randomly initialized network; obtaining a registration mapping matrix corresponding to the target registration type, and generating random registration features of the to-be-detected registration information based on the registration mapping matrix; obtaining a first association relationship between the to-be-detected account information and the to-be-detected subject information, and obtaining a second association relationship between the to-be-detected account information and the to-be-detected registration information; based on the first association relationship and the second association relationship, performing feature fusion on the to-be-detected subject features and the random registration features to obtain neighborhood association information; performing feature fusion on the to-be-detected account features and the neighborhood association information to obtain account fusion features, and performing prediction on the account fusion features to obtain an account detection result of the to-be-detected public broadcast service group account.
2. The method of claim 1, wherein, The extraction of the to-be-detected account features of the to-be-detected account information comprises: performing word segmentation processing on the to-be-detected account information to obtain m to-be-detected account word groups, and performing vector mapping splicing processing on the m to-be-detected account word groups to obtain initial account features of the to-be-detected account information; m is a positive integer; performing convolution conversion on the initial account features to obtain the to-be-detected account features of the to-be-detected account information; The extraction of the to-be-detected subject features of the to-be-detected subject information comprises: performing word segmentation processing on the to-be-detected subject information to obtain m to-be-detected subject word groups, and performing vector mapping splicing processing on the m to-be-detected subject word groups to obtain initial subject features of the to-be-detected subject information; performing convolution conversion on the initial subject features to obtain the to-be-detected subject features of the to-be-detected subject information.
3. The method of claim 1, wherein, Based on the first association relationship and the second association relationship, the to-be-detected subject features and the random registration features are fused to obtain neighborhood association information, comprising: based on the first association relationship, the to-be-detected account features and the to-be-detected subject features are fused to obtain a first feature association degree corresponding to the first association relationship; based on the second association relationship, the to-be-detected account features and the random registration features are fused to obtain a second feature association degree corresponding to the second association relationship; through the first feature association degree and the second feature association degree, the to-be-detected subject features and the random registration features are weighted and summed to obtain neighborhood association information.
4. The method of claim 3, wherein, Based on the first association relationship, the to-be-detected account features and the to-be-detected subject features are fused to obtain a first feature association degree corresponding to the first association relationship, comprising: Map the to-be-detected account feature to a to-be-detected account linear feature, obtain a first information type corresponding to the to-be-detected principal information, map the to-be-detected principal feature to a to-be-detected principal vector based on a first node mapping matrix corresponding to the first information type; Based on the first correlation, the to-be-detected account linear feature and the to-be-detected principal vector are fused to obtain the first feature correlation degree of the first correlation.
5. The method of claim 4, wherein, The first correlation degree of the first correlation is obtained by performing feature fusion on the to-be-detected account linear feature and the to-be-detected principal vector based on the first correlation, including: Based on the first correlation, the to-be-detected account linear feature and the to-be-detected principal vector are fused to obtain the first feature correlation degree of the first correlation. The first initial correlation degree is obtained by performing a continuous multiplication process on the to-be-detected account linear feature and the to-be-detected principal vector based on the first correlation; Obtain the first tuple importance of the first association tuple composed of the to-be-detected account information, the to-be-detected principal information and the first correlation; 6. The method of claim 3, wherein, The first initial correlation degree is obtained by performing a continuous multiplication process on the to-be-detected account linear feature and the to-be-detected principal vector based on the first correlation. The first feature correlation degree and the second feature correlation degree are normalized to obtain a first correlation weight corresponding to the first feature correlation degree and a second correlation weight corresponding to the second feature correlation degree. The first feature correlation degree and the second feature correlation degree are normalized to obtain a first correlation weight corresponding to the first feature correlation degree and a second correlation weight corresponding to the second feature correlation degree.
7. The method of claim 3, wherein, The first feature correlation degree and the second feature correlation degree are normalized to obtain a first correlation weight corresponding to the first feature correlation degree and a second correlation weight corresponding to the second feature correlation degree. The first feature correlation degree and the second feature correlation degree are normalized to obtain a first correlation weight corresponding to the first feature correlation degree and a second correlation weight corresponding to the second feature correlation degree. The first feature correlation degree and the second feature correlation degree are normalized to obtain a first correlation weight corresponding to the first feature correlation degree and a second correlation weight corresponding to the second feature correlation degree. The first feature correlation degree and the second feature correlation degree are normalized to obtain a first correlation weight corresponding to the first feature correlation degree and a second correlation weight corresponding to the second feature correlation degree.
8. The method of claim 1, wherein, The first feature correlation degree and the second feature correlation degree are normalized to obtain a first correlation weight corresponding to the first feature correlation degree and a second correlation weight corresponding to the second feature correlation degree. The first feature correlation degree and the second feature correlation degree are normalized to obtain a first correlation weight corresponding to the first feature correlation degree and a second correlation weight corresponding to the second feature correlation degree. The first feature correlation degree and the second feature correlation degree are normalized to obtain a first correlation weight corresponding to the first feature correlation degree and a second correlation weight corresponding to the second feature correlation degree. The first feature correlation degree and the second feature correlation degree are normalized to obtain a first correlation weight corresponding to the first feature correlation degree and a second correlation weight corresponding to the second feature correlation degree. The first feature correlation degree and the second feature correlation degree are normalized to obtain a first correlation weight corresponding to the first feature correlation degree and a second correlation weight corresponding to the second feature correlation degree. If the normal probability is greater than the abnormal probability, it is determined that an account detection result of the public broadcast service group account to be detected is an account normal result. If the normal probability is less than or equal to the abnormal probability, it is determined that the account detection result of the public broadcast service group account to be detected is an account abnormal result.
9. The method of claim 8, wherein, The method further includes: If the account detection result of the public broadcast service group account to be detected is the account normal result, the public broadcast service group account to be detected is output. If the account detection result of the public broadcast service group account to be detected is the account abnormal result, the public broadcast service group account to be detected is deleted, and an account abnormal message is sent to a service device associated with the public broadcast service group account to be detected based on the to-be-detected registration information.
10. A data detection method characterized by, The method includes: Obtaining a public broadcast service group account sample, obtaining account information samples, subject information samples and registration information samples of the public broadcast service group account sample, and obtaining account sample labels of the public broadcast service group account sample; Inputting the account information samples, the subject information samples and the registration information samples into an initial account detection model, extracting sample account features of the account information samples in a text detection network of the initial account detection model, extracting sample subject features of the subject information samples in the text detection network, and performing random initialization on the registration information samples in a random initialization network of the initial account detection model to obtain sample registration features of the registration information samples; Obtaining a first sample association relationship between the account information samples and the subject information samples, and obtaining a second sample association relationship between the account information samples and the registration information samples; Based on the first sample association relationship and the second sample association relationship, performing feature fusion on the sample subject features and the sample registration features to obtain sample neighborhood association information; Performing feature fusion on the sample account features and the sample neighborhood association information to obtain sample fusion features, and performing prediction on the sample fusion features based on a classification network of the initial account detection model to obtain sample detection results of the public broadcast service group account sample; According to the account sample labels and the sample detection results, performing parameter adjustment on the initial account detection model to obtain an account detection model.
11. The method of claim 10, wherein, The method further includes: Obtaining account promotion data and account menus of the public broadcast service group account sample, and obtaining an account relationship heterogeneous graph associated with the public broadcast service group account sample according to the account promotion data and the account menus; Based on the account relationship heterogeneous graph, obtaining adjacent account information associated with the registration information sample, and performing parameter optimization on the account detection model based on the adjacent account information to obtain an optimized account detection model.
12. A data detection device, characterized by comprising: The device includes: An information acquisition module configured to acquire to-be-detected account information, to-be-detected subject information and to-be-detected registration information of a public broadcast service group account to be detected; An account feature generation module configured to extract to-be-detected account features of the to-be-detected account information; The main body feature generation module is configured to extract main body features of the to-be-detected main body information; The registration feature generation module is configured to input the to-be-detected registration information into a random initialization network, and identify a target registration type of the to-be-detected registration information in the random initialization network; The registration feature generation module is further configured to obtain a registration mapping matrix corresponding to the target registration type, and generate random registration features of the to-be-detected registration information based on the registration mapping matrix; The relationship acquisition module is configured to acquire a first association relationship between the to-be-detected account information and the to-be-detected main body information, and acquire a second association relationship between the to-be-detected account information and the to-be-detected registration information; The information generation module is configured to perform feature fusion on the to-be-detected main body features and the random registration features based on the first association relationship and the second association relationship, to obtain neighborhood association information; The information fusion module is configured to perform feature fusion on the to-be-detected account features and the neighborhood association information, to obtain account fusion features; The account prediction module is configured to predict the account fusion features, to obtain an account detection result of the to-be-detected public broadcast service group account.
13. A data detection device, characterized by comprising: The device comprises: The sample acquisition module is configured to acquire a public broadcast service group account sample, acquire account information samples, main body information samples and registration information samples of the public broadcast service group account sample, and acquire account sample labels of the public broadcast service group account sample; The first feature extraction module is configured to input the account information samples, the main body information samples and the registration information samples into an initial account detection model, and extract sample account features of the account information samples in a text detection network of the initial account detection model; The second feature extraction module is configured to extract sample main body features of the main body information samples in the text detection network; The third feature extraction module is configured to perform random initialization on the registration information samples in a random initialization network of the initial account detection model, to obtain sample registration features of the registration information samples; The sample relationship acquisition module is configured to acquire a first sample association relationship between the account information samples and the main body information samples, and acquire a second sample association relationship between the account information samples and the registration information samples; The sample information acquisition module is configured to perform feature fusion on the sample main body features and the sample registration features based on the first sample association relationship and the second sample association relationship, to obtain sample neighborhood association information; The sample information fusion module is configured to perform feature fusion on the sample account features and the sample neighborhood association information, to obtain sample fusion features; The sample prediction module is configured to predict the sample fusion features based on a classification network of the initial account detection model, to obtain a sample detection result of the public broadcast service group account sample; The model training module is configured to perform parameter adjustment on the initial account detection model according to the account sample labels and the sample detection result, to obtain an account detection model.
14. A computer device, comprising: The device comprises a processor, a memory and an input / output interface; The processor is connected with the memory and the input / output interface respectively, wherein the input / output interface is configured to receive and output data, the memory is configured to store a computer program, and the processor is configured to call the computer program to enable the computer device to execute the method in any one of claims 1-9 or execute the method in any one of claims 10-11.
15. A computer-readable storage medium, characterized in that, The computer readable storage medium stores a computer program, and the computer program is adapted to be loaded and executed by the processor to enable the computer device having the processor to execute the method in any one of claims 1-9 or execute the method in any one of claims 10-11.
16. A computer program product comprising computer programs / instructions, characterized in that, The computer program / instruction is executed by the processor to implement the method in any one of claims 1-9 or execute the method in any one of claims 10-11.
Citation Information
Patent Citations
Data processing method and device and electronic equipment
CN109887112A
Method and device for identifying article containing specified information and storage medium
CN110598095A