Data detection method, device and equipment and computer storage medium

Abstract

The invention discloses a data detection method and device, equipment and a computer storage medium. The method comprises the steps of obtaining a target time sequence of performing target operation on a cloud disk by a target user; and identifying target image information of the target video image frame by using an image identification algorithm to obtain a target frame identification result sequence. And recognizing target character information in the character data by using a character recognition algorithm to obtain a target character recognition sequence. And constructing the three sequences into a user feature matrix according to a preset construction sequence. And inputting the user feature matrix into a target model, and multiplying feature data of the user feature matrix by a corresponding preset weight by using the target model to obtain a target result. It can be known that the target time sequence, the target frame recognition sequence and the target character recognition sequence of the user are obtained to serve as the user feature matrix, and the user feature matrix is input into the target model for recognition, so that the target model can analyze the feature data of the user in multiple aspects, and therefore the accuracy of recognizing the illegal user is improved.

Description

Technical Field

[0001] The present invention relates to the field of cloud computing, and in particular to a method, device, equipment, computer storage medium and computer program product for data detection. Background Art

[0002] With the development of the information age, users usually save pictures, videos or documents in their daily lives to cloud disks. At present, there is a phenomenon that some users save files with illegal information to cloud disks, so it is necessary to monitor and manage cloud disk users.

[0003] In the prior art, a user is usually blacklisted by configuring an untrusted area containing blacklist conditions on a router, that is, when the router forwards a message and obtains characteristic information that satisfies the blacklist, the user is blacklisted and managed. However, there is still a phenomenon that the user whose message forwarded by the router does not meet the characteristic information of the blacklist has violated the rules, resulting in low accuracy in identifying the violating users. Summary of the invention

[0004] The embodiments of the present disclosure provide a method, apparatus, device, computer storage medium, and computer program product for data detection, which can improve the accuracy of identifying illegal users.

[0005] In a first aspect, an embodiment of the present disclosure provides a method for data detection, the method comprising:

[0006] Obtain the target time series of the target user's target operations on the cloud disk within a preset period;

[0007] The target image information of the target video image frame is identified by using an image recognition algorithm to obtain a target frame recognition result sequence, wherein the target video image frame is obtained based on the video data and image data saved by the target user in the cloud disk; the target text information in the text data is identified by using a text recognition algorithm to obtain a target text recognition sequence, wherein the text data is saved in the cloud disk of the target user;

[0008] The target time sequence, the target frame recognition sequence and the target text recognition sequence are constructed into a user feature matrix according to a preset construction order;

[0009] The user feature matrix is ​​input into the target model, and the target model is used to multiply the feature data of the user feature matrix with the corresponding preset weights to obtain the target result, which is used to identify the user type of the target user.

[0010] In one practicable implementation, a training feature matrix is ​​constructed based on the historical target time sequence, the historical target frame recognition sequence, and the historical target text recognition sequence in a preset construction order;

[0011] Determine the label information corresponding to the training feature matrix;

[0012] Based on the training feature matrix and label information, multiple target fitnesses are determined by utilizing the correspondence between the training feature matrix and label information and fitness, and the parameters corresponding to the target fitness with the largest value are selected from the target fitnesses as the target parameters of the initial model to obtain the target model.

[0013] In an achievable implementation, based on the training feature matrix and label information, a plurality of target fitnesses are determined by using the correspondence between the training feature matrix and label information and fitness, and a parameter corresponding to the target fitness with the largest value is selected from the target fitnesses as the target parameter of the initial model to obtain the target model, including:

[0014] Acquire multiple search individuals, where the search individuals represent a set of parameter sets corresponding to the initial model;

[0015] Determine the error function value based on the training feature matrix and label information;

[0016] Based on the error function value, the fitness of multiple search individuals is determined according to the corresponding relationship between the fitness and the error function value, and multiple fitnesses are obtained, and the fitness represents the quality of the search individuals;

[0017] Select the fitness with the largest value from multiple fitness values ​​as the global optimal value;

[0018] According to the corresponding relationship between the population dispersion and fitness corresponding to the multiple search individuals, the multiple fitnesses are updated to obtain the target fitnesses of the multiple first updated search individuals;

[0019] When it is detected that the number of iterations is greater than or equal to a preset iteration threshold, the target fitness with the largest value is selected from the target fitnesses of the plurality of first update search individuals as the target global optimal value;

[0020] The parameters corresponding to the target global optimal value are taken as target parameters to obtain the target model.

[0021] In one possible implementation, obtaining multiple search individuals includes:

[0022] Obtaining the upper and lower thresholds corresponding to the candidate parameters of the initial model;

[0023] According to the upper and lower thresholds corresponding to the candidate parameters, a plurality of search individuals including randomly generated parameters are obtained, wherein the values ​​of the randomly generated parameters are greater than or equal to the lower threshold, and the values ​​of the randomly generated parameters are less than or equal to the upper threshold.

[0024] In an achievable implementation, based on the corresponding relationship between the population dispersion and fitness corresponding to the multiple search individuals, the multiple fitnesses are updated to obtain the target fitnesses of the multiple first updated search individuals, including:

[0025] Based on multiple fitnesses, the neighborhood search values ​​of multiple search individuals are determined by utilizing the corresponding relationship between the fitnesses and the neighborhood search values;

[0026] Based on the global optimal value and the neighborhood search values ​​of the plurality of search individuals, the intersection values ​​corresponding to the neighborhood search values ​​of the plurality of search individuals are determined by utilizing the corresponding relationship between the neighborhood search value and the intersection value;

[0027] Based on a first comparison result between the generated random number and a preset crossover probability, a crossover value or a neighborhood search value corresponding to the first comparison result is selected to replace the fitness of multiple search individuals to obtain the fitness of multiple second updated search individuals;

[0028] Respectively comparing the fitness of each second update search individual in the plurality of second update search individuals with the reverse learning value of each second update search individual, and selecting a larger value to replace the fitness of each second update search individual in the plurality of second update search individuals, to obtain the fitness of a plurality of third update search individuals;

[0029] Based on the fitness of the plurality of third updated search individuals, determining the population dispersion corresponding to the plurality of third updated search individuals according to the corresponding relationship between the fitness and the population dispersion;

[0030] Based on the population dispersion, the fitness of multiple third updated search individuals is updated accordingly using the correspondence between the fine search value and the fitness or the correspondence between the random search value and the fitness, so as to obtain the target fitness of multiple first updated search individuals.

[0031] In an achievable implementation, based on a first comparison result between a generated random number and a preset crossover probability, a crossover value or a neighborhood search value corresponding to the first comparison result is selected to replace the fitness of multiple search individuals to obtain the fitness of multiple second updated search individuals, including:

[0032] Comparing the preset crossover probability with the generated random number to obtain a first comparison result;

[0033] When the first comparison result indicates that the generated random number is less than the crossover probability, the fitness of the corresponding search individual is replaced by the neighborhood search value to obtain the fitness of the second updated search individual;

[0034] When the first comparison result indicates that the generated random number is greater than or equal to the crossover probability, the fitness of the corresponding search individual is replaced by the crossover value to obtain the fitness of the second updated search individual.

[0035] In an achievable implementation, the fitness of each second update search individual in the plurality of second update search individuals is compared with the reverse learning value of each second update search individual, and the item with the larger value is selected to update the fitness of each second update search individual in the plurality of second update search individuals, so as to obtain the fitness of the plurality of third update search individuals, including:

[0036] determining a reverse learning value of each second update search individual in a plurality of second update search individuals;

[0037] Comparing the fitness of each second update search individual in the plurality of second update search individuals and the reverse learning value of each second update search individual to obtain a second comparison result;

[0038] If the second comparison result indicates that the reverse learning value of the second update search individual is greater than the fitness of the second update search individual, the fitness of the second update search individual is replaced by the reverse learning value of the second update search individual to obtain the fitness of the third update search individual.

[0039] In an achievable implementation, based on the population dispersion, using the correspondence between the fine search value and the fitness or using the correspondence between the random search value and the fitness, the fitness of multiple third updated search individuals is correspondingly updated to obtain the target fitness of multiple first updated search individuals, including:

[0040] Based on the fitness of the plurality of third update search individuals, the fitness selection probability is determined by using the corresponding relationship between the fitness and the fitness selection probability;

[0041] When the population dispersion is less than the fitness selection probability, based on the fitness of the third update search individual, the corresponding relationship between the fine search value and the fitness is used to determine the fine search value corresponding to the third update search individual, and the fine search value is used to replace the fitness of the third update search individual to obtain the target fitness of the first update search individual;

[0042] When the population dispersion is greater than or equal to the fitness selection probability, based on the fitness of the third updated search individual, the random search value corresponding to the third updated search individual is determined using the correspondence between the random search value and the fitness, and the fitness of the third updated search individual is replaced by the random search value to obtain the target fitness of the first updated search individual.

[0043] In one practicable implementation, before using an image recognition algorithm to recognize target image information of a target video image frame and obtaining a target frame recognition result sequence, the data detection method further includes:

[0044] The video data and image data saved by the target user in the cloud disk are extracted at a preset frame extraction frequency to obtain the target video image frame.

[0045] In one feasible implementation, the text recognition algorithm includes a sensitive word recognition algorithm, and the target text information includes sensitive word information.

[0046] In a second aspect, a data detection device is provided, the device comprising:

[0047] An acquisition module is used to acquire a target time sequence of a target user performing a target operation on a cloud disk within a preset period;

[0048] The recognition module is used to recognize the target image information of the target video image frame by using the image recognition algorithm to obtain the target frame recognition result sequence, where the target video image frame is obtained based on the video data and image data saved by the target user in the cloud disk; the target text information in the text data is recognized by using the text recognition algorithm to obtain the target text recognition sequence, where the text data is saved in the cloud disk of the target user;

[0049] A construction module, used to construct a target time sequence, a target frame recognition sequence and a target text recognition sequence into a user feature matrix according to a preset construction order;

[0050] The input module is used to input the user feature matrix into the target model, and use the target model to multiply the feature data of the user feature matrix with the corresponding preset weights to obtain the target result, which is used to identify the user type of the target user.

[0051] In a third aspect, an embodiment of the present disclosure provides a data detection device, the device comprising a processor and a memory storing computer program instructions; the processor reads and executes the computer program instructions to implement a data detection method as described in any one of the first aspects.

[0052] In a fourth aspect, an embodiment of the present disclosure provides a computer-readable storage medium, wherein the computer-readable storage medium stores computer program instructions, and when the computer program instructions are executed by a processor, a method for data detection as described in any one of the first aspects is implemented.

[0053] In a fifth aspect, an embodiment of the present disclosure provides a computer program product, including a computer program, which implements a data detection method as described in any one of the first aspects when the computer program is executed by a processor.

[0054] The embodiments of the present disclosure provide a method, apparatus, device, computer storage medium and computer program product for data detection. The method obtains a target time sequence of a target user performing a target operation on a cloud disk within a preset period. The target image information of a target video image frame is identified using an image recognition algorithm to obtain a target frame recognition result sequence. The target text information in the text data is identified using a text recognition algorithm to obtain a target text recognition sequence. The target time sequence, the target frame recognition sequence and the target text recognition sequence are constructed into a user feature matrix according to a preset construction order. The user feature matrix is ​​input into a target model, and the feature data of the user feature matrix is ​​multiplied by the corresponding preset weights using the target model to obtain the target result. It can be seen that the present disclosure obtains the user's target time sequence, target frame recognition sequence and target text recognition sequence as a user feature matrix, and inputs them into the target model for recognition, so that the target model can analyze the feature data of multiple aspects of the user, thereby improving the accuracy of identifying illegal users. BRIEF DESCRIPTION OF THE DRAWINGS

[0055] In order to more clearly illustrate the technical solution of the embodiment of the present invention, the following is a brief introduction to the drawings required for use in the embodiment of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0056] Figure 1 is a flow chart of a data detection method provided by an embodiment of the present disclosure;

[0057] Figure 2 is a flowchart of a method for training a target model provided by an embodiment of the present disclosure;

[0058] Figure 3 is a flow chart of a method for determining target parameters provided by an embodiment of the present disclosure;

[0059] Figure 4 is a flowchart of a method for obtaining a search individual provided by an embodiment of the present disclosure;

[0060] Figure 5 is a flow chart of a method for obtaining a target fitness of a first update search individual provided by an embodiment of the present disclosure;

[0061] Figure 6 is a flow chart of a method for obtaining the fitness of a second update search individual provided by an embodiment of the present disclosure;

[0062] Figure 7 is a flow chart of a method for obtaining the fitness of a third update search individual provided by an embodiment of the present disclosure;

[0063] Figure 8is a flow chart of another method for obtaining the target fitness of a first update search individual provided by an embodiment of the present disclosure;

[0064] Fig. 9 is a structural schematic diagram of a data detection device provided by an embodiment of the present disclosure;

[0065] Fig.10 It is a structural diagram of a data detection device provided in an embodiment of the present disclosure. DETAILED DESCRIPTION

[0066] The features and exemplary embodiments of various aspects of the present invention will be described in detail below. In order to make the purpose, technical solutions and advantages of the present invention clearer, the present invention will be further described in detail below in conjunction with the accompanying drawings and specific embodiments. It should be understood that the specific embodiments described herein are only intended to explain the present invention, rather than to limit the present invention. For those skilled in the art, the present invention can be implemented without the need for some of these specific details. The following description of the embodiments is only to provide a better understanding of the present invention by illustrating examples of the present invention.

[0067] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, the elements defined by the statement "include..." do not exclude the existence of other identical elements in the process, method, article or device including the elements.

[0068] Before describing the technical solutions provided by the embodiments of the present disclosure, in order to facilitate understanding of the embodiments of the present disclosure, the present disclosure specifically describes the problems existing in the related technologies:

[0069] With the development of the information age, users usually save pictures, videos or documents in their daily lives to cloud disks. At present, there is a phenomenon that some users save files with illegal information to cloud disks, so it is necessary to monitor and manage cloud disk users.

[0070] In the prior art, users are usually blacklisted by configuring an untrusted area containing blacklist conditions on a router, that is, when the router forwards a message and obtains characteristic information that satisfies the blacklist, the user is blacklisted and managed. However, there is still a phenomenon that users whose messages forwarded by the router do not meet the characteristic information of the blacklist have violated the rules, and since this method can only monitor quantitative indicators and cannot monitor the overall user behavior, the accuracy of identifying violating users is not high.

[0071] The embodiments of the present disclosure provide a method, apparatus, device, computer storage medium and computer program product for data detection, which can solve the above-mentioned technical problems existing in the related technologies.

[0072] The disclosed embodiment obtains the target time sequence of the target user performing the target operation on the cloud disk within a preset period. The target image information of the target video image frame is identified by an image recognition algorithm to obtain a target frame recognition result sequence. The target text information in the text data is identified by a text recognition algorithm to obtain a target text recognition sequence. The target time sequence, the target frame recognition sequence, and the target text recognition sequence are constructed into a user feature matrix according to a preset construction order. The user feature matrix is ​​input into the target model, and the target model is used to multiply the feature data of the user feature matrix with the corresponding preset weights to obtain the target result. The disclosed embodiment obtains the user's target time sequence, target frame recognition sequence, and target text recognition sequence as the user feature matrix, and inputs them into the target model for recognition, so that the target model can analyze the feature data of multiple aspects of the user, thereby improving the accuracy of identifying illegal users.

[0073] The following is an introduction to the data detection method provided by the embodiment of the present disclosure.

[0074] Figure 1 FIG. 1 is a flow chart of a method for data detection provided by an embodiment of the present disclosure. Figure 1 The method may include the following steps S110-S140.

[0075] S110: Obtain a target time sequence of a target user performing a target operation on a cloud disk within a preset period.

[0076] The component including the collection function is used to collect the target time series of the user's target operation on the cloud disk within a preset period.

[0077] Among them, the target time series is used to reflect the time sequence in which the user performs the target operation.

[0078] The target operation may include multiple operations, such as a delete operation, a modify operation, an add operation, or an access revision operation of a private file, etc. Each target operation corresponds to a target time series.

[0079] The disclosed embodiment obtains the target time sequence of the target user performing target operations on the cloud disk within a preset period, so as to facilitate the subsequent determination of whether the user has violated the rules based on the feature information of the target time sequence corresponding to the user's operation behavior through the target model.

[0080] In one example, the target time sequence for user A to perform a deletion operation may be (8:00, 13:00, ..., 18:00), and the target time sequence for user A to perform an addition operation may be (9:10, 9:20, ..., 9:50).

[0081] S120: using an image recognition algorithm to identify target image information of a target video image frame to obtain a target frame recognition result sequence; using a text recognition algorithm to identify target text information in text data to obtain a target text recognition sequence.

[0082] The target video image frame is recognized by using a preset image recognition algorithm, and the recognition operation includes recognizing the target image information in the target video image frame. A target frame recognition result sequence is obtained according to the result of recognizing the target image information. The text data is recognized by using a preset text recognition algorithm, and the recognition operation includes recognizing the target text information in the text data. A target text recognition sequence is obtained according to the result of recognizing the target text information.

[0083] The target video image frame is obtained based on the video data and image data saved by the target user in the cloud disk. The target image information may include illegal image information. The text data is saved in the cloud disk of the target user. The target text information may include illegal text information.

[0084] In one example, an image recognition algorithm is used to recognize target image information of a target video image frame. The number of target video image frames is 5. The recognition result of the target image information is determined in the following manner: if the target image information is recognized, the recognition result is determined to be 1; if the target image information is not recognized, the recognition result is determined to be 0. Finally, a target frame recognition result sequence (0, 1, 1, 1, 0) is obtained.

[0085] In another example, a text recognition algorithm is used to recognize target text information in text data. The text data can be divided into 6 parts. The recognition result of the target text information is determined in the following manner: if the target text information is recognized, the recognition result is determined to be 1; if the target text information is not recognized, the recognition result is determined to be 0. Finally, the target text recognition sequence (1, 1, 1, 1, 0, 1) is obtained.

[0086] The disclosed embodiment determines the target frame recognition result sequence through the video images stored by the target user in the cloud disk and the target text recognition sequence through the text data stored by the target user in the cloud disk, so as to facilitate the subsequent judgment of whether the user has violated the regulations according to the feature information of the user's stored data through the target model.

[0087] In one embodiment, before using an image recognition algorithm to recognize target image information of a target video image frame to obtain a target frame recognition result sequence, the data detection method further includes step S121.

[0088] S121: Extract frames of the video data and image data saved by the target user in the cloud disk according to a preset frame extraction frequency to obtain target video image frames.

[0089] The frame extraction tool is used to extract frames from the video data and image data saved by the target user in the cloud disk according to the preset frame extraction frequency to obtain a certain number of target video image frames.

[0090] The disclosed embodiment obtains target video image frames by extracting frames from video data and image data, thereby reducing the length of the target frame recognition result sequence and improving the efficiency of subsequent feature information recognition corresponding to the target frame recognition result sequence.

[0091] In one example, the text recognition algorithm may include a sensitive word recognition algorithm, and the target text information may include sensitive word information.

[0092] S130: constructing a user feature matrix from a target time sequence, a target frame recognition sequence and a target text recognition sequence according to a preset construction order.

[0093] After obtaining the target time sequence, the target frame recognition sequence and the target text recognition sequence, the target time sequence, the target frame recognition sequence and the target text recognition sequence are constructed into a user feature matrix according to a preset construction order.

[0094] In one example, if the target time sequence includes time sequences corresponding to two target operations, the time sequences corresponding to the two target operations can be used as the first and second rows of the user feature matrix. The target frame recognition sequence is used as the third row of the user feature matrix, and the target text recognition sequence is used as the fourth row of the user feature matrix. Each row is aligned with the length of the remaining rows according to the standard that if the number of elements is less than the number of elements in the remaining rows, it is padded with zeros, and if the number of elements is greater than the number of elements in the remaining rows, it is deleted.

[0095] S140: Input the user feature matrix into the target model, and use the target model to multiply the feature data of the user feature matrix with the corresponding preset weights to obtain the target result.

[0096] After constructing the user feature matrix, the user feature matrix can be input into the target model. After the input layer of the target model obtains the user feature matrix, the convolution layer in the target model is used to extract feature data from the user feature matrix. The pooling layer in the target model is then used to select important feature data from the feature data. Finally, the important feature data is multiplied by the corresponding preset weights through the fully connected layer of the target model, and a high-dimensional vector, i.e., the target result, is output.

[0097] The target model may include a convolutional neural network model. The target result is used to identify the user type of the target user. The user type may include a blacklist user, a suspected blacklist user, and a normal user.

[0098] The disclosed embodiment obtains the user's target time series, target frame recognition sequence and target text recognition sequence as a user feature matrix, and inputs them into a target model for recognition, so that the target model can analyze multiple aspects of the user's feature data, thereby improving the accuracy of identifying illegal users.

[0099] In one embodiment, before obtaining the target time sequence of the target user performing the target operation on the cloud disk within a preset period, such as Figure 2 As shown, the data detection method further includes steps S150 to S170.

[0100] S150: constructing a training feature matrix based on the historical target time sequence, the historical target frame recognition sequence and the historical target text recognition sequence according to a preset construction order.

[0101] The historical target time series of the user's target operation on the cloud disk within a preset period is obtained, and the historical target frame recognition sequence and the historical target text recognition sequence are respectively obtained according to the video image data and text data saved by the user in the cloud disk within the historical period. After obtaining the historical target time series, the historical target frame recognition sequence and the historical target text recognition sequence, the historical target time series, the historical target frame recognition sequence and the historical target text recognition sequence are constructed into a training feature matrix according to a preset construction order for subsequent training of the initial model, i.e., adjusting the parameters of the initial model, so that the initial model can recognize the user type more accurately.

[0102] S160: Determine label information corresponding to the training feature matrix.

[0103] For the training feature matrix, manual labeling can be used to determine the label information corresponding to the training feature matrix.

[0104] The label information may include blacklist users, suspected blacklist users, and normal users.

[0105] The disclosed embodiment determines the label information corresponding to the training feature matrix, thereby facilitating the subsequent input into the target model so that the target model can accurately identify the user type of the user feature matrix.

[0106] S170: Based on the training feature matrix and label information, multiple target fitnesses are determined using the correspondence between the training feature matrix and label information and fitness, and parameters corresponding to the target fitness with the largest value are selected from the target fitnesses as target parameters of the initial model to obtain the target model.

[0107] After obtaining the training feature matrix and label information, multiple target fitnesses are obtained according to the corresponding relationship between the training feature matrix and label information and fitness. The parameters corresponding to the target fitness with the largest value are selected from the multiple target fitnesses as the target parameters of the initial model, thereby obtaining the target model.

[0108] Among them, each target fitness corresponds to a set of parameters of the initial model.

[0109] In the disclosed embodiment, multiple target fitnesses are determined by training feature matrices and label information, and the parameters corresponding to the target fitness with the largest value are selected as the target parameters of the initial model, so that the parameters of the target model are more applicable and the target results can be more accurately identified to determine the user type of the target user.

[0110] In one embodiment, based on the training feature matrix and label information, multiple target fitnesses are determined by using the corresponding relationship between the training feature matrix and label information and fitness, and the parameters corresponding to the target fitness with the largest value are selected from the target fitnesses as the target parameters of the initial model to obtain the target model, such as Figure 3 As shown, it may include steps S171 to S177.

[0111] S171: Acquire multiple search individuals.

[0112] The search individuals represent a set of parameters corresponding to the initial model.

[0113] In one example, the parameter set included in the search individual may be parameter Y, parameter U, parameter O, parameter P, and parameter I.

[0114] S172: Determine an error function value based on the training feature matrix and label information.

[0115] After obtaining the training feature matrix and label information, the training feature matrix is ​​input into the initial model to obtain the actual output result. According to the actual output result and the corresponding relationship between the label information and the error function, the error function value is determined.

[0116] S173: Based on the error function value, determine the fitness of multiple search individuals according to the corresponding relationship between the fitness and the error function value to obtain multiple fitnesses.

[0117] After the error function value is determined, the fitness of multiple search individuals is determined according to the corresponding relationship between the fitness and the error function value to obtain multiple fitnesses.

[0118] Fitness characterizes the quality of the search individual, that is, the parameter set corresponding to the search individual with high fitness applied to the target model can make the target recognition result of the target model more accurate.

[0119] In one example, the expression of the correspondence between fitness and error function value may include formula (1).

[0120]

[0121] Among them, fit represents fitness, f represents the error function value, and abs(f) represents the absolute value of the error function value.

[0122] S174: Select the fitness with the largest value from multiple fitness values ​​as the global optimal value.

[0123] After obtaining multiple fitness values, compare the values ​​of the multiple fitness values, and select the fitness value with the largest value from the multiple fitness values ​​as the global optimal value.

[0124] S175: Update multiple fitnesses according to the corresponding relationship between the population dispersion and fitnesses corresponding to the multiple search individuals, and obtain target fitnesses of multiple first updated search individuals.

[0125] The population dispersion corresponding to the multiple search individuals is determined according to the fitness of the multiple search individuals. A new fitness is obtained according to the corresponding relationship between the population dispersion and the fitness corresponding to the multiple search individuals. The fitness of the multiple search individuals is updated and replaced with the new fitness to obtain the target fitness of the multiple first updated search individuals.

[0126] S176: When it is detected that the number of iterations is greater than or equal to a preset iteration threshold, a target fitness with the largest value is selected from the target fitnesses of multiple first update search individuals as the target global optimal value.

[0127] The detection module is used to detect the current training iteration number of the model. If the iteration number is detected to be greater than or equal to the preset iteration threshold, the target fitness values ​​of multiple first update search individuals are compared. The target fitness with the largest value is selected from the target fitness values ​​of multiple first update search individuals as the target global fitness.

[0128] S177: Taking the parameters corresponding to the target global optimal value as the target parameters, a target model is obtained.

[0129] After determining the target global fitness, the parameters corresponding to the target global optimal value are used as target parameters, and the values ​​of the parameters of the target model are set to be consistent with the values ​​of the target parameters, thereby obtaining the target model.

[0130] The disclosed embodiment can accurately obtain the target parameters corresponding to the target fitness by updating the fitness of multiple search individuals, thereby obtaining a target model with the best recognition effect.

[0131] In one embodiment, multiple search individuals are obtained, such as Figure 4 As shown, it may include steps S1711 to S1712.

[0132] S1711: Obtain the upper threshold and lower threshold corresponding to the candidate parameters of the initial model.

[0133] An upper threshold and a lower threshold of candidate parameters of the initial model are obtained based on the setting information.

[0134] In one example, the candidate parameters of the initial model may include parameter F, parameter G, and parameter H. Through the setting information, it can be obtained that the upper threshold of parameter F is 100 and the lower threshold is 50. The upper threshold of parameter G is 60 and the lower threshold is 30. The upper threshold of parameter H is 1.0 and the lower threshold is 0.1.

[0135] S1712: According to the upper threshold and the lower threshold corresponding to the candidate parameters, obtain multiple search individuals including randomly generated parameters.

[0136] After obtaining the upper and lower thresholds corresponding to the candidate parameters, parameter values ​​are randomly generated according to the upper and lower thresholds corresponding to the candidate parameters, and different parameter values ​​are combined to obtain multiple search individuals.

[0137] The value of the randomly generated parameter is greater than or equal to the lower threshold, and the value of the randomly generated parameter is less than or equal to the upper threshold. Each search individual corresponds to a set of randomly generated parameters.

[0138] In an example, the parameter information corresponding to the search individual A1 may be (parameter F: 89, parameter G: 41, parameter H: 0.5), and the parameter information corresponding to the search individual B1 may be (parameter F: 65, parameter G: 38, parameter H: 0.3).

[0139] The disclosed embodiment can determine a plurality of search individuals by obtaining upper and lower thresholds corresponding to the candidate parameters of the initial model, thereby facilitating subsequent updating and iteration of the search individuals to select the target parameters.

[0140] In one embodiment, according to the corresponding relationship between the population dispersion and fitness of the multiple search individuals, the multiple fitnesses are updated to obtain the target fitness of the multiple first updated search individuals, such as Figure 5 As shown, steps S1751-S1756 may be included.

[0141] S1751: Based on multiple fitnesses, neighborhood search values ​​of multiple search individuals are determined using the correspondence between fitnesses and neighborhood search values.

[0142] After obtaining multiple fitness values, the neighborhood search values ​​of multiple search individuals are determined according to the corresponding relationship between the fitness values ​​and the neighborhood search values.

[0143] The expression of the corresponding relationship between fitness and neighborhood search value may include formula (2).

[0144]

[0145] in, represents the value of the d-th dimension parameter corresponding to the i-th search individual in the t-th training process, express Neighborhood search value, d = 1, 2, ..., D. D represents the total dimension. j represents a random search individual different from i, and δ represents a random number between (-1, 1).

[0146] S1752: Based on the global optimal value and the neighborhood search values ​​of the multiple search individuals, the intersection values ​​corresponding to the neighborhood search values ​​of the multiple search individuals are determined using the corresponding relationship between the neighborhood search values ​​and the intersection values.

[0147] After the global optimal value and the neighborhood search values ​​of the multiple search individuals are obtained, the intersection values ​​corresponding to the neighborhood search values ​​of the multiple search individuals are determined according to the corresponding relationship between the neighborhood search values ​​and the intersection values.

[0148] The expression of the corresponding relationship between the neighborhood search value and the intersection value may include formula (3).

[0149]

[0150] in, express The crossover value of It represents the value of the d-th dimension parameter corresponding to the G-th search individual in the t-th training process, and is the global optimal value. It represents the value of the d-th dimension parameter corresponding to the j-th search individual in the t-th training process, and δ represents a random number between (-1,1).

[0151] S1753: Based on the first comparison result between the generated random number and the preset crossover probability, a crossover value or a neighborhood search value corresponding to the first comparison result is selected to replace the fitness of multiple search individuals to obtain the fitness of multiple second updated search individuals.

[0152] A random number is randomly generated. The generated random number is compared with a preset crossover probability to obtain a first comparison result. According to the first comparison result, the fitness of multiple search individuals is replaced by the obtained crossover value or neighborhood search value to obtain the fitness of multiple second search individuals.

[0153] S1754: Compare the fitness of each second update search individual in the multiple second update search individuals with the reverse learning value of each second update search individual, and select the item with a larger value to replace the fitness of each second update search individual in the multiple second update search individuals, so as to obtain the fitness of multiple third update search individuals.

[0154] According to the correspondence between fitness and reverse learning value, the reverse learning value of the second update search individual is determined based on the fitness of the second update search individual. The fitness of each second update search individual is compared with the reverse learning value of each second update search individual, and the item with the larger value is selected to replace the fitness of each second update search individual in the plurality of second update search individuals, thereby obtaining the fitness of the plurality of third update search individuals.

[0155] S1755: Based on the fitness of the plurality of third updated search individuals, determine the population dispersion corresponding to the plurality of third updated search individuals according to the corresponding relationship between the fitness and the population dispersion.

[0156] After the fitness of the plurality of third update search individuals is obtained, the population dispersion corresponding to the plurality of third update search individuals is determined according to the corresponding relationship between the fitness and the population dispersion.

[0157] The expression of the corresponding relationship between fitness and population dispersion may include formula (4).

[0158]

[0159] Among them, k represents the population dispersion. The larger k is, the more dispersed the search population is, and it is in the early stage of iteration. The smaller k is, the more concentrated the search population is, and it is in the middle and late stages of iteration. max Indicates the maximum distance between any two search individuals, Dist avg Represents the average distance of the search individuals.

[0160] In one example, determine Dist max The expression of can include formula (5).

[0161]

[0162] Among them, Dist max represents the maximum distance between any two third update search individuals, It represents the value of the d-th dimension parameter corresponding to the G-th search individual in the t-th training process, and is the global optimal value. It represents the value of the d-th dimension parameter corresponding to the j-th search individual in the t-th training process. D represents the total dimension of the parameters of the search individual.

[0163] In one example, determine Dist avg The expression of can include formula (6).

[0164]

[0165] Among them, Dist avg represents the average distance of the searched individual, It represents the value of the d-th dimension parameter corresponding to the G-th search individual in the t-th training process, and is the global optimal value. Indicates the value of the d-th dimension parameter corresponding to the j-th search individual in the t-th training process. D indicates the total dimension of the parameters of the search individuals. SN indicates the total number of search individuals.

[0166] S1756: Based on the population dispersion, using the correspondence between the fine search value and the fitness or using the correspondence between the random search value and the fitness, the fitness of multiple third update search individuals is correspondingly updated to obtain the target fitness of multiple first update search individuals.

[0167] After determining the population dispersion, the correspondence between the fine search value and the fitness or the correspondence between the random search value and the fitness is calculated, and the fitness of multiple third-updated search individuals is replaced by the update corresponding to the fine search value or the random search value, thereby obtaining the target fitness of multiple first-updated search individuals.

[0168] The disclosed embodiment updates the fitness of multiple search individuals through multiple updating methods, which can effectively determine the target fitness of multiple first updated search individuals, facilitating the subsequent determination of the target global optimal value.

[0169] In one embodiment, based on the first comparison result of the generated random number and the preset crossover probability, the crossover value or the neighborhood search value corresponding to the first comparison result is selected to replace the fitness of multiple search individuals, and the fitness of multiple second updated search individuals is obtained, such as Figure 6 As shown, steps S17531-S17533 may be included.

[0170] S17531: Compare the preset crossover probability and the generated random number to obtain a first comparison result.

[0171] The generated random number is compared with the preset crossover probability to obtain a first comparison result for subsequent updating of the fitness of the replacement search individual based on the first comparison result.

[0172] S17532: When the first comparison result indicates that the generated random number is less than the crossover probability, the fitness of the corresponding search individual is replaced with the neighborhood search value to obtain the fitness of the second updated search individual.

[0173] If it is detected that the first comparison result represents that the generated random number is less than the crossover probability, the fitness of the corresponding search individual is replaced by the acquired neighborhood search value, thereby obtaining the fitness of the second updated search individual.

[0174] S17533: When the first comparison result indicates that the generated random number is greater than or equal to the crossover probability, the fitness of the corresponding search individual is replaced by the crossover value to obtain the fitness of the second updated search individual.

[0175] If it is detected that the first comparison result represents that the generated random number is greater than or equal to the crossover probability, the fitness of the corresponding search individual is replaced by the obtained crossover value to obtain the fitness of the second updated search individual.

[0176] By comparing the preset crossover probability and the generated random number, the disclosed embodiment can replace the fitness of the corresponding search individual with the neighborhood search value or the crossover value according to the comparison result, so as to facilitate the subsequent accurate selection of the target fitness.

[0177] In one embodiment, the fitness of each second update search individual in the plurality of second update search individuals is compared with the reverse learning value of each second update search individual, and the one with the larger value is selected to update the fitness of each second update search individual in the plurality of second update search individuals, so as to obtain the fitness of the plurality of third update search individuals, such as Figure 7 As shown, steps S17541-S17543 may be included.

[0178] S17541: Determine the reverse learning value of each second update search individual among multiple second update search individuals.

[0179] According to the correspondence between the fitness and the reverse learning value, the reverse learning value of each second update search individual in the plurality of second update search individuals is determined based on the fitness of the plurality of second update search individuals.

[0180] Among them, the reverse learning value can be expressed as

[0181] In the disclosed embodiment, a reverse learning strategy is adopted, i.e., a reverse learning value is determined, so that a population composed of multiple search individuals can fully utilize population information to find the optimal value, i.e., the potential area of ​​the target fitness, thereby improving the search accuracy of the algorithm.

[0182] S17542: Compare the fitness of each second update search individual in the plurality of second update search individuals and the reverse learning value of each second update search individual to obtain a second comparison result.

[0183] After the reverse learning value of each second update search individual among the plurality of second update search individuals is obtained, the fitness of each second update search individual is compared with the reverse learning value to obtain a second comparison result.

[0184] S17543: If the second comparison result indicates that the reverse learning value of the second update search individual is greater than the fitness of the second update search individual, the fitness of the second update search individual is replaced by the reverse learning value of the second update search individual to obtain the fitness of the third update search individual.

[0185] If it is detected that the second comparison result indicates that the reverse learning value of the second update search individual is greater than the fitness of the second update search individual, the reverse learning value of the second update search individual replaces the fitness of the second update search individual to obtain the fitness of the third update search individual.

[0186] In one embodiment, if it is detected that the second comparison result indicates that the reverse learning value of the second update search individual is less than or equal to the fitness of the second update search individual, the fitness of the second update search individual is not replaced and is regarded as the fitness of the third update search individual.

[0187] In the disclosed embodiment, by comparing the fitness in the second updated search individual with the reverse learning value, the fitness in the second updated search individual can be replaced according to the comparison result, so as to facilitate the subsequent accurate determination of the target fitness.

[0188] In one embodiment, based on the population dispersion, the fitness of multiple third updated search individuals is updated accordingly by using the corresponding relationship between the fine search value and the fitness or the corresponding relationship between the random search value and the fitness, so as to obtain the target fitness of multiple first updated search individuals, such as Figure 8 As shown, steps S17561-S17563 may be included.

[0189] S17561: Based on the fitness of multiple third updated search individuals, the fitness selection probability is determined using the corresponding relationship between the fitness and the fitness selection probability.

[0190] When the fitness of multiple third update search individuals is obtained, the fitness selection probability corresponding to the third update search individual is determined based on the fitness of the multiple third update search individuals and according to the correspondence between the fitness and the fitness selection probability.

[0191] The expression of the corresponding relationship between fitness and fitness selection probability may include formula (7).

[0192]

[0193] Among them, P i represents the fitness selection probability corresponding to the i-th search individual, fit i Indicates the fitness corresponding to the i-th search individual, fit i ′ It represents the scaled fitness corresponding to the i-th search individual, and SN represents the total number of search individuals.

[0194] In one example, the expression for calculating the scaled fitness corresponding to the i-th search individual is shown in formula (8).

[0195]

[0196] Among them, fit i ′ represents the scaled fitness corresponding to the i-th search individual, fit i represents the fitness corresponding to the i-th search individual, represents the average fitness, μ represents the intermediate parameter, and ε represents a non-negative constant.

[0197] In one example, the expression for calculating the intermediate parameter is shown in formula (9).

[0198]

[0199] Among them, μ represents the intermediate parameter, SN represents the total number of search individuals, and fit i represents the fitness corresponding to the i-th search individual, represents the average fitness.

[0200] S17562: When the population dispersion is less than the fitness selection probability, based on the fitness of the third updated search individual, the correspondence between the fine search value and the fitness is used to determine the fine search value corresponding to the third updated search individual, and the fine search value is used to replace the fitness of the third updated search individual to obtain the target fitness of the first updated search individual.

[0201] Compare the population dispersion of the third update search individual and the fitness selection probability of the third update search individual. If the population dispersion of the third update search individual is less than the fitness selection probability of the third update search individual, the corresponding relationship between the fine search value and the fitness is used to determine the fine search value corresponding to the third update search individual. After determining the fine search value, the fitness of the corresponding third update search individual is replaced with the fine search value to obtain the target fitness of the first update search individual.

[0202] The expression for calculating the refined search value is shown in formula (10).

[0203]

[0204] in, express A fine search value of , γ2 represents the second random number between (-1,1).

[0205] It is understandable that when the population dispersion k is smaller, it means that the entire search population is more concentrated, and in the middle and late stages of the search iteration, a detailed search is needed to achieve the global optimum.

[0206] S17563: When the population dispersion is greater than or equal to the fitness selection probability, based on the fitness of the third updated search individual, the random search value corresponding to the third updated search individual is determined using the correspondence between the random search value and the fitness, and the fitness of the third updated search individual is replaced by the random search value to obtain the target fitness of the first updated search individual.

[0207] If the population dispersion of the third update search individual is greater than or equal to the fitness selection probability of the third update search individual, the random search value corresponding to the third update search individual is determined by using the correspondence between the random search value and the fitness. After the random search value is determined, the fitness of the third update search individual is replaced by the random search value to obtain the target fitness of the first update search individual.

[0208] The expression for calculating the random search value is shown in formula (11).

[0209]

[0210] in, express The random search value of x dmin represents the minimum value of the d-th dimension parameter, x dmax represents the maximum value of the d-th dimension parameter, and γ3 represents a random number between (-1,1).

[0211] It is understandable that when the population dispersion k is larger, it means that the entire search population is more dispersed. In the early stage of the search iteration, it is necessary to search for the optimal solution as much as possible. In the early stage of the search, search individuals with poor fitness still have a certain chance of selection, so as to maintain the diversity of the population.

[0212] The disclosure compares the fitness selection probability of the search individual and the population dispersion, and updates the fitness of the search individual according to the comparison result, so that when the population dispersion is large, that is, in the early stage of the search, the search individual with lower fitness still has the opportunity to be selected, thereby improving the diversity of the population. And when the population dispersion is small, that is, in the late stage of the search, the local search rate is increased, and the convergence to the optimal solution is accelerated.

[0213] like Fig. 9 As shown, the embodiment of the present disclosure further provides a data detection device 900, and the data detection device 900 includes:

[0214] An acquisition module 901 is used to acquire a target time sequence of a target user performing a target operation on a cloud disk within a preset period;

[0215] The recognition module 902 is used to recognize the target image information of the target video image frame by using an image recognition algorithm to obtain a target frame recognition result sequence, where the target video image frame is obtained based on the video data and image data saved by the target user in the cloud disk; recognize the target text information in the text data by using a text recognition algorithm to obtain a target text recognition sequence, where the text data is saved in the cloud disk of the target user;

[0216] A construction module 903 is used to construct a user feature matrix by combining the target time sequence, the target frame recognition sequence and the target text recognition sequence according to a preset construction order;

[0217] The input module 904 is used to input the user feature matrix into the target model, and use the target model to multiply the feature data of the user feature matrix with the corresponding preset weights to obtain a target result, which is used to identify the user type of the target user.

[0218] In some embodiments, the construction module 903 is specifically configured to:

[0219] A training feature matrix is ​​constructed based on the historical target time series, the historical target frame recognition sequence and the historical target text recognition sequence according to a preset construction order;

[0220] Determine the label information corresponding to the training feature matrix;

[0221] Based on the training feature matrix and label information, multiple target fitnesses are determined by utilizing the correspondence between the training feature matrix and label information and fitness, and the parameters corresponding to the target fitness with the largest value are selected from the target fitnesses as the target parameters of the initial model to obtain the target model.

[0222] In some embodiments, the acquisition module 901 is specifically used to:

[0223] Acquire multiple search individuals, where the search individuals represent a set of parameter sets corresponding to the initial model;

[0224] Determine the error function value based on the training feature matrix and label information;

[0225] Based on the error function value, the fitness of multiple search individuals is determined according to the corresponding relationship between the fitness and the error function value, and multiple fitnesses are obtained, and the fitness represents the quality of the search individuals;

[0226] Select the fitness with the largest value from multiple fitness values ​​as the global optimal value;

[0227] According to the corresponding relationship between the population dispersion and fitness corresponding to the multiple search individuals, the multiple fitnesses are updated to obtain the target fitnesses of the multiple first updated search individuals;

[0228] When it is detected that the number of iterations is greater than or equal to a preset iteration threshold, the target fitness with the largest value is selected from the target fitnesses of the plurality of first update search individuals as the target global optimal value;

[0229] The parameters corresponding to the target global optimal value are taken as target parameters to obtain the target model.

[0230] In some embodiments, the acquisition module 901 is specifically used to:

[0231] Obtaining the upper and lower thresholds corresponding to the candidate parameters of the initial model;

[0232] According to the upper and lower thresholds corresponding to the candidate parameters, a plurality of search individuals including randomly generated parameters are obtained, wherein the values ​​of the randomly generated parameters are greater than or equal to the lower threshold, and the values ​​of the randomly generated parameters are less than or equal to the upper threshold.

[0233] In some embodiments, the data detection device 900 further includes a determination module, the determination module being used to determine the neighborhood search values ​​of the plurality of search individuals based on the plurality of fitnesses and using the correspondence between the fitnesses and the neighborhood search values;

[0234] Based on the global optimal value and the neighborhood search values ​​of the plurality of search individuals, the intersection values ​​corresponding to the neighborhood search values ​​of the plurality of search individuals are determined by utilizing the corresponding relationship between the neighborhood search value and the intersection value;

[0235] Based on a first comparison result between the generated random number and a preset crossover probability, a crossover value or a neighborhood search value corresponding to the first comparison result is selected to replace the fitness of multiple search individuals to obtain the fitness of multiple second updated search individuals;

[0236] Respectively comparing the fitness of each second update search individual in the plurality of second update search individuals with the reverse learning value of each second update search individual, and selecting a larger value to replace the fitness of each second update search individual in the plurality of second update search individuals, to obtain the fitness of a plurality of third update search individuals;

[0237] Based on the fitness of the plurality of third updated search individuals, determining the population dispersion corresponding to the plurality of third updated search individuals according to the corresponding relationship between the fitness and the population dispersion;

[0238] Based on the population dispersion, the fitness of multiple third updated search individuals is updated accordingly using the correspondence between the fine search value and the fitness or the correspondence between the random search value and the fitness, so as to obtain the target fitness of multiple first updated search individuals.

[0239] In some embodiments, the data detection device 900 further includes a comparison module, which is used to compare the preset crossover probability and the generated random number to obtain a first comparison result;

[0240] When the first comparison result indicates that the generated random number is less than the crossover probability, the fitness of the corresponding search individual is replaced by the neighborhood search value to obtain the fitness of the second updated search individual;

[0241] When the first comparison result indicates that the generated random number is greater than or equal to the crossover probability, the fitness of the corresponding search individual is replaced by the crossover value to obtain the fitness of the second updated search individual.

[0242] In some embodiments, the data detection device 900 further includes a determination module, the determination module being used to determine a reverse learning value of each second update search individual in the plurality of second update search individuals;

[0243] Comparing the fitness of each second update search individual in the plurality of second update search individuals and the reverse learning value of each second update search individual to obtain a second comparison result;

[0244] If the second comparison result indicates that the reverse learning value of the second update search individual is greater than the fitness of the second update search individual, the fitness of the second update search individual is replaced by the reverse learning value of the second update search individual to obtain the fitness of the third update search individual.

[0245] In some embodiments, the data detection device 900 further includes a determination module, the determination module being used to search for fitness of individuals based on the plurality of third updates, and determine the fitness selection probability using the correspondence between the fitness and the fitness selection probability;

[0246] When the population dispersion is less than the fitness selection probability, based on the fitness of the third update search individual, the corresponding relationship between the fine search value and the fitness is used to determine the fine search value corresponding to the third update search individual, and the fine search value is used to replace the fitness of the third update search individual to obtain the target fitness of the first update search individual;

[0247] When the population dispersion is greater than or equal to the fitness selection probability, based on the fitness of the third updated search individual, the random search value corresponding to the third updated search individual is determined using the correspondence between the random search value and the fitness, and the fitness of the third updated search individual is replaced by the random search value to obtain the target fitness of the first updated search individual.

[0248] In some embodiments, the data detection device 900 further includes a frame extraction module, which is used to extract frames of the video data and image data saved by the target user in the cloud disk according to a preset frame extraction frequency to obtain target video image frames.

[0249] In some embodiments, the text recognition algorithm includes a sensitive word recognition algorithm, and the target text information includes sensitive word information.

[0250] Fig. 9 Each module / unit in the device shown has the function of realizing Figures 1 to 8 The functions of each step executed by the application system in the present invention can achieve the corresponding technical effects. For the sake of brevity, they will not be repeated here.

[0251] Fig.10 A schematic diagram of the hardware structure of data detection provided by an embodiment of the present disclosure is shown.

[0252] The data detection device may include a processor 1001 and a memory 1002 storing computer program instructions.

[0253] Specifically, the processor 1001 may include a central processing unit (CPU), or an application specific integrated circuit (ASIC), or may be configured to implement one or more integrated circuits of the embodiments of the present application.

[0254] The memory 1002 may include a large capacity memory for data or instructions. By way of example and not limitation, the memory 1002 may include a hard disk drive (HDD), a floppy disk drive, a flash memory, an optical disk, a magneto-optical disk, a magnetic tape, or a universal serial bus (USB) drive or a combination of two or more of these. In appropriate cases, the memory 1002 may include a removable or non-removable (or fixed) medium. In appropriate cases, the memory 1002 may be inside or outside the integrated gateway disaster recovery device. In a specific embodiment, the memory 1002 is a non-volatile solid-state memory.

[0255] The memory 1002 may include a read-only memory (ROM), a random access memory (RAM), a magnetic disk storage medium device, an optical storage medium device, a flash memory device, an electrical, optical or other physical / tangible memory storage device. Thus, typically, the memory includes one or more tangible (non-transitory) computer-readable storage media (e.g., a memory device) encoded with software including computer-executable instructions, and when the software is executed (e.g., by one or more processors), it is operable to perform the operations described with reference to the method according to one aspect of the present disclosure.

[0256] The processor 1001 implements any one of the data detection methods in the above embodiments by reading and executing computer program instructions stored in the memory 1002 .

[0257] In one example, the data detection device may further include a communication interface 1003 and a bus 1004. Fig.10 As shown, the processor 1001, the memory 1002, and the communication interface 1003 are connected via a bus 1004 and communicate with each other.

[0258] The communication interface 1003 is mainly used to implement communication between various modules, devices, units and / or equipment in the embodiments of the present application.

[0259] Bus 1004 includes hardware, software or both, and couples the components of online data traffic billing equipment to each other. For example, but not limitation, the bus may include Accelerated Graphics Port (AGP) or other graphics bus, Enhanced Industry Standard Architecture (EISA) bus, Front Side Bus (FSB), Hyper Transport (HT) interconnection, Industry Standard Architecture (ISA) bus, InfiniBand interconnection, Low Pin Count (LPC) bus, Memory bus, Micro Channel Architecture (MCA) bus, Peripheral Component Interconnect (PCI) bus, PCI-Express (PCI-X) bus, Serial Advanced Technology Attachment (SATA) bus, Video Electronics Standards Association Local (VLB) bus or other suitable bus or two or more of these combinations. In appropriate cases, bus 1004 may include one or more buses. Although the embodiment of the present invention describes and illustrates a specific bus, the present invention considers any suitable bus or interconnection. In addition, in conjunction with the method for data detection in the above-mentioned embodiment, the embodiment of the present application also provides a computer storage medium to realize. The computer storage medium stores computer program instructions; when the computer program instructions are executed by the processor, any one of the data detection methods in the above embodiments is implemented.

[0260] An embodiment of the present application also provides a computer program product, including a computer program, which implements any one of the data detection methods in the above embodiments when the computer program is executed by a processor.

[0261] It should be clear that the present invention is not limited to the specific configuration and processing described above and shown in the figures. For the sake of simplicity, a detailed description of the known method is omitted here. In the above embodiments, several specific steps are described and shown as examples. However, the method process of the present invention is not limited to the specific steps described and shown, and those skilled in the art can make various changes, modifications and additions, or change the order between the steps after understanding the spirit of the present invention.

[0262] The functional blocks shown in the above structural block diagram can be implemented as hardware, software, firmware or a combination thereof. When implemented in hardware, it can be, for example, an electronic circuit, an application specific integrated circuit (Application Specific Integrated Circuit, ASIC), appropriate firmware, plug-in, function card, etc. When implemented in software, the elements of the present invention are programs or code segments used to perform the required tasks. The program or code segment can be stored in a machine-readable medium, or transmitted on a transmission medium or communication link by a data signal carried in a carrier. "Machine-readable medium" can include any medium capable of storing or transmitting information. Examples of machine-readable media include electronic circuits, semiconductor memory devices, read-only memory (Read-Only Memory, ROM), flash memory, erasable read-only memory (Erasable ReadOnly Memory, EROM), floppy disks, compact disc read-only memory (Compact Disc Read-Only Memory, CD-ROM), optical discs, hard disks, optical fiber media, radio frequency (Radio Frequency, RF) links, etc. The code segment can be downloaded via a computer network such as the Internet, an intranet, etc.

[0263] It should also be noted that the exemplary embodiments mentioned in the present invention describe some methods or systems based on a series of steps or devices. However, the present invention is not limited to the order of the above steps, that is, the steps can be performed in the order mentioned in the embodiments, or in a different order from the embodiments, or several steps can be performed simultaneously.

[0264] Aspects of the present disclosure are described above with reference to the flowchart and / or block diagram of the method, device (system) and computer program product according to the embodiment of the present disclosure. It should be understood that each box in the flowchart and / or block diagram and the combination of each box in the flowchart and / or block diagram can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable data processing device to produce a machine so that these instructions executed by the processor of the computer or other programmable data processing device enable the implementation of the function / action specified in one or more boxes of the flowchart and / or block diagram. Such a processor can be, but is not limited to, a general-purpose processor, a special-purpose processor, a special application processor, or a field programmable logic circuit. It can also be understood that each box in the block diagram and / or flowchart and the combination of boxes in the block diagram and / or flowchart can also be implemented by dedicated hardware that performs a specified function or action, or can be implemented by a combination of dedicated hardware and computer instructions.

[0265] The above is only a specific implementation of the present invention. Those skilled in the art can clearly understand that for the convenience and simplicity of description, the specific working process of the system, module and unit described above can refer to the corresponding process in the aforementioned method embodiment, and will not be repeated here. It should be understood that the protection scope of the present invention is not limited to this. Any technician familiar with the technical field can easily think of various equivalent modifications or replacements within the technical scope disclosed by the present invention, and these modifications or replacements should be covered within the protection scope of the present invention.

Claims

1. A method for data detection, characterized in that: include: Obtain the target time series of the target user's target operations on the cloud disk within a preset period; Using an image recognition algorithm to identify target image information of a target video image frame, and obtaining a target frame recognition result sequence, wherein the target video image frame is obtained based on the video data and image data saved by the target user in the cloud disk; Using a text recognition algorithm to recognize target text information in the text data to obtain a target text recognition sequence, wherein the text data is stored in a cloud disk of the target user; Constructing the target time sequence, the target frame recognition sequence and the target text recognition sequence into a user feature matrix according to a preset construction order; The user feature matrix is ​​input into a target model, and the target model is used to multiply the feature data of the user feature matrix with corresponding preset weights to obtain a target result, and the target result is used to identify the user type of the target user.

2. The method according to claim 1, characterized in that: Before obtaining the target time sequence of the target user performing the target operation on the cloud disk within a preset period, the method further includes: Constructing a training feature matrix based on the historical target time sequence, the historical target frame recognition sequence and the historical target text recognition sequence according to the preset construction order; Determine label information corresponding to the training feature matrix; Based on the training feature matrix and the label information, multiple target fitnesses are determined using the corresponding relationship between the training feature matrix and the label information and the fitness, and the parameters corresponding to the target fitness with the largest value are selected from the target fitnesses as the target parameters of the initial model to obtain the target model.

3. The method according to claim 2, characterized in that The method of determining a plurality of target fitnesses based on the training feature matrix and the label information and using the corresponding relationship between the training feature matrix and the label information and the fitnesses, and selecting a parameter corresponding to the target fitness with the largest value from the target fitnesses as the target parameter of the initial model to obtain the target model includes: Acquire a plurality of search individuals, wherein the search individuals represent a set of parameter sets corresponding to the initial model; Determining an error function value based on the training feature matrix and the label information; Based on the error function value, determining the fitness of the multiple search individuals according to the corresponding relationship between the fitness and the error function value, to obtain multiple fitnesses, wherein the fitnesses represent the quality of the search individuals; Selecting the fitness with the largest value from the multiple fitness values ​​as the global optimal value; According to the corresponding relationship between the population dispersion and fitness corresponding to the plurality of search individuals, the plurality of fitnesses are updated to obtain the target fitness of the plurality of first updated search individuals; When it is detected that the number of iterations is greater than or equal to a preset iteration threshold, selecting the target fitness with the largest value from the target fitnesses of the plurality of first update search individuals as the target global optimal value; The target model is obtained by taking the parameter corresponding to the target global optimal value as the target parameter.

4. The method according to claim 3, characterized in that: The obtaining of multiple search individuals comprises: Obtaining the upper and lower thresholds corresponding to the candidate parameters of the initial model; According to the upper threshold and lower threshold corresponding to the candidate parameters, multiple search individuals including randomly generated parameters are obtained, the value of the randomly generated parameters is greater than or equal to the lower threshold, and the value of the randomly generated parameters is less than or equal to the upper threshold.

5. The method according to claim 3, characterized in that: The method of updating the plurality of fitnesses according to the corresponding relationship between the population dispersion and fitnesses corresponding to the plurality of search individuals to obtain the target fitnesses of the plurality of first updated search individuals comprises: Based on the multiple fitnesses, determining the neighborhood search values ​​of the multiple search individuals using a correspondence between the fitnesses and the neighborhood search values; Based on the global optimal value and the neighborhood search values ​​of the plurality of search individuals, determining the intersection values ​​corresponding to the neighborhood search values ​​of the plurality of search individuals by using the correspondence between the neighborhood search values ​​and the intersection values; Based on a first comparison result between the generated random number and a preset crossover probability, selecting the crossover value or the neighborhood search value corresponding to the first comparison result to replace the fitness of the plurality of search individuals, to obtain the fitness of a plurality of second updated search individuals; Respectively comparing the fitness of each second update search individual in the plurality of second update search individuals with the reverse learning value of each second update search individual, and selecting an item with a larger value to replace the fitness of each second update search individual in the plurality of second update search individuals, to obtain the fitness of a plurality of third update search individuals; Based on the fitness of the plurality of third updated search individuals, determining the population dispersion corresponding to the plurality of third updated search individuals according to the corresponding relationship between the fitness and the population dispersion; Based on the population dispersion, the fitness of the multiple third updated search individuals is updated accordingly using the correspondence between the fine search value and the fitness or using the correspondence between the random search value and the fitness to obtain the target fitness of the multiple first updated search individuals.

6. The method according to claim 5, characterized in that The first comparison result based on the generated random number and the preset crossover probability, selecting the crossover value or the neighborhood search value corresponding to the first comparison result to replace the fitness of the plurality of search individuals, and obtaining the fitness of the plurality of second updated search individuals, comprises: Comparing the preset crossover probability with the generated random number to obtain a first comparison result; When the first comparison result indicates that the generated random number is less than the crossover probability, the fitness of the corresponding search individual is replaced by the neighborhood search value to obtain the fitness of the second updated search individual; When the first comparison result indicates that the generated random number is greater than or equal to the crossover probability, the fitness of the corresponding search individual is replaced by the crossover value to obtain the fitness of the second updated search individual.

7. The method according to claim 5, characterized in that The step of respectively comparing the fitness of each second update search individual in the plurality of second update search individuals with the reverse learning value of each second update search individual, and selecting a larger value to update the fitness of each second update search individual in the plurality of second update search individuals, to obtain the fitness of a plurality of third update search individuals, comprises: determining a reverse learning value of each second update search individual in the plurality of second update search individuals; Comparing the fitness of each second update search individual in the plurality of second update search individuals with the reverse learning value of each second update search individual to obtain a second comparison result; If the second comparison result indicates that the reverse learning value of the second update search individual is greater than the fitness of the second update search individual, the fitness of the second update search individual is replaced by the reverse learning value of the second update search individual to obtain the fitness of the third update search individual.

8. The method according to claim 5, characterized in that The method of updating the fitness of the plurality of third updated search individuals based on the population dispersion by using the correspondence between the fine search value and the fitness or by using the correspondence between the random search value and the fitness to obtain the target fitness of the plurality of first updated search individuals comprises: Based on the fitness of the plurality of third update search individuals, the fitness selection probability is determined by using the corresponding relationship between the fitness and the fitness selection probability; In the case where the population dispersion is less than the fitness selection probability, based on the fitness of the third update search individual, a fine search value corresponding to the third update search individual is determined by using the correspondence between the fine search value and the fitness, and the fitness of the third update search individual is replaced by the fine search value to obtain the target fitness of the first update search individual; When the population dispersion is greater than or equal to the fitness selection probability, based on the fitness of the third updated search individual, the random search value corresponding to the third updated search individual is determined using the correspondence between random search value and fitness, and the fitness of the third updated search individual is replaced by the random search value to obtain the target fitness of the first updated search individual.

9. The method according to claim 1, characterized in that: Before using the image recognition algorithm to identify the target image information of the target video image frame and obtaining the target frame recognition result sequence, the method further includes: The video data and image data stored by the target user in the cloud disk are extracted at a preset frame extraction frequency to obtain the target video image frame.

10. The method according to claim 1, characterized in that The text recognition algorithm includes a sensitive word recognition algorithm, and the target text information includes sensitive word information.

Images