Data processing methods, computer storage media, electronic devices and computer program products

Abstract

This application provides a data processing method, a computer storage medium, an electronic device, and a computer program product. It involves receiving an information review instruction from a server based on exposure information of a target object; then, according to the information review instruction, invoking a client-side large model configured on the client side for information review of the target object. This client-side large model is obtained by loading basic data resources based on the client's configuration information; finally, the client-side large model performs information review on the target object, and based on the review results, determines the exposure processing method for the target object. This solution allows for further information review of the target object on the client side after exposure, using the client-side large model to determine the exposure processing method based on the review results. This addresses any missed reviews before exposure and effectively prevents the large-scale exposure of risky information.

Description

Technical Field

[0001] This application relates to the field of computer technology, and in particular to a data processing method, a computer storage medium, an electronic device, and a computer program product. Background Technology

[0002] With the continuous development of Internet technology, users are increasingly using the Internet and various computer applications in their daily work and life. Through these applications, users can perform corresponding operations, such as working, watching audio and video, reading e-books, and browsing online information. At the same time, they can also see other information embedded by the application's platform or partners, including but not limited to advertising information.

[0003] To ensure the compliance of this information, it is currently mostly reviewed manually before being displayed. However, this review method is costly due to the large amount of manpower required; moreover, it often results in the omission of risky information, leading to the widespread exposure of such information. Summary of the Invention

[0004] In view of this, embodiments of this application provide a data processing scheme to at least partially solve the above-mentioned problems.

[0005] According to a first aspect of the embodiments of this application, a data processing method is provided, applied to a client. The method of this embodiment includes: receiving an information review instruction for a target object sent by a server based on the exposure information of the target object; invoking a client-side large model set on the client for information review of the target object according to the information review instruction, wherein the client-side large model is obtained by loading basic data resources according to the client's configuration information; performing information review on the target object through the client-side large model, and determining the exposure processing method for the target object based on the information review result.

[0006] According to a second aspect of the embodiments of this application, a data processing method is provided, applied to a server. The method of this embodiment includes: obtaining configuration information reported by a client, wherein the configuration information is used to instruct the client to dynamically load a client configuration of a large client-side model, and the large client-side model is used to perform information verification on a target object on the client; based on the configuration information, dynamically determining the basic data resources of the corresponding large client-side model for the client, and sending the basic data resources to the client, so that the client can obtain the large client-side model by loading the basic data resources, and perform information verification on the target object through the large client-side model.

[0007] According to a third aspect of the embodiments of this application, an electronic device is provided, including: a processor, a communication interface, a memory, and a communication bus, wherein the processor, the communication interface, and the memory communicate with each other through the communication bus; the memory is used to store a computer program; and the processor is used to execute the method described in the first or second aspect by running the computer program stored in the memory.

[0008] According to a fourth aspect of the embodiments of this application, a computer storage medium is provided having a computer program stored thereon, which, when executed by a processor, implements the method described in the first or second aspect.

[0009] According to a fifth aspect of the embodiments of this application, a computer program product is provided, including a computer program that, when executed by a processor, implements the method as described in the first or second aspect.

[0010] According to the data processing scheme provided in this application embodiment, the system receives an information review instruction for a target object sent by the server based on the exposure information of the target object; then, according to the information review instruction, it calls a client-side large model set up on the client for information review of the target object. This client-side large model is obtained by loading basic data resources based on the client's configuration information; finally, the client-side large model performs information review on the target object, and the exposure processing method for the target object is determined based on the review results. Therefore, through this data processing method, after the target object is exposed, the client can further review the target object using the client-side large model, and determine the exposure processing method based on the review results. This can compensate for any missed reviews before the target object was exposed, effectively preventing the large-scale exposure of risky information. Furthermore, the client-side large model set up on the client is obtained by loading basic data resources based on the client's configuration information, ensuring that the client's configuration meets the operating conditions of the client-side large model, thereby improving the efficiency of information review by the client-side large model without requiring additional hardware investment. In addition, using the client-side large model to review the target object can also reduce labor costs. Attached Figure Description

[0011] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in the embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings.

[0012] Figure 1This is a schematic diagram of a data processing system according to an embodiment of this application.

[0013] Figure 2 This is a flowchart of the steps of a data processing method according to an embodiment of this application.

[0014] Figure 3 This is a flowchart illustrating the steps of another data processing method according to an embodiment of this application.

[0015] Figure 4 This is a schematic diagram illustrating an example of a data processing scheme according to an embodiment of this application.

[0016] Figure 5 This is a schematic diagram of the structure of an electronic device according to an embodiment of this application. Detailed Implementation

[0017] To enable those skilled in the art to better understand the technical solutions in the embodiments of this application, the technical solutions in the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art should fall within the protection scope of the embodiments of this application.

[0018] The specific implementation of the embodiments of this application will be further described below with reference to the accompanying drawings.

[0019] Figure 1 An exemplary system applicable to embodiments of this application is shown. For example... Figure 1 As shown, the system 100 may include a server 102, a communication network 104, and / or one or more clients 106. Figure 1 The example shows multiple clients 106, each with an application set up to display the target object.

[0020] Server 102 can be any suitable device for storing information, data, programs, and / or any other suitable type of content, including but not limited to distributed storage system devices, server clusters, computing server clusters, etc. In some embodiments, server 102 can perform any suitable function. For example, when server 102 implements the scheme of the embodiments of this application, in some embodiments, server 102 can be used to perform data processing methods. As an optional example, in some embodiments, server 102 can first obtain configuration information reported by the client, the configuration information being used to instruct the client to dynamically load the client configuration of the client-side large model, the client-side large model being used to review information on the target object; then, based on the configuration information, dynamically determine the basic data resources of the corresponding client-side large model for the client, and send the basic data resources to the client, so that the client can obtain the client-side large model by loading the basic data resources, and review the target object through the client-side large model to determine the exposure processing method for the target object. In some embodiments, server 102 may receive configuration information sent by client 106, and after dynamically determining the basic data resources of the corresponding edge-side large model for the client in the manner described above, send the basic data resources to client 106 so that client 106 can obtain the edge-side large model by loading the basic data resources, and perform information review on the target object through the edge-side large model to determine the exposure processing method for the target object.

[0021] In some embodiments, communication network 104 can be any suitable combination of one or more wired and / or wireless networks. For example, communication network 104 can include any one or more of the following: the Internet, intranet, wide area network (WAN), local area network (LAN), wireless network, digital subscriber line (DSL) network, frame relay network, asynchronous transfer mode (ATM) network, virtual private network (VPN), and / or any other suitable communication network. Client 106 can connect to communication network 104 via one or more communication links (e.g., communication link 112), which can be linked to server 102 via one or more communication links (e.g., communication link 114). Communication links can be any communication link suitable for transmitting data between client 106 and server 102, such as network links, dial-up links, wireless links, hardwired links, any other suitable communication links, or any suitable combination of such links.

[0022] Optionally, client 106 may be configured with an application for displaying the target object. Client 106 may include any one or more clients suitable for displaying the target object and interacting with the user. In some embodiments, client 106 may include any suitable type of device. For example, in some embodiments, client 106 may include a mobile device, tablet computer, laptop computer, desktop computer, and / or any other suitable type of client. In some embodiments, client 106 may perform any appropriate function. For example, when the solution of the embodiments of this application is implemented by client 106, in some embodiments, client 106 may be used to perform a data processing method. As an optional example, in some embodiments, client 106 may first receive an information review instruction for the target object sent by the server based on the exposure information of the target object; then, according to the information review instruction, call the client-side large model set on the client for information review of the target object, wherein the client-side large model is obtained by loading basic data resources according to the client's configuration information; finally, the client-side large model performs information review on the target object, and determines the exposure processing method for the target object based on the information review result.

[0023] Based on the above system, this application provides a data processing scheme, which will be described below through several embodiments.

[0024] Figure 2 This is a flowchart illustrating the steps of a data processing method according to an embodiment of this application. According to a first aspect of this application, a data processing method is provided, which is applied to a client. (Refer to...) Figure 2 As shown, the method includes steps S202, S204, and S206, specifically:

[0025] S202: Receive information review instructions for the target object sent by the server based on the exposure information of the target object.

[0026] In this embodiment, the target object can refer to any form of object that can be displayed to the user. For example, the target object can be a text / image object, a video object, an audio object, etc. The exposure information of the target object can be information indicating the exposure status of the target object, such as exposure volume. The information review instruction for the target object is used to instruct the client to perform information review of the target object.

[0027] The server can generate an information review instruction for the target object based on its exposure information and preset exposure conditions, and send the instruction to the client. Optionally, if the target object's exposure information meets the preset exposure conditions, an information review instruction is generated and sent to the client. Optionally, the target object's exposure information can be the target object's exposure volume, and the preset exposure condition can be that the target object's exposure volume is greater than or equal to an exposure volume threshold. That is, if the target object's exposure volume is greater than or equal to the exposure volume threshold, an information review instruction is generated. The exposure volume threshold can be flexibly set by those skilled in the art according to actual conditions, and this embodiment does not impose any limitations on it.

[0028] S204. According to the information review instruction, call the client-side large model set up on the client side for information review of the target object.

[0029] The client-side large model is obtained by loading basic data resources based on the client's configuration information. It can be implemented as an appropriate machine learning model capable of verifying target object information on the client side. In one example, it can be an LLM (Large Language Model) model.

[0030] The client's configuration information is used to indicate the client's configuration for dynamically loading large client-side models. Optionally, the client's configuration information may include available resource information, system locale information, and information about resource files and / or loaded class files of the large client-side models used in the past.

[0031] In this embodiment, the basic data resource refers to the portion of data resources in the pre-set client-side large model on the server that is used for information verification of the target object and matches the client's configuration information. This portion of data resources is a subset of the entire large model resources. The client can pre-load the basic data resources to obtain the client-side large model based on the client's configuration information. After receiving the information verification instruction, the client calls the pre-loaded client-side large model based on the information verification instruction.

[0032] In some optional embodiments, before receiving the information review instruction for the target object sent by the server based on the exposure information of the target object, the method of this embodiment further includes: obtaining basic data resources dynamically determined based on the client's configuration information from the server, wherein the configuration information is sent by the client to the server, and the configuration information includes at least one of the following: available resource information, system language environment information; and loading basic data resources.

[0033] For example, available resource information is used to indicate the device resource information of the client that can dynamically load large models on the client side, such as storage space, GPU computing power resources, CPU computing power resources, etc. The basic data resources determined by this information can ensure smooth operation on the client side.

[0034] System language environment information indicates the client's system language environment. This information can be configuration information within the operating system used to determine the language used for the system interface, text display, input / output, and related language settings. The basic data resources determined by this information do not need to consider resources in other language environments, significantly reducing the amount of basic data resources.

[0035] For example, basic data resources may include resource files, class files, language data packages, etc. of the large model on the client side.

[0036] The client can pre-send configuration information to the server. The server can then dynamically determine the underlying data resources that match the client's configuration information. For example, it can dynamically determine the underlying data resources based on the client's available resource information; or it can dynamically determine the underlying data resources based on the client's system language environment information; or it can dynamically determine the underlying data resources that match both the client's available resource information and system language environment information. This embodiment does not impose any limitations on this. Afterwards, the client can obtain the dynamically determined underlying data resources based on the client's configuration information from the server and load them.

[0037] In this embodiment, the client obtains basic data resources dynamically determined based on available resource information and / or system language environment information, and loads the basic data resources to obtain the client's terminal-side large model. This ensures that the client's configuration meets the operating conditions of the terminal-side large model, thereby improving the efficiency of the terminal-side large model in information review.

[0038] In some optional embodiments, the configuration information also includes information about resource files of the client-side large model that have been used historically, and / or information about loaded class files. Based on this, obtaining client-side configuration information from the server and dynamically determining the basic data resources can be achieved by: receiving at least one of the resource files, class files, and language data packets of the client-side large model, dynamically determined according to the configuration information, returned by the server; wherein the determined resource files match the information in the configuration information, the determined class files match the information in the configuration information, and the language data packets match the system language environment information. Through this information, resources that are highly likely to be used by the client can be effectively determined, while resources with low probability of use can be excluded, further reducing the amount of basic data resources sent to the client.

[0039] For example, the resource files of a large edge model may include model files, configuration files, vocabulary files, and word segmenter files. The model file contains information such as the model's pre-training weights. The configuration file defines the model's architecture and hyperparameter settings during training. The vocabulary file provides the model with the vocabulary information needed to process text. The word segmenter file contains configuration information for the word segmenter.

[0040] A class file refers to various classes defined in a programming language, used to implement different functions and modules of the model. Class files can include model architecture classes, data processing classes, training classes, evaluation classes, etc. The model architecture class defines the overall architecture of the model, such as the number of layers in the neural network, the number of neurons, activation functions, and attention mechanisms. The data processing class is responsible for data reading, preprocessing, word segmentation, encoding, and other operations. The training class implements the model's training logic, including defining loss functions, optimizers, learning rate schedulers, executing training loops, and updating model parameters. The evaluation class evaluates the trained model and calculates various evaluation metrics.

[0041] Based on this, in one example, the resource file information includes the type and identification information of resource files of the historically used client-side large model, and the class file information can include the type and identification information of loaded class files. Optionally, the client can obtain all resource files and class files of the client-side large model, and statistically analyze the historical usage of all resource files and the historical loading of class files. Unused resource files refer to resource files that are not referenced in the program, and unloaded class files refer to components whose classes have not been loaded during program execution. Based on the statistical results, information on historically used client-side large model resource files and / or loaded class files is filtered and obtained.

[0042] The client can send configuration information to the server. The server can dynamically determine the resource files of the large-scale client model based on information about previously used resource files in the client's database; and / or, based on information about loaded class files, dynamically determine the class files of the large-scale client model. Furthermore, it can dynamically determine the language data package of the large-scale client model to be loaded based on the client's system locale information. Afterwards, the client can receive at least one of the following from the server: the resource files, class files, and the language data package matching the system locale information, and load it.

[0043] In this embodiment, the client can receive at least one of the following: resource files, class files, and language data packets that match the system language environment information of the large client-side model dynamically determined by the server according to the configuration information. That is, the client can obtain the data resources of the large client-side model that match the client's configuration information, which reduces the amount of data resources obtained and thus reduces the size of the loaded large client-side model.

[0044] In some optional embodiments, when the configuration information includes available resource information, the method of this embodiment further includes: determining whether the available resources indicated by the available resource information of the client meet the basic data resource loading requirements of the large model on the client side; if they meet the requirements, then obtaining the basic data resources dynamically determined based on the configuration information of the client from the server.

[0045] For example, based on the client's available resource information and the basic data resource loading requirements of the large-scale client-side model, it is determined whether the client's available resources meet the loading requirements of the large-scale client-side model. For instance, it is determined whether the client's storage space meets the loading requirements, whether the client's GPU meets the loading requirements, and so on. If it is determined that the client's available resources meet the loading requirements of the large-scale client-side model, then the basic data resources, dynamically determined based on the client's configuration information, are obtained from the server and loaded. The basic data resources include at least one of the following: resource files of the large-scale client-side model, class files, and language data packages matching the system's language environment information.

[0046] In this embodiment, before acquiring the basic data resources of the large-scale client model, it is pre-determined whether the available resources indicated by the available resource information of the client meet the basic data resource loading requirements of the large-scale client model. This ensures that the client configuration meets the operating conditions of the large-scale client model and avoids the adverse effects of directly loading basic data resources on the normal operation of the client.

[0047] S206. Review the information of the target object through the large-scale model on the edge, and determine the exposure processing method for the target object based on the results of the information review.

[0048] This embodiment uses a large-scale model on the edge to review the information of the target object, determining whether it contains preset information. For example, the preset information can be risk information, such as information that does not comply with established laws and regulations. If it is included, the target object is considered a risk object; if not, it is considered a qualified object. The result of the information review indicates whether the target object is a risk object. The result can include a review score for the target object, and multiple score ranges can be preset, such as a risk score range for a risk object and a score range for an object awaiting review. Furthermore, the result can also include exposure information for the target object, and multiple exposure ranges can be preset, such as a first exposure range for a risk object and a second exposure range for an object awaiting review.

[0049] For example, if the target object's review score falls within the risk score range and / or the target object's exposure information falls within the first exposure range, then the target object is considered a risky object. For risky objects, a large client-side model can be used to select at least one candidate object from multiple pre-acquired candidate objects. The selected at least one candidate object is considered a qualified object. Then, the selected at least one candidate object replaces the target object and is displayed to the user on the client side. If the target object's review score falls within the score range of the object awaiting review and the target object's exposure information falls within the second exposure range, then the target object is considered a subject awaiting review. For objects awaiting review, a report can be submitted to the server for further evaluation.

[0050] In some optional embodiments, the target object is a multimodal target object; information review of the target object through the edge-side large model includes: performing multimodal information detection on the target object through the edge-side large model to obtain multimodal information; performing structured processing on the multimodal information; and performing information review on the target object based on the results of the structured processing.

[0051] For example, the target object is a multimodal target object, meaning it can be a text / image object, a video object, an audio object, etc. Multimodal information detection is performed on the target object using a large-scale model on the edge, identifying the multimodal information contained within it. This multimodal information can include image information, text information, audio information, etc. If the target object is a video object, frames can be extracted from the video object to obtain a text / image object, and the audio in the video object can be converted into an audio object. Multimodal information detection is then performed on both the text / image object and the audio object. Next, the multimodal information undergoes structured processing. For example, data cleaning is performed to remove noise, duplicate, or invalid information; semantic relationships are constructed for the various entities and concepts contained in the multimodal information; natural language generation techniques or specific structured description templates are used to convert the multimodal information into a structured text description, and so on. Based on the results of the structured processing, pre-acquired structured review rules are invoked to review the results, thus achieving information review of the target object.

[0052] In this embodiment, multimodal information detection of the target object is performed by a large edge model to obtain multimodal information. The multimodal information is then structured to obtain the information contained in the target object in a comprehensive and accurate manner. Based on the results of the structured processing, the target object is then reviewed, which can improve the accuracy of the information review results. This can effectively avoid the large-scale exposure of risky information, while ensuring the exposure rate of target objects that do not contain risky information.

[0053] In some optional embodiments, the information review of the target object is performed by the client-side large model, including: instructing the client-side large model to call the review rules pre-stored by the client, so that the client-side large model can review the information of the target object according to the review rules.

[0054] For example, the client can pre-obtain and store the review rules from the server. These review rules can be obtained as follows: based on object categories, objects from at least one of different forms, platforms, and scenarios are classified by risk type; based on the risk type classification results, structured review rules are generated for each object category. That is, the review rules include different structured review rules corresponding to different risk type classifications. For example, for an advertisement object classified as financial, the structured review rule could be that if the advertisement object contains words such as "0 risk," "zero risk," or "no risk," then the advertisement object is a risky object.

[0055] In this embodiment, the review rules can be stored in advance on the client side and can be directly called when the large model on the client side is used for information review, which reduces the consumption of computing resources on the client side.

[0056] In some optional embodiments, the exposure processing method for the target object is determined based on the information review result, including: if the information review result indicates that the target object is a risk object, then at least one candidate object is selected from multiple candidate objects through the edge big model, and the selected at least one candidate object is used to replace the target object.

[0057] For example, the multiple candidate objects can be objects pre-obtained by the client to replace the target object. Taking the target object as the target advertisement as an example, the multiple candidate objects are multiple candidate advertisements. If the information review result indicates that the target advertisement is a risky object, that is, the target object contains risky information, then the client-side large model performs information review on the multiple candidate advertisements, filters out the risky objects among the multiple candidate advertisements, leaving at least one candidate advertisement, and selects at least one candidate advertisement from the remaining target advertisements to replace the target advertisement, and displays it to the user on the client. As for the target advertisement, it can be reported to the server for further review, such as manual review, etc., and this embodiment does not limit this.

[0058] In this embodiment, when the information review result indicates that the target object is a risk object, at least one candidate object is selected from multiple candidate objects to replace the target object through the large-scale model on the edge. This can reduce the exposure of the risk object in a timely manner. Furthermore, by selecting the candidate object to replace the target object through the large-scale model on the edge, the situation where the candidate object is a risk object is avoided.

[0059] In summary, the data processing scheme provided in this application embodiment can further review the target object's information on the client side using a large-scale client-side model after the target object has been exposed. Based on the review results, the exposure processing method for the target object is determined, thus mitigating any missed reviews before exposure and effectively preventing the large-scale exposure of risky information. Furthermore, the large-scale client-side model is loaded with basic data resources based on the client's configuration information, ensuring that the client's configuration meets the operating conditions of the large-scale client-side model. This improves the efficiency of information review by the large-scale client-side model without requiring additional hardware investment. Additionally, using a large-scale client-side model for information review of the target object can reduce labor costs.

[0060] Figure 3 This is a flowchart illustrating the steps of a data processing method according to an embodiment of this application. According to a second aspect of this application, a data processing method is provided, which is applied to a server communicating with the aforementioned client. (Refer to...) Figure 3 As shown, the method includes steps S302 and S304, specifically:

[0061] S302. Obtain the configuration information reported by the client.

[0062] The configuration information is used to instruct the client to dynamically load the client configuration of the client-side large model, which is used to review the information of the target object on the client side.

[0063] For example, the client's configuration information may include available resource information, system language environment information, etc. Available resource information is used to indicate the device resource information that the client can dynamically load large models from the client side, such as storage space, GPU resources, CPU resources, etc. System language environment information refers to information used to indicate the client's system language environment, which refers to the configuration information in the operating system used to determine the language used for system interface, text display, input and output, and related language settings.

[0064] In some optional embodiments, before obtaining the configuration information reported by the client, the method of this embodiment further includes: obtaining historical audit samples and performing data augmentation on the historical audit samples to obtain training samples for training the original client-side large model on the server; using the training samples to train the original client-side large model to obtain a client-side large model that can be dynamically loaded by the client.

[0065] For example, historical review samples can be obtained, which may include risk information such as sensitive words. The historical review samples will undergo data cleaning, followed by data augmentation processing. Taking sensitive words as an example, broad triggering processing can be applied, such as reverse order processing, synonym replacement processing, and superscript / hyperscript replacement processing. This embodiment does not impose limitations on this. The processed data samples are used as training samples to train the original client-side large model on the server. After training, a client-side large model that can be dynamically loaded by the client can be obtained.

[0066] In this embodiment, data augmentation is performed on historical review samples to obtain training samples for training the original client-side large model, thereby improving the comprehensiveness of the training samples. Then, the training samples are used to train the original client-side large model to obtain a client-side large model that can be dynamically loaded by the client, which improves the accuracy of the client-side large model in information review.

[0067] In some optional embodiments, before obtaining the configuration information reported by the client, the method of this embodiment further includes: classifying objects from at least one of different forms, platforms, and scenarios into risk categories based on object categories; generating structured audit rules for object categories based on the results of risk category classification; and sending the structured audit rules to the client for the client to call when using the client-side large model for information auditing.

[0068] For example, an object category refers to the type of object. Objects can be of different forms, such as text / image objects, video objects, or audio objects. Objects can also come from different platforms, such as different ad networks for an ad object. Alternatively, objects can be objects in different scenarios, such as ad placement scenarios for an ad object. An ad placement scenario refers to the specific location, environment, and various related conditions and factors of an ad display. Online ad placement scenarios can include search engine ad placement scenarios, social media ad placement scenarios, video platform ad placement scenarios, e-commerce platform ad placement scenarios, reading app ad placement scenarios, and gaming app ad placement scenarios. Taking an object as an ad object as an example, based on the ad category, objects from different ad placement scenarios can be classified by risk type to obtain risk classification results. For example, risk classification results can include financial, gaming, etc. Based on the risk classification results, structured audit rules can be generated for that object category. That is, different structured audit rules are generated for different risk classifications. For example, for an advertisement object whose risk classification result is financial, the structured audit rule can be that if the advertisement object contains words such as "0 risk", "zero risk", or "no risk", then the advertisement object is a risk object.

[0069] In this embodiment, structured review rules for object categories can be generated in advance on the server side and sent to the client. When the client uses the client-side large model for information review, it can directly call the rules, which reduces the computational resources consumed by the client.

[0070] S304. Based on the configuration information, dynamically determine the basic data resources of the corresponding client-side large model for the client, and send the basic data resources to the client so that the client can obtain the client-side large model by loading the basic data resources, and perform information verification on the target object through the client-side large model.

[0071] The server can dynamically determine the basic data resources to be loaded that match the client's configuration information. For example, it can dynamically determine the basic data resources to be loaded that match the client's available resource information; or, it can dynamically determine the basic data resources to be loaded that match the client's system language environment information; or, it can dynamically determine the basic data resources to be loaded that match both the client's available resource information and system language environment information. This embodiment does not impose any limitations on this. Afterwards, the server sends the basic data resources to the client. The client loads the basic data resources to obtain the client-side large model, performs information verification on the target object using the client-side large model, and determines the exposure processing method for the target object based on the information verification result.

[0072] In some optional embodiments, the configuration information includes available resource information of the client; based on the configuration information, dynamically determining the basic data resources of the corresponding large-scale client-side model for the client includes: dynamically determining the basic data resources of the corresponding large-scale client-side model for the client based on the available resource information.

[0073] For example, basic data resources may include resource files, class files, etc. of the large client-side model. The resource files and class files of the large client-side model have been described in the foregoing embodiments and will not be repeated here.

[0074] When the server receives available resource information reported by the client, it can determine whether the available resources indicated by the client's available resource information meet the basic data resource loading requirements of the client-side large model. If they do, the server dynamically determines the corresponding basic data resources of the client-side large model for the client. Optionally, the server can obtain all resource files and class files of the preset client-side large model, and statistically analyze the historical usage (used by the client) of all resource files and the historical loading (loaded by the client) of class files. Based on the statistical results, the server filters and obtains the information of resource files and / or class files that have been used and loaded in the past. Then, based on the information of resource files and / or class files, the server determines the resource files and / or class files corresponding to the client from all resource files and class files of the preset client-side large model and sends them to the client so that the client can load the basic data resources of the client-side large model.

[0075] In this embodiment, based on the available resource information of the client, the basic data resources of the corresponding large-scale client-side model are dynamically determined for the client, which reduces the amount of data resources acquired by the client and thus reduces the size of the large-scale client-side model loaded by the client.

[0076] In some optional embodiments, the configuration information also includes the client's system language environment information; based on the available resource information, dynamically determining the basic data resources of the corresponding client-side large model for the client also includes: dynamically determining the basic data resources of the corresponding client-side large model for the client based on the available resource information and the system language environment information.

[0077] For example, the basic data resources may also include language data packages that match the system's locale information.

[0078] The server can determine whether the available resources indicated by the client's configuration information meet the basic data resource loading requirements of the client-side large model, based on the available resource information reported by the client. If so, the server can dynamically determine the resource files and / or class files of the client-side large model. Furthermore, based on the client's system language environment information in the configuration information, the server can determine the language data package that matches that system language environment. Then, the server can send the language data package matching the system language environment information, along with the resource files and / or class files, as the basic data resources to the client, enabling the client to load these basic data resources.

[0079] In this embodiment, based on the system language environment information, basic data resources that match the system language environment information can be determined, eliminating the need to load other data resources that do not match the system language environment information, and further reducing the amount of data resources that the client needs to load.

[0080] In some optional embodiments, the configuration information further includes: information on resource files of the historically used client-side large model and / or information on loaded class files; based on the configuration information, dynamically determine at least one of the resource files, class files, and language data packets of the corresponding client-side large model; wherein the determined resource files match the resource file information in the configuration information, the determined class files match the class file information in the configuration information, and the language data packets match the system language environment information.

[0081] For example, the information of resource files may include the type and identification information of resource files of the historically used client-side large model, and the information of class files may include the type and identification information of loaded class files. Optionally, the client can obtain all resource files and class files of the client-side large model, and statistically analyze the historical usage of all resource files and the historical loading of class files. Unused resource files refer to resource files that are not referenced in the program, and unloaded class files refer to components whose classes have not been loaded during program execution. Based on the statistical results, information of historically used client-side large model resource files and / or loaded class files is filtered and obtained.

[0082] After receiving the configuration information from the client, the server can dynamically determine the resource files of the large-scale client model based on information about previously used resource files in the client's database; and / or, dynamically determine the class files of the large-scale client model based on information about loaded class files. Furthermore, it can dynamically determine the language data package of the large-scale client model to be loaded based on the client's system locale information. Then, it can send at least one of the following to the client: the resource files, class files, and the language data package matching the system locale information.

[0083] In this embodiment, based on the information of resource files and / or loaded class files, combined with configuration information, the basic data resources are dynamically determined for the client. This eliminates the need to load resource files and / or class files that the client has not used, thus reducing the amount of data resources obtained by the client and consequently reducing the size of the large client-side model loaded by the client.

[0084] In some optional embodiments, after sending basic data resources to the client, the method of this embodiment further includes: obtaining information about target objects that the client requests to be exposed and that have not been reviewed, and determining multiple candidate objects corresponding to the target objects; and when receiving a request from the client to request candidate objects, sending at least one candidate object from the multiple candidate objects to the client.

[0085] For example, when the server receives an exposure request for a target object from the client, it can obtain information about the target object and information about multiple candidate objects. Then, after the client reviews the target object using the client-side big data model, if the review result indicates that the target object is a risky object, it sends a request to the server to request candidate objects. Upon receiving this request, the server sends one or more candidate objects from the multiple candidate objects to the client. The client can further review the received candidate objects using the client-side big data model and select at least one candidate object to replace the target object (risky object) currently being exposed by the client.

[0086] In this embodiment, information about target objects that the client requests to be exposed but have not been reviewed is obtained, and multiple candidate objects are determined for the target objects. Then, when a request for candidate objects is received from the client, at least one of the multiple candidate objects can be sent to the client so that the client can replace the risky objects in a timely manner, thereby reducing the exposure of the risky objects.

[0087] Optionally, information about a target object that the client requests to be exposed but has not been reviewed is obtained. The target object's information includes the target object's topic information. If the target object's topic information contains information that conforms to the review rules, then a target candidate object whose topic information does not contain information that conforms to the topic review rules is determined from multiple candidate objects. If the current cumulative exposure of the target candidate object is less than a preset exposure threshold, then the target candidate object is sent to the client to replace the target object for exposure. Alternatively, if the target object's topic information does not contain information that conforms to the topic review rules, and the target object's current cumulative exposure is less than the preset exposure threshold, then the target object is sent to the client for exposure.

[0088] For example, the system receives an exposure request for a target object from a client, obtains information about the target object and information about multiple candidate objects. It determines whether the target object has been reviewed, meaning it has undergone on-device large-scale model review or manual review. If the target object has not been reviewed, it checks whether the target object's topic information contains information that conforms to topic review rules, which may include sensitive words, etc. If the target object's topic information contains information that conforms to topic review rules, it indicates that the target object is a risky object. From the multiple candidate objects, it identifies target candidate objects whose topic information does not contain information that conforms to topic review rules. Then, it compares the current cumulative exposure of the target candidate object with a preset exposure threshold. If the current cumulative exposure of the target candidate object is less than the preset exposure threshold, the target candidate object is sent to the client to replace the target object for exposure. Alternatively, if the target object's topic information does not contain information that conforms to topic review rules, it compares the current cumulative exposure of the target object with the preset exposure threshold. If the current cumulative exposure of the target object is less than the preset exposure threshold, the target object is sent to the client for exposure. The preset exposure threshold can be flexibly set by those skilled in the art according to actual conditions; this embodiment does not limit this.

[0089] In this embodiment, when a client requests exposure of a target object, the server performs a preliminary review of the target object. If the target object is deemed a risky object after the preliminary review, a candidate target object that has passed the preliminary review is used to replace the target object for exposure. If the target object is deemed a qualified object after the preliminary review, the target object is exposed. Utilizing the server for preliminary review further reduces the exposure rate of risky objects and does not consume client resources.

[0090] In some optional embodiments, the method of this embodiment further includes: performing cumulative exposure statistics on objects exposed by multiple different clients, wherein the multiple different clients include clients that report configuration information, and the objects include target objects; if the cumulative exposure reaches a preset exposure threshold, an instruction is sent to the client that reports configuration information to instruct the use of the terminal-side large model to review the information of the target object or target candidate object.

[0091] For example, suppose there are multiple clients that report configuration information and request the server to expose a target object (an unapproved object). While sending the target object to each client that reported configuration information for exposure, the cumulative exposure count of the target object is simultaneously calculated. When the cumulative exposure count of the target object reaches a preset exposure threshold, an instruction is sent to each client that reported configuration information, instructing them to use the client-side large model to review the target object's information. This allows each client to invoke the client-side large model to review the target object's information and determine the appropriate exposure processing method based on the review results.

[0092] In this embodiment, when the cumulative exposure of a target object reaches a preset exposure threshold, the client is instructed to conduct further review of the target object, which can prevent unreviewed risky objects from being exposed in large quantities.

[0093] In summary, the data processing scheme provided in this application embodiment, after obtaining the configuration information reported by the client, dynamically determines the corresponding basic data resources of the client-side large model based on the configuration information, and sends the basic data resources to the client. This allows the client to obtain the client-side large model by loading the basic data resources, and then uses the client-side large model to review the information of the target object to determine the exposure processing method for the target object. Since the client's client-side large model is loaded from the basic data resources determined by the client's configuration information, it ensures that the client's configuration meets the operating conditions of the client-side large model, thereby improving the efficiency of information review by the client-side large model without requiring additional hardware investment. Furthermore, the client's use of the client-side large model to review the information of the target object can also reduce manual costs.

[0094] Below, refer to Figure 4 The illustrated scenario provides an example of the overall implementation process of the data processing scheme in this application embodiment. This overall implementation process can be understood by applying it to a scenario involving information verification of advertising targets. Figure 4 As shown, the client requests an ad from the server, and the server retrieves the target ad and multiple candidate ads. If the target ad has passed the client-side large-scale model review or manual review, it is sent to the client for exposure. If it has not passed review, the topic information of the target ad is reviewed. If the topic information of the target ad does not contain information that meets the topic review rules, and the current cumulative exposure of the target ad is less than a preset exposure threshold, the target ad is sent to the client for exposure on the user interface. Otherwise, a candidate ad that meets the criteria of not containing information that meets the topic review rules and whose current cumulative exposure is less than the preset exposure threshold is selected from the multiple candidate ads and sent to the client for exposure on the user interface. Subsequently, when the current cumulative exposure of the target ad or the candidate ad exposed on the client reaches the preset exposure threshold, an information review instruction for the target ad or the candidate ad is sent to the client.

[0095] Based on the received information review instructions, the client invokes a pre-loaded client-side large model to review the target ad or target candidate ad. Based on the review results, the client determines the appropriate exposure method for the target ad or target candidate ad. For example, if the review indicates that the target ad or target candidate ad is at risk, the client retrieves several candidate ads (ads whose topic information does not contain information that meets the topic review rules and whose current cumulative exposure is less than a preset exposure threshold). The client-side large model selects one candidate ad to replace the target ad and displays it on the client's user interface. Afterward, the replaced target ad or target candidate ad can be reported to the server for further review. If the review is successful, the target ad or target candidate ad is resent to the client for exposure.

[0096] It should be understood that Figure 4 Further details of the overall implementation process can be understood in conjunction with the preceding embodiments, and Figure 4 The overall implementation process shown is merely an example for easy understanding of the embodiments of this application and is not intended to limit the embodiments of this application in any way.

[0097] It is understood that the foregoing description of the data processing method is merely an exemplary description of the embodiments of this application and is not intended to limit the embodiments of this application in any way.

[0098] According to a third aspect of the embodiments of this application, an electronic device is provided, comprising: a processor, a communication interface, a memory, and a communication bus, wherein the processor, the communication interface, and the memory communicate with each other via the communication bus; the memory is used to store a computer program; and the processor is used to execute the data processing method described in the first or second aspect by running the computer program stored in the memory.

[0099] Figure 5 A structural block diagram of an optional electronic device according to an embodiment of this application is shown. This application does not limit the specific implementation of the electronic device 500; however, as an example, reference is made to... Figure 5 The electronic device 500 provided in this application embodiment includes: a processor 502, a communications interface 504, a memory 506, and a communication bus 508. Wherein:

[0100] The processor 502, communication interface 504, and memory 506 communicate with each other via communication bus 508.

[0101] Communication interface 504 is used to communicate with other electronic devices or servers.

[0102] The processor 502 is used to execute the computer program 510, specifically the relevant steps in any of the aforementioned data processing method embodiments.

[0103] Specifically, computer program 510 may include program code that includes computer operation instructions.

[0104] The processor 502 may be a CPU, a GPU (Graphics Processing Unit), an Application Specific Integrated Circuit (ASIC), or one or more integrated circuits configured to implement the embodiments of this application. The one or more processors included in the smart device may be processors of the same type, such as one or more CPUs; or they may be processors of different types, such as one or more CPUs and one or more ASICs.

[0105] Memory 506 is used to store computer program 510. Memory 506 may include high-speed RAM memory, and may also include non-volatile memory, such as at least one disk storage device.

[0106] Specifically, computer program 510 can be used to cause processor 502 to execute the data processing method in any of the foregoing embodiments.

[0107] The specific implementation of each step in computer program 510 can be found in the corresponding descriptions of the steps and units in any of the foregoing data processing method embodiments, and will not be repeated here. Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working process of the devices and modules described above can be referred to the corresponding process descriptions in the foregoing method embodiments, and will not be repeated here.

[0108] The electronic device 500 in this application embodiment has been described in detail in the foregoing data processing method embodiment. Therefore, its related content and beneficial effects can be understood by referring to the above method embodiment, and will not be repeated here.

[0109] According to a fourth aspect of the embodiments of this application, the embodiments of this application also provide a computer storage medium storing a computer program thereon, which, when executed by a processor, implements the data processing method described in any of the foregoing method embodiments. The computer storage medium includes, but is not limited to, compact disc read-only memory (CD-ROM), random access memory (RAM), floppy disk, hard disk, or magneto-optical disk, etc.

[0110] According to a fifth aspect of the embodiments of this application, the embodiments of this application also provide a computer program product, including a computer program that, when executed by a processor, implements the data processing method as described in any of the embodiments of the plurality of method embodiments described above.

[0111] The electronic device 500 / computer storage medium / computer program product embodiment in this application has been described in detail in the foregoing data processing method embodiment. Therefore, its related content and beneficial effects can be understood by referring to the above method embodiment, and will not be repeated here.

[0112] Furthermore, it should be noted that the user-related information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to sample data used for training the model, data used for analysis, stored data, displayed data, etc.) involved in the embodiments of this application are all information and data authorized by the user or fully authorized by all parties. Moreover, the collection, use and processing of related data must comply with the relevant laws, regulations and standards of the relevant countries and regions, and corresponding operation entry points are provided for users to choose to authorize or refuse.

[0113] It should be noted that, depending on the implementation needs, the various components / steps described in the embodiments of this application can be broken down into more components / steps, or two or more components / steps or parts of the operation of components / steps can be combined into new components / steps to achieve the purpose of the embodiments of this application.

[0114] The methods described in the embodiments of this application can be implemented in hardware, firmware, or as software or computer code that can be stored in a recording medium (such as a CD-ROM, RAM, floppy disk, hard disk, or magneto-optical disk), or as computer code downloaded over a network that is originally stored in a remote recording medium or a non-transitory machine-readable medium and will be stored in a local recording medium. Thus, the methods described herein can be stored on a recording medium using a general-purpose computer, a dedicated processor, or programmable or dedicated hardware (such as an Application Specific Integrated Circuit (ASIC) or a Field Programmable Gate Array (FPGA)). It is understood that the computer, processor, microprocessor controller, or programmable hardware includes storage components (e.g., Random Access Memory (RAM), Read-Only Memory (ROM), Flash Memory, etc.) capable of storing or receiving software or computer code, which, when accessed and executed by the computer, processor, or hardware, implements the methods described herein. Furthermore, when a general-purpose computer accesses code used to implement the methods shown herein, the execution of the code transforms the general-purpose computer into a dedicated computer for executing the methods shown herein.

[0115] Those skilled in the art will recognize that the units and method steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for specific applications, but such implementations should not be considered beyond the scope of the embodiments of this application.

[0116] The term "comprising" and its variations as used herein are open-ended, meaning "including but not limited to". The term "based on" means "at least partially based on". The term "one embodiment" means "at least one embodiment"; the term "another embodiment" means "at least one additional embodiment"; the term "some embodiments" means "at least some embodiments". It should be noted that the concepts of "first", "second", etc., mentioned in the embodiments of this application are only used to distinguish different devices, modules, or units, and are not used to limit the order of functions performed by these devices, modules, or units or their interdependencies. It should be noted that the modifications of "a" and "a plurality" mentioned in the embodiments of this application are illustrative rather than restrictive, and those skilled in the art should understand that unless otherwise expressly indicated in the context, they should be understood as "one or more".

[0117] The above embodiments are only used to illustrate the embodiments of this application, and are not intended to limit the embodiments of this application. Those skilled in the art can make various changes and modifications without departing from the spirit and scope of the embodiments of this application. Therefore, all equivalent technical solutions also fall within the scope of the embodiments of this application, and the patent protection scope of the embodiments of this application should be defined by the claims.

Claims

1. A data processing method applied to a client, the method comprising: Receive information review instructions for the target object sent by the server based on the exposure information of the target object; According to the information review instruction, the client-side large model set on the client for reviewing information of the target object is invoked, wherein the client-side large model is obtained by loading basic data resources according to the configuration information of the client; The target object is reviewed by the large-scale model on the edge, and the exposure processing method for the target object is determined based on the review results.

2. The method according to claim 1, wherein, Before receiving the information review instruction for the target object sent by the receiving server based on the exposure information of the target object, the method further includes: The basic data resources are dynamically determined by obtaining configuration information from the client from the server, wherein the configuration information is sent from the client to the server and includes at least one of the following: available resource information and system language environment information; Load the aforementioned basic data resources.

3. The method according to claim 2, wherein, The configuration information also includes information about the resource files of the client-side large model used in the past, and / or information about the loaded class files; The step of obtaining the basic data resources dynamically determined from the server based on the client's configuration information includes: The system receives at least one of the resource file, class file, and language data packet of the client-side large model, which is dynamically determined based on the configuration information and returned by the server; wherein the determined resource file matches the information of the resource file in the configuration information, the determined class file matches the information of the class file in the configuration information, and the language data packet matches the system language environment information.

4. The method according to claim 3, wherein, When the configuration information includes the available resource information, the method further includes: Determine whether the available resources indicated by the available resource information of the client meet the basic data resource loading requirements of the large model on the client side; If the conditions are met, the basic data resources are dynamically determined based on the configuration information of the client obtained from the server.

5. The method according to claim 1, wherein, The target object is a multimodal target object; The information verification of the target object through the terminal-side large model includes: Multimodal information is obtained by detecting the target object using the large end-side model. The multimodal information is structured, and the target object is audited based on the results of the structured processing.

6. The method according to any one of claims 1-5, wherein, The information verification of the target object through the terminal-side large model includes: The client-side large model is instructed to call the audit rules pre-stored by the client, so that the client-side large model can audit the target object according to the audit rules.

7. The method according to any one of claims 1-5, wherein, The step of determining the exposure processing method for the target object based on the information review results includes: If the result of the information review indicates that the target object is a risk object, then at least one candidate object is selected from multiple candidate objects through the terminal large model, and the selected at least one candidate object is used to replace the target object.

8. A data processing method, applied on a server, the method comprising: Obtain configuration information reported by the client. The configuration information is used to instruct the client to dynamically load the client configuration of the terminal-side large model. The terminal-side large model is used to review the information of the target object on the client. Based on the configuration information, the corresponding basic data resources of the client-side large model are dynamically determined for the client, and the basic data resources are sent to the client so that the client can obtain the client-side large model by loading the basic data resources, and perform information verification on the target object through the client-side large model.

9. The method according to claim 8, wherein, The configuration information includes the available resource information of the client; the step of dynamically determining the basic data resources of the corresponding client-side large model based on the configuration information includes: Based on the available resource information, the corresponding basic data resources of the client-side large model are dynamically determined for the client.

10. The method according to claim 9, wherein, The configuration information also includes the client's system language environment information; the step of dynamically determining the basic data resources of the corresponding client-side large model based on the available resource information includes: Based on the available resource information and the system language environment information, the basic data resources of the corresponding client-side large model are dynamically determined for the client.

11. The method according to any one of claims 8-10, wherein, The configuration information also includes: information on the resource files of the historically used client-side large model and / or information on the loaded class files in the client; The step of dynamically determining the basic data resources of the corresponding client-side large model for the client based on the configuration information includes: dynamically determining at least one of the resource files, class files, and language data packets of the corresponding client-side large model for the client based on the configuration information; wherein the determined resource files match the information of the resource files in the configuration information, the determined class files match the information of the class files in the configuration information, and the language data packets match the system language environment information.

12. The method according to claim 8, wherein, After sending the basic data resources to the client, the method further includes: Obtain information about the target object that the client requests to be exposed but has not been reviewed, and determine multiple candidate objects corresponding to the target object; Upon receiving a request from the client for a candidate object, at least one of the plurality of candidate objects is sent to the client.

13. The method according to claim 12, wherein, The method further includes: The cumulative exposure of objects exposed by multiple different clients is statistically analyzed, wherein the multiple different clients include the client that reported the configuration information, and the objects include the target objects; If the cumulative exposure reaches a preset exposure threshold, an instruction is sent to the client that reported the configuration information to instruct the use of the terminal-side large model to review the information of the target object.

14. The method according to claim 8, wherein, Before obtaining the configuration information reported by the client, the method further includes: Obtain historical review samples and perform data augmentation on the historical review samples to obtain training samples for training the original client-side large model of the server. The original large client-side model is trained using the training samples to obtain a large client-side model that can be dynamically loaded by the client.

15. The method according to claim 8, wherein, Before obtaining the configuration information reported by the client, the method further includes: Based on object categories, risk types are classified for objects from at least one of different forms, platforms, and scenarios; Based on the risk category classification results, structured audit rules for the object category are generated, and the structured audit rules are sent to the client for the client to call when using the client-side large model for information auditing.

16. An electronic device comprising: The processor, the communication interface, the memory, and the communication bus are provided, wherein the processor, the communication interface, and the memory communicate with each other via the communication bus. The memory is used to store computer programs; The processor is configured to perform the method of any one of claims 1-7 by running the computer program stored in the memory, or to perform the method of any one of claims 8-15.

17. A computer storage medium having a computer program stored thereon, which, when executed by a processor, implements the method according to any one of claims 1-7, or implements the method according to any one of claims 8-15.

18. A computer program product comprising computer instructions that instruct a computing device to perform an operation corresponding to the method according to any one of claims 1-7, or to perform an operation corresponding to the method according to any one of claims 8-15.

Images