Data review method, device, and equipment

Abstract

An embodiment of the present invention provides a method, device, and equipment for auditing data. The method includes: determining at least one target selection function from a plurality of preset selection functions, where the selection function is used to select sample data to be labeled; determining at least one target sample data from a plurality of sample data to be labeled according to the target selection function; determining labeled sample data corresponding to the target sample data; using the labeled sample data to optimize and train a preset auditing model, and using the optimized and trained auditing model to audit data to be processed. The auditing method for data provided in this embodiment realizes that different target selection functions can be determined in different application scenarios, and the labeled sample data determined by the target selection function can be used to optimize and train the auditing model, thereby ensuring the auditing accuracy of the auditing model in different application scenarios and improving the quality and efficiency of the auditing model for auditing data.

Description

Technical Field

[0001] The present invention relates to the field of communication technology, and in particular to a data audit method, device and equipment. Background Art

[0002] With the continuous development of multimedia information, e-commerce platforms are developing more and more rapidly, and e-commerce platforms can realize the transaction of goods. In order to increase the transaction rate, sellers on e-commerce platforms will edit and publish pictures and / or text information about the goods. In order to ensure the safety of e-commerce platform operations and ensure that the goods sold by sellers on the platform are in compliance with relevant national and platform regulations, e-commerce platforms will manually review the titles and contents of the goods published by sellers to determine whether they are in violation of regulations.

[0003] In the prior art, when using an audit model to audit data, in order to ensure the audit accuracy of the audit model, the audit model needs to be optimized and trained; specifically, sample data to be labeled is first obtained, and then the audit model is optimized and trained using the sample data. However, when obtaining sample data to be labeled, the sample data to be labeled selected for different application scenarios are often the same, which will affect the audit accuracy of the audit model in different application scenarios. Summary of the invention

[0004] The embodiments of the present invention provide a data audit method, device and equipment, which can be applied to different application scenarios and ensure the audit accuracy of the audit model.

[0005] In a first aspect, an embodiment of the present invention provides a data audit method, comprising:

[0006] Determine at least one target selection function from a plurality of preset selection functions, wherein the selection function is used to select sample data to be labeled;

[0007] Determining at least one target sample data from a plurality of sample data to be labeled according to the target selection function;

[0008] Determining labeled sample data corresponding to the target sample data;

[0009] The labeled sample data is used to optimize and train a preset audit model, and the audit model after the optimization training is used to audit the data to be processed.

[0010] In a second aspect, an embodiment of the present invention provides a data auditing device, including:

[0011] A first selection module, used to determine at least one target selection function from a plurality of preset selection functions, wherein the selection function is used to select sample data to be labeled;

[0012] A first determination module, configured to determine at least one target sample data from multiple sample data to be labeled according to the target selection function;

[0013] The first determination module is further configured to determine labeled sample data corresponding to the target sample data;

[0014] An auditing module, configured to optimize and train a preset auditing model by using the labeled sample data, and audit data to be processed by using the optimized and trained auditing model.

[0015] In a third aspect, an embodiment of the present invention provides an electronic device, including: a memory and a processor; wherein, the memory is used to store one or more computer instructions, and when the one or more computer instructions are executed by the processor, the data auditing method in the first aspect above is implemented.

[0016] In a fourth aspect, an embodiment of the present invention provides a computer storage medium, configured to store a computer program, and when the computer program is executed by a computer, the data auditing method in the first aspect above is implemented.

[0017] By determining at least one target selection function from a preset plurality of selection functions, determining at least one target sample data from the multiple sample data to be labeled according to the target selection function, then determining the labeled sample data corresponding to the target sample data, and then optimizing and training a preset auditing model by using the labeled sample data, it is realized that different target selection functions can be determined based on different application scenarios, and the labeled sample data determined by the target selection function can be used to optimize and train the auditing model, ensuring the auditing accuracy of the auditing model in different application scenarios; when auditing the data to be processed by using the auditing model, the quality and efficiency of the auditing model for auditing the data are effectively ensured, and on the premise of ensuring that the accuracy of the auditing model does not decrease, the cost of manual labeling is effectively reduced, thereby improving the practicability of the method and being conducive to the market promotion and application.

[0018] In a fifth aspect, an embodiment of the present invention provides a model training method, including:

[0019] Determining at least one target selection function from a preset plurality of selection functions, where the selection function is used to select sample data to be labeled;

[0020] Determining at least one target sample data from the multiple sample data to be labeled according to the target selection function;

[0021] Determining labeled sample data corresponding to the target sample data;

[0022] Optimize and train a preset review model using the labeled sample data.

[0023] In a sixth aspect, an embodiment of the present invention provides a training device for a model, including:

[0024] A second selection module, configured to determine at least one target selection function from a preset plurality of selection functions, where the selection function is used to select sample data to be labeled;

[0025] A second determination module, configured to determine at least one target sample data from a plurality of sample data to be labeled according to the target selection function;

[0026] The second determination module is further configured to determine the labeled sample data corresponding to the target sample data;

[0027] A training module, configured to optimize and train a preset review model using the labeled sample data.

[0028] In a seventh aspect, an embodiment of the present invention provides an electronic device, including: a memory and a processor; wherein, the memory is used to store one or more computer instructions, and when the one or more computer instructions are executed by the processor, the training method of the model in the fifth aspect above is implemented.

[0029] In an eighth aspect, an embodiment of the present invention provides a computer storage medium for storing a computer program, and when the computer program is executed by a computer, the training method of the model in the fifth aspect above is implemented.

[0030] In a ninth aspect, an embodiment of the present invention provides a data processing method, including:

[0031] Obtain data to be processed, where the data to be processed includes at least one data object;

[0032] Determine a neural network corresponding to the data object, where the neural network is trained through sample data;

[0033] Analyze and process the data object using the neural network to obtain object information corresponding to the data object.

[0034] In a tenth aspect, an embodiment of the present invention provides a data processing device, including:

[0035] An acquisition module, configured to acquire data to be processed, where the data to be processed includes at least one data object;

[0036] A third determination module, configured to determine a neural network corresponding to the data object, where the neural network is trained through sample data;

[0037] A processing module for analyzing and processing the data object by using the neural network to obtain object information corresponding to the data object.

[0038] In a tenth aspect, an embodiment of the present invention provides an electronic device, including: a memory and a processor; wherein, the memory is used to store one or more computer instructions, and when the one or more computer instructions are executed by the processor, the data processing method in the above ninth aspect is implemented.

[0039] In an eleventh aspect, an embodiment of the present invention provides a computer storage medium for storing a computer program, and when the computer program is executed by a computer, the data processing method in the above ninth aspect is implemented.

[0040] By obtaining the data to be processed and determining the neural network corresponding to the data object, the neural network is obtained by training with sample data, so that different neural networks can be determined based on different data objects, and then the neural network is used to analyze and process the data object to obtain the object information corresponding to the data object, effectively ensuring the quality and efficiency of using the neural network to process data, further expanding the applicable range of the neural network, improving the practicability of the method, and being conducive to market promotion and application. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required for use in the description of the embodiments or the prior art. Obviously, the following drawings are some embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0042] Figure 1a It is a flowchart of a data review method provided by an embodiment of the present invention;

[0043] Figure 1b It is a schematic diagram of an application scenario of a data review method provided by an embodiment of the present invention;

[0044] Figure 2 It is a flowchart of determining at least one target selection function from a preset plurality of selection functions provided by an embodiment of the present invention;

[0045] Figure 3 It is a flowchart of another data review method provided by an embodiment of the present invention;

[0046] Figure 4 It is a flowchart of determining at least one target sample data from a plurality of sample data to be labeled according to the target selection function provided by an embodiment of the present invention;

[0047] Figure 5 Flow chart for determining the sample selection probability corresponding to each sample data according to the target selection function provided by the embodiment of the present invention;

[0048] Figure 6 Flow chart for a training method of a model provided by the embodiment of the present invention;

[0049] Figure 7 Flow chart for a data processing method provided by the embodiment of the present invention;

[0050] Figure 8 Scenario schematic diagram for a data processing method provided by the embodiment of the present invention;

[0051] Figure 9 Flow chart one for a data auditing method provided by the application embodiment of the present invention;

[0052] Figure 10 Process of a data auditing method provided by the application embodiment of the present invention Figure Two ;

[0053] Figure 11 Structure schematic diagram of a data auditing device provided by the embodiment of the present invention;

[0054] Figure 12 For Figure 11 Structure schematic diagram of an electronic device corresponding to the data auditing device shown in the embodiment;

[0055] Figure 13 Structure schematic diagram of a model training device provided by the embodiment of the present invention;

[0056] Figure 14 For Figure 13 Structure schematic diagram of an electronic device corresponding to the model training device shown in the embodiment;

[0057] Figure 15 Structure schematic diagram of a data processing device provided by the embodiment of the present invention;

[0058] Figure 16 For Figure 15 Structure schematic diagram of an electronic device corresponding to the data processing device shown in the embodiment. Detailed implementation manners

[0059] To make the objectives, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention. Apparently, the described embodiments are some but not all of the embodiments of the present invention. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without any creative efforts shall fall within the protection scope of the present invention.

[0060] The terms used in the embodiments of the present invention are only for the purpose of describing specific embodiments and are not intended to limit the present invention. The singular forms "a", "said" and "the" used in the embodiments of the present invention and the appended claims are also intended to include the plural forms unless the context clearly indicates otherwise. "Plural" generally includes at least two, but does not exclude the case of including at least one.

[0061] It should be understood that the term "and / or" used herein is only a description of the association relationship of associated objects, indicating that three relationships may exist. For example, A and / or B may represent: A exists alone, A and B exist simultaneously, and B exists alone. In addition, the character " / " herein generally represents an "or" relationship between the associated objects before and after.

[0062] Depending on the context, the words "if" and "when" as used herein may be interpreted as "when" or "when...", or "in response to determining" or "in response to detecting". Similarly, depending on the context, the phrase "if determined" or "if detecting (stated condition or event)" may be interpreted as "when determined" or "in response to determining", or "when detecting (stated condition or event)" or "in response to detecting (stated condition or event)".

[0063] It should also be noted that the term "comprising", "including" or any other variant thereof is intended to cover non-exclusive inclusion, so that a commodity or system including a series of elements not only includes those elements, but also includes other elements not explicitly listed, or also includes elements inherent to such commodity or system. Without further limitation, the element defined by the statement "including one..." does not exclude the existence of another identical element in the commodity or system including the said element.

[0064] In addition, the step timings in the following method embodiments are only examples and not strictly limited.

[0065] Figure 1a It is a flowchart of a method for auditing a kind of data provided by the embodiments of the present invention; Figure 1b It is a schematic diagram of an application scenario of a method for auditing a kind of data provided by the embodiments of the present invention; Refer to the appendix Figure 1a - 1bAs shown in the figure, this embodiment provides a method for auditing data. The execution subject of this auditing method is a data auditing device. It can be understood that this auditing device can be implemented as software, or a combination of software and hardware. Moreover, this auditing device can adaptively select a selection function that matches the business scenario for different business scenarios. Through the selection function, the most valuable sample data can be selected. Then, on the premise of ensuring that the accuracy of the auditing model does not decrease, the auditing model is optimized using the most valuable sample data. Specifically, this data auditing method may include:

[0066] S101: Determine at least one target selection function from a preset plurality of selection functions. The selection function is used to select sample data to be labeled.

[0067] In this embodiment, the preset plurality of selection functions are used to select sample data to be labeled that matches the application scenario in different application scenarios; and the above-mentioned plurality of selection functions may include at least one of the following functions: margin function, uniform function, bandit_discrete function, kcenter_greedy greedy function, hierarchical_clustering hierarchical clustering function. Among them, the margin function can calculate the amplitude margin, phase margin, and the corresponding frequency from the frequency response data; the uniform(x,y) function can randomly generate a real number within the range of [x,y]; the bandit_discrete function is a simple online learning algorithm and is often used to try to solve the exploit-explore selection problem and the cold start problem.

[0068] Specifically, when selecting the target selection function according to the application scenario, the number of target selection functions can be one or more; specifically, refer to the appendix Figure 2 As shown in the figure, determining at least one target selection function from a preset plurality of selection functions may include:

[0069] S1011: Use the auditing model to obtain the estimated accuracy corresponding to each selection function.

[0070] S1012: Determine the selection function corresponding to the maximum estimated accuracy as the target selection function.

[0071] Specifically, an audit model for auditing data is preset. For each selection function, sample data corresponding to the selection function can be obtained in advance. Then, the sample data corresponding to the selection function is input into the audit model, so that the estimated accuracy corresponding to the selection function can be obtained. After obtaining the estimated accuracy corresponding to each selection function, the selection function corresponding to the maximum estimated accuracy can be determined as the target selection function, where the number of target selection functions is one or more. In this way, the accurate and reliable acquisition of the target selection function can be effectively guaranteed, and the accuracy of the method can be further improved.

[0072] It can be understood that the specific determination method of the target selection function is not limited to the above implementation method. For example, multiple selection functions are preset, and each selection function corresponds to a selection probability. In different application scenarios, different selection functions can correspond to different selection probabilities. In a specific application scenario, the selection function with the maximum selection probability can be determined as the target selection function. Of course, those skilled in the art can also use other methods to determine at least one target selection function, as long as the accurate and reliable determination of the target selection function can be ensured, which will not be elaborated here.

[0073] S102: Determine at least one target sample data from multiple sample data to be labeled according to the target selection function.

[0074] After determining the target selection function corresponding to a specific application scenario, at least one target sample data can be determined from multiple sample data to be labeled according to the target selection function. Specifically, a mapping relationship between the target selection function and the target sample data can be preset, and the target selection function and the mapping relationship are used to determine at least one target sample data from multiple sample data to be labeled; or, the multiple sample data are directly screened by the target selection function, and the sample data that meet the target selection function are determined as at least one target sample data. For example, in the application scenario of e-commerce, multiple sample data sets can be obtained through an e-commerce platform (such as a server, a processor, etc.). Each sample data set can include multiple sample data to be labeled. Then, the target selection function can be used to select at least one sample data from the multiple sample data as the target sample data, which effectively ensures that the selected target sample data is suitable for a specific application scenario (the application scenario of e-commerce).

[0075] S103: Determine the labeled sample data corresponding to the target sample data.

[0076] Among them, the labeled sample data is the labeled sample data corresponding to the target sample data, and the number thereof can be one or more. The labeled sample data refers to the sample data corresponding to a labeling result, and the labeling result can meet a preset standard. Specifically, the specific content of the preset standard met by the labeling result is not limited, and those skilled in the art can set it according to specific application requirements. When the labeling result corresponding to the sample data meets the preset standard, the labeling result at this time can be confirmed as the standard labeling result of the sample data. And the above labeling result can be the review result obtained after pre-reviewing the labeled sample data. The specific implementation method of the above pre-review can include: manual review or model review. For example: the above labeling result can be the review result obtained by manually pre-labeling and reviewing the labeled sample data; or, the above labeling result can also be the review result obtained after analyzing and processing the labeled sample data by using a preset model. More preferably, in order to ensure the accuracy of the labeling result, the labeled sample data can be reviewed manually, so as to accurately obtain the labeling result that meets the preset requirements.

[0077] S104: Optimize and train a preset review model by using the labeled sample data, and review the data to be processed by using the optimized and trained review model.

[0078] Among them, the optimization training means using the labeled sample data as a reference benchmark and continuously adjusting the parameters of the review model to make the performance index of the review model for data review reach a higher level. For example: the review is more accurate, the review efficiency is higher, the review resources are less, etc. Therefore, after obtaining the labeled sample data, the review model can be optimized and trained by using the labeled sample data. When performing the optimization training, the review model can be directly optimized and trained by using the labeled sample data. At this time, if the number of labeled sample data is multiple, the review model can be optimized and trained multiple times, thereby increasing the number and complexity of the optimization training of the review model, but also reducing the efficiency of the optimization training. Therefore, to avoid the above situation, when optimizing and training a preset review model by using the labeled sample data, the labeled sample data can be added to a preset training sample set; and then the review model can be optimized and trained by using the training sample set.

[0079] Specifically, the training sample set is used to store sample data for optimizing the training of the audit model. The sample data may include labeled sample data and unlabeled non-sample data. It can be understood that the labeling results of the non-sample data stored in the training sample set also meet the preset criteria. Therefore, after obtaining one or more labeled sample data, the one or more labeled sample data can be added to the training sample set, and then the training sample set is used to uniformly optimize the training of the audit model, that is: the audit model is optimized and trained using all the sample data included in the training sample. The specific implementation process includes: when the number of labeled sample data included in the training sample set reaches a preset quantity threshold, the training sample set is used to optimize the training of the audit model; or, another feasible way is to optimize and train the audit model using the training sample set according to a preset optimization period. This effectively reduces the number and complexity of optimizing the training of the audit model, and also ensures the quality and efficiency of optimizing the training of the audit model.

[0080] In addition, after the optimization training of the audit model is completed, the optimized audit model can be used to audit the data to be processed. Since the audit model can be continuously optimized and trained through labeled sample data, the quality and efficiency of the audit model for auditing the data to be processed are effectively improved.

[0081] The data auditing method provided in this embodiment determines at least one target selection function from a preset plurality of selection functions, determines at least one target sample data from the plurality of sample data to be labeled according to the target selection function, then determines the labeled sample data corresponding to the target sample data, and then uses the labeled sample data to optimize the training of the preset audit model, thereby realizing that different target selection functions can be determined based on different application scenarios, and the labeled sample data determined by the target selection function can be used to optimize the training of the audit model, ensuring the audit accuracy of the audit model in different application scenarios; when using the audit model to audit the data to be processed, the quality and efficiency of the audit model for auditing the data are effectively ensured, and on the premise of ensuring that the accuracy of the audit model does not decrease, the cost of manual labeling is effectively reduced, thereby improving the practicability of this method and facilitating the market promotion and application.

[0082] Optionally, before determining at least one target selection function from a preset plurality of selection functions, the method in this embodiment may further include:

[0083] S001: Configure a selection probability and a minimum selection probability for each selection function in the plurality of selection functions, and the minimum selection probability is greater than 0.

[0084] Specifically, for each selection function, an initial selection probability can be pre-configured for the selection function. Taking five selection functions as an example, the selection probability of each selection function is denoted as p i = 0.2 (i = 1, 2, 3, 4, 5) and a minimum selection probability p min . Configuring the minimum selection probability for each selection function is to prevent the probability of a certain selection function being selected from being 0. It should be noted that the selection probabilities configured for each selection function among multiple selection functions can be the same or different, and in different application scenarios, the same selection function can correspond to different selection probabilities.

[0085] By configuring the selection probability and the minimum selection probability for each selection function, different selection functions can have the chance of being selected in different application scenarios, which is conducive to the comprehensive reliability of the determination of labeled sample data, and further improves the accuracy of optimizing and training the audit model.

[0086] Figure 3 FIG. is a flowchart of another data auditing method provided by an embodiment of the present invention; on the basis of the above embodiment, continue to refer to the attached Figure 3 As shown, in order to further improve the quality of optimizing and training the audit model, after optimizing and training the preset audit model with the labeled sample data, the method in this embodiment may further include:

[0087] S201: Obtain the model feedback information of the audit model.

[0088] Among them, the model feedback information may include at least one of the following: model feedback accuracy, model feedback performance information, model feedback overhead resource information, etc. Specifically, the model feedback accuracy may refer to the accuracy of the audit model in auditing data, and the model feedback performance information may refer to the performance characterization information of the audit model when auditing data; the model feedback overhead resource may refer to the audit resource information saved when the audit model audits data. It can be understood that the model feedback information may not only include the above information, and those skilled in the art can also set the specific content of the model feedback information according to specific application requirements, which will not be elaborated here.

[0089] S202: Adjust the selection probability of the target selection function according to the model feedback information.

[0090] After obtaining the model feedback information, the selection probability of the target selection function can be adjusted according to the model feedback information to determine whether the target selection function contributes to the optimization training of the audit model. Specifically, when adjusting the selection probability of the target selection function, the selection probability of the target selection function can be the pre-configured initial selection probability, or the historical selection probability after adjustment. For example, the initial selection probability of the target selection function is 0.2. After obtaining the model feedback information, the selection probability of the target selection function can be adjusted. For example, the selection probability of the target selection probability can be adjusted from 0.2 to 0.3. When the model feedback information is obtained again, the historical selection probability of the target selection function can be adjusted according to the model feedback information, that is, 0.3 is adjusted. For example, the selection probability can be adjusted from 0.3 to 0.25.

[0091] Specifically, the steps of adjusting the selection probability of the target selection function according to the model feedback information may include:

[0092] S2021: When the model feedback information is positive feedback, increase the selection probability of the target selection function.

[0093] S2022: When the model feedback information is negative feedback, decrease the selection probability of the target selection function.

[0094] In the specific implementation process, when determining whether the model feedback information is positive feedback or negative feedback, the model audit information of the audit model before optimization training can be obtained first. The model audit information corresponds to the model feedback information. For example, when the model feedback information includes the model feedback accuracy, the model audit information can include the model audit accuracy; when the model feedback information includes the model feedback performance information, the model audit information can include the model performance information. After obtaining the model audit information and the model feedback information, the model feedback information and the model audit information can be compared, and based on the comparison result, it can be determined whether the model feedback information is positive feedback or negative feedback. Specifically, when the model feedback information is greater than the model audit information, it indicates that the contribution degree of the labeled sample data corresponding to the selection function to the optimization training of the audit model is relatively large. That is, it can be determined that the model feedback information at this time is positive feedback, so the weight information corresponding to the selection function can be increased. When the model feedback information is less than the model audit information, it indicates that the contribution degree of the labeled sample data corresponding to the selection function to the optimization training of the audit model is relatively small. That is, it can be determined that the model feedback information at this time is negative feedback, so the weight information corresponding to the selection function can be decreased. When the model feedback information is equal to the model audit information, it indicates that the labeled sample data corresponding to the selection function has no influence on the optimization training of the audit model. At this time, the weight information corresponding to the selection function can be not adjusted.

[0095] By obtaining the model feedback information of the audit model and adjusting the selection probability of the target selection function according to the model feedback information, the selection probability of the selection function is effectively adjusted according to the application scenario, thereby improving the quality of the optimization training of the audit model.

[0096] Figure 4 It is a flowchart for determining at least one target sample data from multiple sample data to be labeled according to the target selection function provided by an embodiment of the present invention; Figure 5 It is a flowchart for determining the sample selection probability corresponding to each sample data according to the target selection function provided by an embodiment of the present invention; On the basis of the above embodiments, continue to refer to FIG. 4-5. In order to ensure the accurate and reliable determination of the target sample data, in this embodiment, determining at least one target sample data from multiple sample data to be labeled according to the target selection function may include:

[0097] S1021: Determine the sample selection probability corresponding to each sample data according to the target selection function.

[0098] Among them, determining the sample selection probability corresponding to each sample data according to the target selection function may include:

[0099] S10211: Determine the to-be-selected probability that each sample data is selected by all selection functions.

[0100] S10212: Determine the sum of the products of the to-be-selected probabilities of each sample data and the selection probability of the target selection function as the sample selection probability corresponding to each sample data.

[0101] When optimizing and training the audit function each time, for sample data, each sample data has a certain probability of being selected by all selection functions. For each sample data, the corresponding to-be-selected probability can be configured according to the specific application scenario. After obtaining the to-be-selected probabilities of each sample data, the sum of the products of the to-be-selected probabilities of each sample data and the selection probability of the target selection function can be determined as the sample selection probability corresponding to each sample data, that is: the sample selection probability corresponding to each sample data is related to the probability of the sample data itself being selected by the selection function and the selection probability of the selection function. When the sample selection probability is relatively large, it indicates that the probability of the sample data being selected is relatively large.

[0102] S1022: Determine at least one sample data corresponding to the maximum sample selection probability as the target sample data.

[0103] After obtaining the sample selection probabilities corresponding to each sample data, the sample selection probabilities of all sample data can be compared, and at least one sample data corresponding to the maximum sample selection probability can be determined as the target sample data. It can be understood that when the maximum sample selection probability is one, the corresponding target sample data is one, and when the maximum sample selection probability is multiple, the corresponding target sample data is multiple.

[0104] Further, when the target sample data is multiple, the method in this embodiment may further include:

[0105] S1023: Screen the multiple target sample data to obtain the screened sample data.

[0106] Specifically, when screening the multiple target sample data, the multiple target sample data can be screened according to a preset screening strategy, so as to obtain the screened sample data. Among them, the screening strategy can be a preset random selection strategy, or the screening strategy can be set with a screening number, and the multiple target sample data can be screened based on the screening number, so as to obtain the screened sample data that meets the screening number.

[0107] Determine the sample selection probability corresponding to each sample data according to the target selection function, and then determine at least one sample data corresponding to the maximum sample selection probability as the target sample data, effectively realizing the accurate and reliable determination of the target sample data, and further ensuring the stable and reliable implementation of the audit method.

[0108] Figure 6 Flow chart of a training method for a model provided by an embodiment of the present invention; refer to the attached Figure 6 As shown, this embodiment provides a training method for a model. The execution subject of this method is a model training device, and this training device can be implemented as software, or a combination of software and hardware. Specifically, this method may include:

[0109] S301: Determine at least one target selection function from a preset plurality of selection functions. The selection function is used to select sample data to be labeled.

[0110] S302: Determine at least one target sample data from a plurality of sample data to be labeled according to the target selection function.

[0111] S303: Determine the labeled sample data corresponding to the target sample data.

[0112] S304: Optimize and train a preset audit model by using the labeled sample data.

[0113] The specific implementation process and implementation effect of the method in this embodiment are similar to those of S101 - S104 in the above - mentioned embodiment. For specific reference, see the description content in the above - mentioned embodiment, and details are not described herein again.

[0114] Figure 7 Flow chart of a data processing method provided by an embodiment of the present invention; Figure 8 Scenario schematic diagram of a data processing method provided by an embodiment of the present invention; refer to the attached Figure 7 - 8 As shown, this embodiment provides a data processing method. The execution subject of this method is a data processing device, and this processing device can be implemented as software, or a combination of software and hardware. Specifically, this method may include:

[0115] S401: Obtain data to be processed, where the data to be processed includes at least one data object.

[0116] Among them, the data to be processed can be any one of the following types of data: image data, video data, application program data, etc.; when the data to be processed is image data, the data object may refer to a person, an animal, an object, or others in the image data; when the data to be processed is video data, the data object may refer to a video frame image in the video data; when the data to be processed is an application program, the data object may refer to the application implementation function corresponding to the application program. Of course, those skilled in the art can also set the data to be processed as other types of data according to specific application requirements, and details are not described herein again.

[0117] S402: Determine a neural network corresponding to the data object, where the neural network is obtained by training with sample data.

[0118] After obtaining the data objects included in the data to be processed, since the number of data objects is one or more, in order to accurately identify and process the data objects, a neural network corresponding to the data objects can be determined. It can be understood that the mapping relationship between the data objects and the neural network can be a one-to-one mapping, a many-to-one mapping, or a one-to-many mapping. The neural network corresponding to the data object can be obtained by training with sample data, for example: the neural network is obtained by optimizing the training in the embodiment corresponding to FIG. 1 above, so that the data object can be analyzed and processed using the neural network.

[0119] S403: Analyze and process the data object using the neural network to obtain object information corresponding to the data object.

[0120] After obtaining the neural network, the data object can be analyzed and processed using the neural network, so as to obtain object information corresponding to the data object. It is understandable that the object information can change with the different uses of the neural network. Specifically, it includes but is not limited to the following application scenarios:

[0121] Application scenario 1: When the neural network is used to identify the price information of a data object, the object information may refer to the price information. Figure 8 As shown, after obtaining the neural network corresponding to the data object, after using the neural network to analyze and process the data object, the price information corresponding to the data object can be obtained, such as: the price of a tennis ball is 1 yuan, the price of an ice hockey puck is 1 yuan, the price of an ice hockey stick is 9 yuan, the price of a football is 10 yuan, and so on.

[0122] It can be imagined that after the price information of the data object can be identified, the data object and the corresponding price information can be uploaded to the network (trading platform or second-hand trading platform) to generate the pending transaction information corresponding to the data object; in order to display the data object more conveniently, the data object can also be photographed from multiple angles to obtain multi-angle object images, and the object images are associated with the pending transaction information and stored, thereby using virtual reality (VR) technology to form the pending transaction information for the data object.

[0123] Furthermore, regarding the information to be traded of the data object, in order to facilitate the management of the information to be traded, the user can establish a corresponding entity warehouse for the information to be traded and store the entity warehouse information in association with the information to be traded. For example: The user can take timed photos of the area used to store multiple data objects to be traded to obtain a real picture of the warehouse, and then store the real picture of the warehouse in association with the information to be traded, so as to facilitate the management and maintenance of the data object.

[0124] Application Scenario 2: When the neural network is used to identify the source information of the data object, the object information can refer to the source information; or, when the neural network is used to identify the historical information of the data object, the object information can refer to the historical information. For example: The data object can be an object whose archaeological value needs to be verified (abbreviation: "archaeology"). After determining the neural network corresponding to the object, the neural network can be used to analyze and identify the object, so as to obtain the source information or historical information corresponding to the object.

[0125] Application Scenario 3: When the neural network is used to give suggestions on the matching of data objects, the object information can be the suggestion information for the matching of objects. For example: The data object includes: the first component, the second component, the third component, and the fourth component to be matched, where the first component can be used in combination with the second component. After determining the neural network corresponding to the data object, the neural network can be used to analyze and identify the matching status between components, so as to obtain the suggestion information on the matching status between components, such as: the first component and the second component can be matched, and the third component and the fourth component cannot be matched, etc.

[0126] Application Scenario 4: When the neural network is used to classify data objects, the object information can refer to the classification information. For example: When the data object is multiple garbage objects to be classified, when determining the neural network corresponding to the garbage objects (the neural network at this time can be determined by the appearance information of the garbage objects), the neural network can be used to classify the garbage objects, such as: Garbage 1 is kitchen waste, and Garbage 2 is recyclable waste.

[0127] Application Scenario 5: When the neural network is used to extract video frame images from a video, the object information can refer to the extracted video frame images. For example: There are currently Video 1, Video 2, and Video 3 from which images are to be extracted, where the shooting angles of Video 1, Video 2, and Video 3 are different. The shooting angle of Video 1 is a close-up, the shooting angle of Video 2 is a medium shot, and the shooting angle of Video 3 is a long shot. At this time, different neural networks can be determined according to the shooting angles of different videos, and then different neural networks can be used to perform image extraction operations on the corresponding videos, so as to obtain the extracted images corresponding to the videos.

[0128] Application Scenario 6: When a neural network can implement a multi-point focusing function based on an application program, the object information may refer to the image information after focusing. For example: In an existing shooting application program, when using the application program to shoot different shooting objects in different scenarios and at different shooting perspectives, the neural network corresponding to the shooting object can be determined. For example, for shooting object 1, a neural network that can implement single-point focusing is used for image processing, and for shooting object 2, a neural network that can implement multi-point focusing is used for image processing, so that the captured images after different focusing strategies can be obtained.

[0129] It can be conceived that those skilled in the art can train different neural networks according to different application requirements, so as to perform different data processing operations on data objects, and then different object information can be obtained, which will not be elaborated here.

[0130] The data processing method provided in this embodiment realizes that different neural networks can be determined based on different data objects by obtaining the data to be processed and determining the neural network corresponding to the data object, where the neural network is trained by sample data. Then, the neural network is used to analyze and process the data object to obtain the object information corresponding to the data object, effectively ensuring the quality and efficiency of using the neural network to process data, further expanding the applicable range of the neural network, improving the practicability of this method, and being conducive to market promotion and application.

[0131] Based on the above embodiments, continue to refer to the appendix Figure 7 - 8 As shown, this embodiment does not limit the specific implementation manner of determining the neural network corresponding to the data object. Those skilled in the art can set it according to specific application requirements. Among them, when the data to be processed is image data, a way to implement determining the neural network corresponding to the data object includes:

[0132] S4021: Obtain the shooting angle of the data to be processed.

[0133] S4022: Determine the neural network corresponding to the data object according to the shooting angle.

[0134] Specifically, when the data to be processed is image data, the shooting angle of the data to be processed can be determined first, and the neural network for processing the data object can be determined based on the shooting angle. For example: First, use an image acquisition device to shoot the same object at a first shooting angle and a second shooting angle, so that the first image corresponding to the first shooting angle and the second image corresponding to the second shooting angle can be obtained. When processing the data object in the first image, the first neural network can be determined according to the first shooting angle, so that the data object can be processed using the first neural network; when processing the data object in the second image, the second neural network can be determined according to the second shooting angle, so that the data object can be processed using the second neural network. Thus, it is possible to use different neural networks to process image data at different angles (or image data with interactions at different angles), effectively improving the quality and efficiency of processing the data object.

[0135] In addition, when there are multiple data objects, another way to determine the neural network corresponding to the data object includes:

[0136] S4023: Obtain the object type of the data object.

[0137] S4024: Determine the neural network corresponding to the data object according to the object type.

[0138] Specifically, when the data to be processed is image data, a neural network can be used to analyze and process the data objects in the image data, so that the relevant information corresponding to each data object in the image data can be obtained. For example, when at least one of the data objects in the image data includes: football, ice hockey, ice hockey stick, tennis, different neural networks can be determined according to the different data types to which the data objects belong. For example: football corresponds to the first neural network, ice hockey corresponds to the second neural network, ice hockey stick corresponds to the third neural network, and tennis corresponds to the fourth neural network. Furthermore, different neural networks can be used to process the data objects, so that the price information corresponding to each data object can be determined.

[0139] It should be noted that the mapping relationship between data objects and neural networks is not limited to the one-to-one mapping described above, but can also be a many-to-one mapping or a one-to-many mapping. For example, football corresponds to the first neural network, ice hockey and hockey stick correspond to the second neural network, and tennis corresponds to the third neural network. The corresponding relationship between the above-mentioned ice hockey and hockey stick and the second neural network is a many-to-one mapping relationship. Or, football corresponds to the first neural network, ice hockey and hockey stick correspond to the second neural network, and tennis can correspond to the first neural network and the third neural network. The corresponding relationship between the above-mentioned tennis and the first neural network and the third neural network is a one-to-many mapping relationship. It can be envisioned that those skilled in the art can arbitrarily set the mapping relationship between the above-mentioned data objects and neural networks according to specific design requirements, which will not be elaborated here.

[0140] In addition, when there are multiple data objects, there is another implementation method for determining the neural network corresponding to the data objects, including:

[0141] S4025: Obtain the interaction information input by the user for the data objects.

[0142] S4026: Group the multiple data objects according to the interaction information to obtain object grouping information.

[0143] S4027: Determine the neural network corresponding to the data objects according to the object grouping information.

[0144] Specifically, when the data objects included in the data to be processed are obtained, the user can input interaction information for the data objects. After the interaction information input by the user is obtained, the data objects can be grouped according to the user's interaction information, so that object grouping information can be obtained. Then, the neural network corresponding to the data objects can be determined according to the object grouping information. It can be envisioned that one or more data objects in the same object grouping information can correspond to the same neural network.

[0145] For example: The data objects include: a hockey puck, a hockey stick, a football, and a tennis ball. When the user performs an associated interaction on the hockey puck and the hockey stick, the hockey puck and the hockey stick can be determined as the first object grouping information, and this object grouping information includes the hockey puck and the hockey stick; when the user performs an associated interaction on the football and the tennis ball, the football and the tennis ball can be determined as the second object grouping information, and this object grouping information includes the football and the tennis ball; then, a first neural network corresponding to the hockey puck and the hockey stick can be determined according to the first object grouping information (this neural network can be determined by the common keywords in the first object grouping information), and a second neural network corresponding to the football and the tennis ball can be determined according to the second grouping information. Or, the data objects include: a hockey puck, a hockey stick, a football, and a tennis ball. When the user performs an associated interaction on the hockey puck, the football, and the tennis ball, the hockey puck, the football, and the tennis ball can be determined as the first object grouping information, and this object grouping information includes the hockey puck, the football, and the tennis ball; at this time, the remaining one data object (the hockey stick) can be automatically used as the second object grouping information. Then, a third neural network corresponding to the hockey puck, the football, and the tennis ball can be determined according to the first object grouping information (this neural network can be determined by the common keywords in the first object grouping information), and a fourth neural network corresponding to the hockey stick can be determined according to the second grouping information.

[0146] Through the above process, it is effectively realized that the neural network corresponding to the data object can be determined through the interaction information input by the user, effectively realized that the determined neural network can meet the needs of the user, and can also ensure the quality and efficiency of processing the data object.

[0147] In specific applications, refer to Att Figure 9 - 10 As shown in the figure, taking the application scenario in the e-commerce field as an example, this application embodiment provides a data review method, which realizes a semi-artificial (manual annotation of sample data so that the sample data can obtain a labeling result that meets the preset standard) and semi-machine review data review method. It can not only effectively reduce the manual annotation cost on the premise of ensuring that the accuracy of the review model (classifier) does not decrease; but also be applicable to different application scenarios. Specifically, the method includes the following steps:

[0148] step1: When a new service is accessed, all data corresponding to the new service can be obtained. At this time, all data is not labeled. Then, through the method of hierarchical clustering, the TOP K samples in each cluster can be selected and given to manual annotation (expert annotation), so that labeled samples can be obtained.

[0149] step2: Using the above labeled samples as the initial training example set, extract the features of each dimension of the commodity respectively to construct a benchmark classifier (review model). Specifically, the construction steps of the benchmark classifier are as follows:

[0150] (1) Extract the original features of the images by using the Convolutional Neural Network - Inception V3 algorithm on the labeled samples.

[0151] (2) Extract the text - type data by using the fasttext algorithm on the labeled samples. The text - type data includes product title information, category information, and so on.

[0152] (3) Quantify the obtained category information by using the Weight of Evidence (WOE) algorithm.

[0153] (4) For the labeled data, extract the product dimension and the data features of the product dimension. The data features of the product dimension include: product star rating, merchant sales volume, product price, and other features.

[0154] (5) Extract the features of the structured data (the above - mentioned text - type data and the original image features) through multiple hidden layers, and pass them together with the unstructured data through the attention model network. Finally, obtain the embedding layer embedding, and classify it through the softmax function, so as to obtain a baseline classifier, which is used to audit the data.

[0155] Step 3: Select the sample data to be labeled through a selection strategy.

[0156] Considering the scenario where there are many services on the audit platform and there are large differences between services, in order to obtain better results in the selection of the sample data to be labeled, this application embodiment can adopt multiple selection functions. In this way, it can not only meet the requirements of multiple different service scenarios at the same time, but also has a high efficiency in data selection. Specifically, the selection function and the samples can be regarded as a multi - armed bandit problem, obtain the feedback information after each selection, and can update the probability of the selection function according to the feedback information after each selection, so as to adaptively select the most effective selection function, and then select the optimal sample data to be labeled according to the most effective selection function.

[0157] Specifically, it includes the following steps:

[0158] (3.1) Determine at least one target selection function among the preset multiple selection functions. The selection function is used to select the sample data to be labeled.

[0159] Specifically, by using the audit model, the estimated accuracy corresponding to each selection function is obtained; the selection function corresponding to the maximum estimated accuracy is determined as the target selection function. Additionally, before determining at least one target selection function among the preset multiple selection functions, a selection probability and a minimum selection probability are configured for each selection function among the multiple selection functions, and the minimum selection probability is greater than 0.

[0160] (3.2) Determine at least one target sample data from the multiple sample data to be labeled according to the target selection function.

[0161] Among them, the multiple selection functions may include at least one of the following: margin function, uniform function, bandit_discrete function, kcenter_greedy function, hierarchical_clustering function. For each selection function, an initial selection probability value can be set and denoted as p i and a minimum probability value p min (to prevent the probability of a certain selection function being selected from being 0). Then, for the above-mentioned baseline classifier (audit model), denote the probability that the k-th function can be used to select the sample data to be labeled in the t-th iteration process as p k (t); in each iteration, for each unlabeled sample data, it can be selected by all the selection functions h(t). Denote the probability that the k-th selection function selects the sample data in t iterations as h k (t); then the final probability of selecting sample data j That is, the sum of the products of the to-be-selected probabilities of each sample data and the selection probability of the target selection function is determined as the sample selection probability corresponding to each sample data, so that the sample selection probability corresponding to selecting sample data j can be obtained. Then, the sample data corresponding to the maximum sample selection probability can be determined as the target sample data.

[0162] step4: The expert labels the selected samples.

[0163] After determining the target sample data, the expert can be enabled to determine the labeled sample data corresponding to the target sample data.

[0164] step5: After obtaining the labeled sample data, the baseline classifier can be optimized and trained using the labeled sample data, and the optimized and trained baseline classifier can be used to audit other data to be processed.

[0165] step6: Detect whether the model accuracy of the audit model meets the preset standard.

[0166] Step 7: If the model accuracy does not meet the preset standard, continue to train the review model.

[0167] If the model accuracy meets the preset standard, use the review model to review the unlabeled data.

[0168] Step 8: Conduct a gray-box sampling inspection on the review results of the review model for the unlabeled data.

[0169] Step 9: Determine whether the machine review accuracy (actual accuracy) meets the preset standard according to the gray-box sampling inspection results.

[0170] If the machine review accuracy does not meet the preset standard, the review model can be continuously trained, and the labeled results that meet the preset standard corresponding to the data can also be obtained; the data is directly determined as new sample data. Or, the data is directly used as alternative data for new sample data. If the machine review accuracy meets the preset standard, other unlabeled data can be continuously reviewed.

[0171] Step 10: After using the review model to review the data to be processed, obtain the model feedback information of the review model; adjust the selection probability of the target selection function according to the model feedback information.

[0172] Specifically, obtain the model review information of the review model before the optimization training; when the model feedback information is greater than the model review information, increase the selection probability of the target selection function; or, when the model feedback information is less than the model review information, decrease the selection probability of the target selection function; thus, the selection probability of the selection function is updated.

[0173] The data processing method provided in this application embodiment can select the optimal selection function and the most valuable sample to be labeled from multiple selection functions through the multi-armed bandit method according to the requirements of different scenarios in each iteration, optimize and train the review model according to the most valuable sample to be labeled, and then adaptively update the selection probability of the corresponding selection function; the selected sample data is better and more general; it not only effectively solves the problem of a large amount of labeling and high cost in supervised learning, but also effectively avoids the problem of unstable accuracy in semi-supervised learning and the problem that traditional active learning cannot adapt to multiple scenarios; it realizes the training and optimization of the review model, and then uses the review model to process the data, thus forming an effective closed loop, and the entire process can be automatically and efficiently carried out, further reflecting the self-learning ability of this method and ensuring the intelligent degree of the use of this method.

[0174] Figure 11 It is a schematic structural diagram of a data review device provided in an embodiment of the present invention; refer to the appendixFigure 11 As shown in the figure, this embodiment provides a data review device, which can execute the above-mentioned data review method. Specifically, the device may include: a first selection module 11, a first determination module 12, and a review module 13.

[0175] The first selection module 11 is configured to determine at least one target selection function from a plurality of preset selection functions, and the selection function is used to select sample data to be labeled;

[0176] The first determination module 12 is configured to determine at least one target sample data from a plurality of sample data to be labeled according to the target selection function;

[0177] The first determination module 12 is further configured to determine labeled sample data corresponding to the target sample data;

[0178] The review module 13 is configured to optimize and train a preset review model by using the labeled sample data, and review the data to be processed by using the optimized and trained review model.

[0179] Optionally, when the first selection module 11 determines at least one target selection function from a plurality of preset selection functions, the first selection module 11 may be configured to execute: using the review model to obtain the estimated accuracy corresponding to each selection function; determining the selection function corresponding to the maximum estimated accuracy as the target selection function.

[0180] Optionally, before determining at least one target selection function from a plurality of preset selection functions, the review module 13 in this embodiment may further be configured to: configure a selection probability and a minimum selection probability for each selection function in the plurality of selection functions, and the minimum selection probability is greater than 0.

[0181] Optionally, after optimizing and training the preset review model by using the labeled sample data, the review module 13 in this embodiment may further be configured to execute: obtaining model feedback information of the review model; adjusting the selection probability of the target selection function according to the model feedback information.

[0182] Optionally, when the review module 13 adjusts the selection probability of the target selection function according to the model feedback information, the review module 13 may be configured to execute: when the model feedback information is positive feedback, increasing the selection probability of the target selection function; or, when the model feedback information is negative feedback, decreasing the selection probability of the target selection function.

[0183] Optionally, when the first determination module 12 determines at least one target sample data from multiple sample data to be labeled according to the target selection function, the first determination module 12 may be configured to: determine the sample selection probability corresponding to each sample data according to the target selection function; determine at least one sample data corresponding to the maximum sample selection probability as the target sample data.

[0184] Optionally, when the first determination module 12 determines the sample selection probability corresponding to each sample data according to the target selection function, the first determination module 12 may be configured to: determine the to-be-selected probability that each sample data is selected by all selection functions; determine the sum of the products of the to-be-selected probability of each sample data and the selection probability of the target selection function as the sample selection probability corresponding to each sample data.

[0185] Optionally, when there are multiple target sample data, the review module 13 in this embodiment may also be configured to: screen the multiple target sample data to obtain the screened sample data.

[0186] Figure 11 The device shown can execute Figure 1a 、 Figure 1b 、 Figure 2 - Figure 5 the method of the embodiment shown. For parts not described in detail in this embodiment, reference may be made to the relevant descriptions of Figure 1a 、 Figure 1b 、 Figure 2 - Figure 5 the embodiments shown. For the execution process and technical effects of this technical solution, refer to the descriptions in Figure 1a 、 Figure 1b 、 Figure 2 - Figure 5 the embodiments shown, which will not be elaborated here.

[0187] In a possible design, Figure 11 the structure of the data review device shown may be implemented as an electronic device, and the electronic device may be various devices such as a mobile phone, a tablet computer, a server, etc. As Figure 12 shown, the electronic device may include: a first processor 21 and a first memory 22. Among them, the first memory 22 is used to store a program that supports the electronic device to execute the data review method provided in the above Figure 1a 、 Figure 1b 、 Figure 2 - Figure 5 embodiments, and the first processor 21 is configured to execute the program stored in the first memory 22.

[0188] The program includes one or more computer instructions, and when one or more computer instructions are executed by the first processor 21, the following steps can be implemented:

[0189] Determine at least one target selection function from a plurality of preset selection functions, where the selection function is used to select sample data to be labeled;

[0190] Determine at least one target sample data from a plurality of sample data to be labeled according to the target selection function;

[0191] Determine the labeled sample data corresponding to the target sample data;

[0192] Use the labeled sample data to optimize and train a preset audit model, and use the optimized and trained audit model to audit the data to be processed.

[0193] Optionally, the first processor 21 is further configured to execute all or part of the steps in the foregoing Figure 1a 、 Figure 1b 、 Figure 2 - Figure 5 illustrated embodiments.

[0194] Wherein, the structure of the electronic device may further include a first communication interface 23 for the electronic device to communicate with other devices or communication networks.

[0195] In addition, an embodiment of the present invention provides a computer storage medium for storing computer software instructions used by an electronic device, which includes programs for executing the data auditing method in the foregoing Figure 1a 、 Figure 1b 、 Figure 2 - Figure 5 illustrated method embodiments.

[0196] Figure 13 is a schematic structural diagram of a model training device provided by an embodiment of the present invention; referring to Appendix Figure 13 illustrated, an embodiment of the present invention provides a model training device, which can execute the above-mentioned model training method. Specifically, the device may include: a second selection module 31, a second determination module 32, and a training module 33.

[0197] The second selection module 31 is configured to determine at least one target selection function from a plurality of preset selection functions, where the selection function is used to select sample data to be labeled;

[0198] The second determination module 32 is configured to determine at least one target sample data from a plurality of sample data to be labeled according to the target selection function;

[0199] The second determination module 32 is further configured to determine the labeled sample data corresponding to the target sample data;

[0200] The training module 33 is configured to optimize and train a preset audit model by using the labeled sample data.

[0201] Figure 13 The illustrated device can executeFigure 6 For the method of the illustrated embodiment, for parts not described in detail in this embodiment, reference may be made to the relevant descriptions of the Figure 6 illustrated embodiment. For the execution process and technical effects of this technical solution, refer to the Figure 6 description in the illustrated embodiment and will not be elaborated here.

[0202] In a possible design, Figure 13 the structure of the training device of the illustrated model can be implemented as an electronic device, which can be various devices such as a mobile phone, a tablet computer, a server, etc. As Figure 14 illustrated, the electronic device may include: a second processor 41 and a second memory 42. Among them, the second memory 42 is used to store a program that supports the electronic device to execute the Figure 6 training method of the model provided in the illustrated embodiment, and the second processor 41 is configured to execute the program stored in the second memory 42.

[0203] The program includes one or more computer instructions. When one or more computer instructions are executed by the second processor 41, the following steps can be implemented:

[0204] Determine at least one target selection function among a preset plurality of selection functions, where the selection function is used to select sample data to be labeled;

[0205] Determine at least one target sample data among the plurality of sample data to be labeled according to the target selection function;

[0206] Determine the labeled sample data corresponding to the target sample data;

[0207] Use the labeled sample data to optimize and train a preset audit model.

[0208] Among them, the structure of the electronic device may further include a second communication interface 43 for the electronic device to communicate with other devices or communication networks.

[0209] In addition, an embodiment of the present invention provides a computer storage medium for storing computer software instructions used by an electronic device, which includes a program involved in executing the Figure 6 training method of the model in the illustrated method embodiment.

[0210] Figure 15 It is a schematic structural diagram of a data processing device provided by an embodiment of the present invention; this embodiment provides a data processing device, which can execute the above data processing method. Specifically, the device may include: an acquisition module 51, a third determination module 52, and a processing module 53.

[0211] An acquisition module 51, configured to acquire data to be processed, where the data to be processed includes at least one data object;

[0212] A third determination module 52, configured to determine a neural network corresponding to the data object, where the neural network is obtained by training with sample data;

[0213] A processing module 53, configured to analyze and process the data object by using the neural network to obtain object information corresponding to the data object.

[0214] Optionally, when the data to be processed is image data, when the third determination module 52 determines the neural network corresponding to the data object, the third determination module 52 may be configured to: obtain the shooting angle of the data to be processed; determine the neural network corresponding to the data object according to the shooting angle.

[0215] Optionally, when there are multiple data objects, when the third determination module 52 determines the neural network corresponding to the data object, the third determination module 52 may be configured to: obtain the object type of the data object; determine the neural network corresponding to the data object according to the object type.

[0216] Optionally, when there are multiple data objects, when the third determination module 52 determines the neural network corresponding to the data object, the third determination module 52 may be configured to: obtain interaction information input by a user for the data object; group the multiple data objects according to the interaction information to obtain object grouping information; determine the neural network corresponding to the data object according to the object grouping information.

[0217] Figure 15 The illustrated device may execute Figure 7 - 8 the method of the illustrated embodiment. For parts not described in detail in this embodiment, reference may be made to the relevant descriptions of Figure 7 - 8 the illustrated embodiment. For the execution process and technical effects of this technical solution, reference may be made to the descriptions in Figure 7 - 8 the illustrated embodiment and will not be elaborated herein.

[0218] In a possible design, Figure 15 the structure of the data processing device illustrated may be implemented as an electronic device, and the electronic device may be various devices such as a mobile phone, a tablet computer, a server, etc. As Figure 16 illustrated, the electronic device may include: a third processor 61 and a third memory 62. Among them, the third memory 62 is configured to store a program for supporting the electronic device to execute the data processing method provided in the above Figure 7 - 8 illustrated embodiment, and the third processor 61 is configured to execute the program stored in the third memory 62.

[0219] The program includes one or more computer instructions, and when the one or more computer instructions are executed by the third processor 61, the following steps can be implemented:

[0220] Obtain the data to be processed, where the data to be processed includes at least one data object;

[0221] Determine the neural network corresponding to the data object, and the neural network is obtained by training with sample data;

[0222] Use the neural network to analyze and process the data object to obtain the object information corresponding to the data object.

[0223] Wherein, the structure of the electronic device may further include a third communication interface 63 for the electronic device to communicate with other devices or communication networks.

[0224] In addition, an embodiment of the present invention provides a computer storage medium for storing computer software instructions used by an electronic device, which includes a program for executing the data processing method involved in the method embodiment shown above. Figure 7 - 8 The program involved in the data processing method in the above-mentioned method embodiment.

[0225] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed to multiple network units. Some or all of the modules can be selected according to actual needs to achieve the purpose of the solution of this embodiment. Those of ordinary skill in the art can understand and implement it without creative labor.

[0226] Through the description of the above embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of adding a necessary general hardware platform, and of course, it can also be implemented by a combination of hardware and software. Based on such an understanding, the above technical solution essentially or the part that contributes to the prior art can be embodied in the form of a computer product. The present invention can be implemented in the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk memories, CD-ROMs, optical memories, etc.) containing computer-usable program codes.

[0227] The present invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It should be understood that each flow and / or block in the flowchart illustrations and / or block diagrams, and combinations of flows and / or blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions may be provided to a processor of a general purpose computer, special purpose computer, embedded processor, or other programmable device to produce a machine, such that the instructions executed by the processor of the computer or other programmable device create means for implementing the functions specified in one or more flows of the flowchart and / or one or more blocks of the block diagram.

[0228] These computer program instructions may also be stored in a computer-readable memory that can direct a computer or other programmable device to function in a particular manner, such that the instructions stored in the computer-readable memory produce an article of manufacture including instruction means that implement the functions specified in one or more flows of the flowchart and / or one or more blocks of the block diagram.

[0229] These computer program instructions may also be loaded onto a computer or other programmable device, such that a series of operational steps are performed on the computer or other programmable device to produce a computer-implemented process, and thus the instructions executed on the computer or other programmable device provide steps for implementing the functions specified in one or more flows of the flowchart and / or one or more blocks of the block diagram.

[0230] In a typical configuration, a computing device includes one or more processors (CPUs), an input / output interface, a network interface, and memory.

[0231] The memory may include non-permanent memory in the form of computer-readable media, random access memory (RAM), and / or non-volatile memory, such as read-only memory (ROM) or flash RAM. The memory is an example of computer-readable media.

[0232] A computer-readable medium includes permanent and non-permanent, removable and non-removable media that can implement information storage by any method or technology. The information can be computer-readable instructions, data structures, program modules, or other data. Examples of computer storage media include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technologies, compact disc read-only memory (CD-ROM), digital versatile disc (DVD) or other optical storage, magnetic cassettes, magnetic tape magnetic disk storage or other magnetic storage devices, or any other non-transitory medium that can be used to store information accessible by a computing device. As defined herein, a computer-readable medium does not include transitory computer-readable media, such as modulated data signals and carrier waves.

[0233] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that they can still modify the technical solutions recorded in the foregoing embodiments, or perform equivalent replacements for some of the technical features; and these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for auditing data, characterized in that, Including: Determine at least one target selection function from a preset plurality of selection functions, each of the selection functions being used to evaluate the value of the sample data to be labeled, so as to select the sample data to be labeled that is adapted to the application scenario in different application scenarios; Determine at least one target sample data from a plurality of sample data to be labeled according to the target selection function; Determine the labeled sample data corresponding to the target sample data; Use the labeled sample data to optimize and train a preset audit model, and use the optimized and trained audit model to audit the data to be processed; Wherein, determining at least one target sample data from a plurality of sample data to be labeled according to the target selection function includes: Determine the sample selection probability corresponding to each of the sample data according to the target selection function; Determine at least one sample data corresponding to the maximum sample selection probability as the target sample data.

2. The method according to claim 1, wherein Determining at least one target selection function from a preset plurality of selection functions includes: Using the audit model, obtain the estimated accuracy corresponding to each selection function; Determine the selection function corresponding to the maximum estimated accuracy as the target selection function.

3. The method according to claim 1, characterized in that, Before determining at least one target selection function from a preset plurality of selection functions, the method further includes: Configure a selection probability and a minimum selection probability for each of the plurality of selection functions, and the minimum selection probability is greater than 0.

4. The method according to claim 3, wherein After using the labeled sample data to optimize and train a preset audit model, the method further includes: Obtain the model feedback information of the audit model; Adjust the selection probability of the target selection function according to the model feedback information.

5. The method according to claim 4, wherein Adjusting the selection probability of the target selection function according to the model feedback information includes: When the model feedback information is positive feedback, increase the selection probability of the target selection function; or, When the model feedback information is negative feedback, decrease the selection probability of the target selection function.

6. The method according to claim 1, characterized in that, Determining the sample selection probability corresponding to each sample data according to the target selection function includes: Determine the to-be-selected probability that each sample data is selected by all selection functions; Determine the sum of the products of the to-be-selected probability of each sample data and the selection probability of the target selection function as the sample selection probability corresponding to each sample data.

7. The method according to any one of claims 1-6, characterized in that, When there are multiple target sample data, the method further includes: Screen the multiple target sample data to obtain the screened sample data.

8. A training method for a model, characterized in that, Including: Determine at least one target selection function from a preset plurality of selection functions, each of the selection functions being used to evaluate the value of the sample data to be labeled, so as to select the sample data to be labeled that is adapted to the application scenario in different application scenarios; Determine at least one target sample data from a plurality of sample data to be labeled according to the target selection function; Determine the labeled sample data corresponding to the target sample data; Use the labeled sample data to optimize and train a preset audit model; Among them, determining at least one target sample data from multiple sample data to be labeled according to the target selection function includes: Determining the sample selection probability corresponding to each of the sample data according to the target selection function; Determining at least one sample data corresponding to the maximum sample selection probability as the target sample data.

9. An auditing device for data, characterized in that, Including: A first selection module, configured to determine at least one target selection function from a plurality of preset selection functions, where each selection function is used to evaluate the value of the sample data to be labeled, so as to select the sample data to be labeled that is adapted to the application scenario in different application scenarios; A first determination module, configured to determine at least one target sample data from multiple sample data to be labeled according to the target selection function; The first determination module is further configured to determine the labeled sample data corresponding to the target sample data; An audit module, configured to optimize and train a preset audit model by using the labeled sample data, and audit the data to be processed by using the optimized and trained audit model; The first determination module is further configured to: determine the sample selection probability corresponding to each of the sample data according to the target selection function; determine at least one sample data corresponding to the maximum sample selection probability as the target sample data.

10. An electronic device, characterized in that, Including: A memory and a processor; wherein, the memory is used to store one or more computer instructions, and when the one or more computer instructions are executed by the processor, the data auditing method according to any one of claims 1 to 7 is implemented.

11. A training device for a model, characterized in that, Including: A second selection module, configured to determine at least one target selection function from a plurality of preset selection functions, where each selection function is used to evaluate the value of the sample data to be labeled, so as to select the sample data to be labeled that is adapted to the application scenario in different application scenarios; A second determination module, configured to determine at least one target sample data from multiple sample data to be labeled according to the target selection function; The second determination module is further configured to determine the labeled sample data corresponding to the target sample data; A training module, configured to optimize and train a preset audit model by using the labeled sample data; The second determination module is further configured to: determine the sample selection probability corresponding to each of the sample data according to the target selection function; determine at least one sample data corresponding to the maximum sample selection probability as the target sample data.

12. An electronic device, characterized in that, Including: A memory and a processor; wherein, the memory is used to store one or more computer instructions, and when the one or more computer instructions are executed by the processor, the model training method according to claim 8 is implemented.

Images