Risk control service data scheduling method and its device, equipment, medium, product

Abstract

The present application relates to a risk control service data scheduling method, its device, equipment, medium, and product. The method includes: obtaining the storage address of a target picture by the risk control service and transmitting it to the image encoding service; generating a pointer identifier by the image encoding service according to the version identifier of the image encoder it references and the storage address; detecting by the image encoding service whether the pointer identifier exists in the cache list, and when it exists, returning the image feature data pointed to by the pointer identifier; when the pointer identifier does not exist, inputting the picture data of the target picture into the image encoder to obtain the corresponding image feature data and returning it, and mapping and storing the obtained image feature data and the pointer identifier in the cache list; determining by the picture comparison service whether the target picture constitutes an infringing picture in the commodity picture library according to the returned image feature data. The present application can greatly improve the efficiency of picture infringement analysis and significantly improve the response speed of infringement analysis.

Description

Technical Field

[0001] This application relates to information security technology, and particularly to a risk control service data scheduling method, its device, equipment, medium, and product. Background Art

[0002] In the risk control mechanism of an e-commerce platform, infringement analysis is often performed on product images to achieve the purpose of information security control.

[0003] In traditional technologies, since a large amount of image data needs to be processed in each link of implementing infringement analysis, including feature extraction of images, obtaining results through infringement comparison and recognition of images, etc., and the upgrade and replacement of program versions used in each link, a large amount of data accumulates, resulting in huge system overhead of the backend server. Further consequences often lead to slower response speed and affect the user experience.

[0004] The backend architecture of an e-commerce platform is relatively complex. Various services are deployed in units of containers, and various data are stored in a distributed system. Therefore, if various data in the intermediate process cannot be effectively organized, it is difficult to effectively solve existing problems. Summary of the Invention

[0005] The purpose of this application is to solve the above problems and provide a risk control service data scheduling method, its corresponding device, equipment, non-volatile readable storage medium, and computer program product.

[0006] According to one aspect of this application, a risk control service data scheduling method is provided, including the following steps:

[0007] Obtain the storage address of the target image from the risk control service and transmit it to the image encoding service;

[0008] The image encoding service generates a pointer identifier based on the version identifier of the image encoder it references and the storage address;

[0009] The image encoding service detects whether the pointer identifier exists in the cache list. When it exists, return the image feature data pointed to by the pointer identifier;

[0010] When the pointer identifier does not exist in the cache list, the image encoding service inputs the image data of the target image into the image encoder to obtain corresponding image feature data and return it, and maps the obtained image feature data to the pointer identifier and stores it in the cache list;

[0011] The risk control service transmits the returned image feature data to the image comparison service, and the image comparison service determines whether the target image is an infringing image of the product image in the product image library based on the returned image feature data.

[0012] Optionally, the image encoding service generates a pointer identifier based on the version identifier of the image encoder it references and the storage address, including:

[0013] The image encoding service performs a preset mapping operation on the storage address to convert the storage address into a unique feature identifier with a unified character length;

[0014] The image encoding service concatenates the version identifier of the image encoder and the unique feature identifier into a pointer identifier.

[0015] Optionally, after the risk control service transmits the returned image feature data to the image comparison service and the image comparison service determines whether the target image constitutes an infringing image of the product images in the product image library based on the returned image feature data, including:

[0016] The risk control service maps and stores the determination result indicating whether the target image is an infringing image and the pointer identifier in a temporary storage pool, and the determination result includes the download address of the infringed product image.

[0017] Optionally, before the risk control service transmits the returned image feature data to the image comparison service and the image comparison service determines whether the target image constitutes an infringing image of the product images in the product image library based on the returned image feature data, including:

[0018] The risk control service calculates the semantic similarity between the returned image feature data and each piece of image feature data in the cache list;

[0019] The risk control service determines the pointer identifiers of some of the image feature data whose semantic similarity exceeds a preset threshold as candidate objects;

[0020] The risk control service queries the determination result corresponding to the candidate object from the temporary storage pool as the determination result of the target image and ends the subsequent processing.

[0021] Optionally, the image comparison service determines whether the target image constitutes an infringing image of the product images in the product image library based on the returned image feature data, including:

[0022] The image comparison service calculates the data distance between the returned image feature data and the image feature data of each product image in the product image library as the semantic similarity, and the image feature data of each product image is pre-stored in the product image library;

[0023] The picture comparison service determines whether there is at least one product picture in the product picture library with a semantic similarity higher than a preset threshold. When there are such product pictures, these product pictures constitute the pictures being infringed, and the target picture constitutes an infringing picture.

[0024] The picture comparison service returns the download address of the picture being infringed as the determination result to the risk control service, or returns a null value as the determination result when there is no picture being infringed.

[0025] Optionally, mapping the obtained image feature data to the pointer identifier and storing it in the cache list includes:

[0026] The image encoding service compresses the obtained image feature data using a preset compression algorithm to obtain compressed data.

[0027] The image encoding service stores the compressed data mapped to the pointer identifier in the cache list.

[0028] Optionally, after the picture comparison service determines whether the target picture constitutes an infringing picture of a product picture in the product picture library according to the returned image feature data, it includes:

[0029] When the target picture constitutes an infringing picture, the risk control service sends a picture deletion instruction to the distributed storage server pointed to by the storage address to delete the target picture.

[0030] The risk control service generates a bad record of the user who uploaded the target picture and statistically updates the total amount of the user's bad records.

[0031] The risk control service detects whether the total amount of the user's bad records exceeds a preset threshold. When it exceeds the preset threshold, it sends a permission modification instruction to the distributed storage server to restrict the user's picture upload permission.

[0032] According to another aspect of the present application, a risk control service data scheduling device is provided, including:

[0033] An address acquisition module, configured to acquire the storage address of the target picture by the risk control service and transmit it to the image encoding service.

[0034] A pointer generation module, configured to generate a pointer identifier by the image encoding service according to the version identifier of the image encoder it references and the storage address.

[0035] A cache call module, configured to detect whether the pointer identifier exists in the cache list by the image encoding service, and return the image feature data pointed to by the pointer identifier when it exists.

[0036] The encoding processing module is configured such that when the pointer identifier does not exist in the cache list, the image encoding service inputs the picture data of the target picture to the image encoder to obtain corresponding image feature data and returns the data, and stores the obtained image feature data mapped to the pointer identifier in the cache list.

[0037] The infringement analysis module is configured such that the risk control service transmits the returned image feature data to the picture comparison service, and the picture comparison service determines whether the target picture is an infringing picture of the commodity pictures in the commodity picture library according to the returned image feature data.

[0038] According to another aspect of the present application, there is provided a risk control service data scheduling device, including a central processing unit and a memory. The central processing unit is configured to call and run a computer program stored in the memory to execute the steps of the risk control service data scheduling method described in the present application.

[0039] According to another aspect of the present application, there is provided a non-volatile readable storage medium, which stores a computer program implemented according to the risk control service data scheduling method in the form of computer-readable instructions. When the computer program is called and run by a computer, it executes the steps included in the method.

[0040] According to another aspect of the present application, there is provided a computer program product, including a computer program / instructions. When the computer program / instructions are executed by a processor, the steps of the method described in any embodiment of the present application are implemented.

[0041] Compared with the prior art, for the target pictures that need to be identified for infringement, the risk control service of the present application generates a pointer identifier by using its storage address and the version identifier of the image encoder. First, it queries in the cache list whether there is image feature data corresponding to the pointer identifier. When there is no image feature data, the image encoding service calls the image encoder to generate the image feature data of the target picture, and stores the generated image feature data associated with the pointer identifier of the target picture in the cache list for subsequent reuse. After obtaining the image feature data of the target picture, the risk control service performs infringement identification on the target picture. Thus, adapting to the characteristics of high-frequency dynamic changes of pictures related to commodities in the e-commerce platform, it can avoid calling the image encoder to obtain the image feature data of pictures frequently to the greatest extent, and the image encoder is exactly the reason for the large system overhead. By quickly and conveniently calling the image feature data of pictures from the cache list through the present application, the infringement analysis efficiency of pictures can be greatly improved, and the response speed of infringement analysis can be significantly improved, thereby improving the user experience. Description of the Drawings

[0042] To more clearly illustrate the technical solutions in the embodiments of the present application, the following will briefly introduce the accompanying drawings required for the description of the embodiments. Obviously, the accompanying drawings in the following description are only some embodiments of the present application. For those of ordinary skill in the art, without creative efforts, other drawings can be obtained based on these drawings.

[0043] Figure 1 It is a schematic diagram of the network architecture of the application environment of the technical solution of the present application;

[0044] Figure 2 It is a schematic diagram of an exemplary service architecture of the present application;

[0045] Figure 3 It is a schematic flowchart of an embodiment of the risk control service data scheduling method of the present application;

[0046] Figure 4 It is a schematic flowchart of implementing infringement analysis for a target picture depending on a cache mechanism in the embodiments of the present application;

[0047] Figure 5 It is a schematic flowchart of implementing infringement analysis by a picture comparison service in the embodiments of the present application;

[0048] Figure 6 It is a schematic flowchart of adjusting user permissions according to a determination result in the embodiments of the present application;

[0049] Figure 7 It is a principle block diagram of the risk control service data scheduling device of the present application;

[0050] Figure 8 It is a schematic diagram of the structure of a risk control service data scheduling device adopted by the present application. Detailed implementation manners

[0051] The models cited or possibly cited in the present application, including traditional machine learning models or deep learning models, unless expressly specified, can be deployed on a remote server and remotely called on the client, or can be directly called on the client capable of the device. In some embodiments, when it runs on the client, its corresponding intelligence can be obtained through transfer learning to reduce the requirements for the client's hardware operation resources and avoid over-occupying the client's hardware operation resources.

[0052] Please refer to Figure 1, The network architecture adopted by an exemplary application scenario of this application includes a terminal device 80, an independent station server 81, and an application server 82. The application server 82 can be used to deploy risk control services, picture comparison services, image encoding services, etc. The risk control service provides services by running a computer program product implemented according to the risk control service data scheduling method of this application and opening corresponding interfaces. The independent station server 81 can be used to deploy and open an online store. Users on the terminal device 80 can submit product pictures to the independent station server 81 on the page of the online store. The independent station server 81 further calls the corresponding interface provided by the risk control service of the application server 82 to process the product pictures uploaded by the user, obtains a determination result indicating whether the product pictures constitute an infringement of the product pictures in the product picture library preset by the e-commerce platform, and returns it to the terminal device 80.

[0053] The product picture library can be product pictures pre-collected by the e-commerce platform, such as pictures of registered products. The product picture library can be set individually for an independent station or can centrally serve a large number of independent stations in the entire e-commerce platform. The product picture library can be stored in a distributed storage system.

[0054] The application server can be a service cluster that runs various services through a microservices architecture. Accordingly, the risk control service, picture comparison service, and image encoding service can run in the same container or different containers provided by the microservices architecture respectively.

[0055] As Figure 2 shown in the schematic diagram, the risk control service is responsible for the centralized response to requests triggered on each independent station, and can realize the on-demand invocation of the image encoding service and the picture comparison service. The product pictures uploaded by users to the independent station can pass the storage address as a parameter to the risk control service. The risk control service can then download the target picture according to this storage address, obtain the binary data of the target picture, and can also generate corresponding image feature data according to the binary data by calling the image encoding service, and find out the infringed product pictures corresponding to the target picture by calling the image comparison service, and obtain its download address, etc.

[0056] The image encoding service can be constructed with various deep learning-based models, such as Convolutional Neural Network (CNN), Residual Network (ResNet), etc. It is pre-trained in advance to acquire the ability to extract deep semantic information for product pictures and obtain corresponding image feature data before being put into use. The model can be iteratively trained multiple times at different stages, thus deriving multiple upgraded versions, each version having its corresponding inference ability and corresponding version identifier.

[0057] The image comparison service is mainly used to perform semantic matching operations. By calculating the data distance between the image feature data of pairwise product images, the semantic similarity between each two images is determined. According to the semantic similarity, it can be determined whether two product images are similar, so as to determine whether one product image constitutes an infringing image of another product image.

[0058] Based on the above principle introduction, please refer to Figure 3 , in an embodiment of a risk control service data scheduling method provided by this application, it includes the following steps:

[0059] Step S1100: The risk control service obtains the storage address of the target image and transmits it to the image encoding service;

[0060] When a user of an independent station uploads a product image, the product image is used as the target image and scheduled and stored by the independent station to obtain its storage address. Then, the independent station calls the audit interface provided by the risk control service deployed on the application server, and transmits the storage address of the target image through this audit interface to instruct the risk control service to detect whether the target image is suspected of infringement.

[0061] After receiving the storage address of the target image, the risk control service deployed on the application server can first download the binary data of the target image from this storage address to the local, or directly transmit the storage address to the image encoding service, and the image encoding service downloads the binary data of the target image from this storage address to its local.

[0062] Step S1200: The image encoding service generates a pointer identifier according to the version identifier of the image encoder it references and the storage address;

[0063] As mentioned above, the image encoding service is implemented to obtain corresponding image feature data by calling an image encoder to extract deep semantic information for the target image. The image encoder is implemented using a neural network model, and this neural network model can be trained multiple times to achieve upgrading. Each time training obtains a version or generates a corresponding version identifier. The image feature data generated by different versions of the image encoder for the same image may be different. Therefore, establishing the mapping relationship data between the version identifier of the image encoder and the image feature data generated by the corresponding version helps to eliminate the problem of inaccurate data caused by different versions of the image encoder.

[0064] In one embodiment, for the target image, the image encoding service generates a pointer identifier with unique features for it according to the following process, so as to correspond to the currently adopted version of the image encoder and the specific target image:

[0065] Step S1210: The image encoding service performs a preset mapping operation on the storage address to convert the storage address into a unique feature identifier with a unified character length.

[0066] First, a preset mapping algorithm can be used to perform a mapping operation on the storage address of the target picture. The mapping algorithm can be, for example, a digital digest algorithm or other hash algorithms. Through the operation of the mapping algorithm, the storage address is converted into a unique feature identifier with a unified character length. It is not difficult to understand that different storage addresses will obtain different unique feature identifiers, and a target picture can be uniquely pointed to through the unique feature identifier.

[0067] Step S1220: The image encoding service concatenates the version identifier of the image encoder with the unique feature identifier to form a pointer identifier.

[0068] Since the version identifier of the image encoder has a differentiating effect, further, the version identifier of the image encoder currently in use by the image encoding service can be concatenated with the unique feature identifier generated according to the storage address to form a pointer identifier. This pointer identifier can not only indicate the version of the image encoder used but also determine the specific target picture.

[0069] It is not difficult to understand that the pointer identifier itself has a unique differentiating effect and provides the version information of the image encoder used. Therefore, it can play a role in indicating a specific object in the follow-up.

[0070] Step S1300: The image encoding service detects whether the pointer identifier exists in the cache list. When it exists, it returns the image feature data pointed to by the pointer identifier.

[0071] The image encoding service pre-creates a cache list in its local memory or other storage devices to store the image feature data of each target picture passed in by the risk control service, so as to avoid frequently calling the image encoder for the same target picture. In the cache list, the pointer identifier of the target picture can be mapped and stored with the image feature data of the target picture, so that the corresponding image feature data can be quickly called through the pointer identifier.

[0072] Accordingly, the image encoding service first performs a query in the cache list using the pointer identifier of the target picture. When the pointer identifier exists in the cache list, it means that the corresponding target picture has already had its image feature data generated by the image encoder and stored in the cache list. Therefore, the image feature data mapped to this pointer identifier is directly called and returned to the risk control service, and then step S1500 can be executed by jumping. If the pointer identifier of the target picture does not exist in the cache list, it means that the target picture has not had its image feature data generated by the current version of the image encoder. Therefore, step S1400 can be executed.

[0073] Step S1400: When the pointer identifier does not exist in the cache list, the image encoding service inputs the picture data of the target picture to the image encoder to obtain the corresponding image feature data and returns it, and stores the obtained image feature data mapped to the pointer identifier in the cache list.

[0074] When the pointer identifier does not exist in the cache list, the image encoding service needs to call the current version of the image encoder to generate the corresponding image feature data for the target picture. Thus, after performing necessary format preprocessing on the binary data of the target picture, it is transmitted to the image encoder, and the image encoder performs feature representation on the target picture, extracts its deep semantic information, and obtains the corresponding image feature data. The image encoding service then returns this image feature data to the risk control service.

[0075] To facilitate the reuse of the image feature data generated by the image encoder for the target picture, the image encoding service further stores the image feature data obtained by the target picture passing through the image encoder mapped to the pointer identifier in the cache list.

[0076] In one embodiment, to further save the system storage overhead, the image feature data generated by the image encoder is cached according to the following process:

[0077] Step S1410: The image encoding service uses a preset compression algorithm to compress the obtained image feature data to obtain compressed data.

[0078] For the image feature data generated by the image encoder for the target picture, a preset compression algorithm can be used to compress this image feature data to obtain compressed data with fewer bytes. Generally speaking, the image feature data generated by the image encoder is matrix - type data with various values. Therefore, any common compression algorithm can be used to compress it, and it is expected to obtain fewer bytes. Of course, the compression algorithm should use a lossless compression algorithm.

[0079] Step S1420: The image encoding service maps and stores the compressed data and the pointer identifier in the cache list.

[0080] After obtaining the compressed data of the image feature data, map and store the compressed data and the pointer identifier of the target picture in the cache list. When it is necessary to obtain the image feature data with a certain pointer identifier from the cache list, decompress the compressed data pointed to by the pointer identifier according to the decompression algorithm corresponding to the compression algorithm to obtain the original image feature data.

[0081] Since the number of pictures that need to be audited by the e-commerce platform daily is huge, and the image feature data generated by each target picture also occupies rare storage space after all, therefore, compressing the image feature data in the cache list before storing can further save the system storage resources and improve the data caching ability of the limited storage resources for the picture audit service of the entire e-commerce platform.

[0082] Step S1500: The risk control service transmits the returned image feature data to the picture comparison service, and the picture comparison service determines whether the target picture is an infringing picture of the product pictures in the product picture library according to the returned image feature data.

[0083] Whether the image feature data of the target picture is obtained by querying from the cache list or the image feature data of the target picture is generated immediately by the image encoding service calling the image encoder, it will ultimately be returned to the risk control service. The risk control service further calls the picture comparison service according to the returned image feature data. After receiving the returned image feature data, the picture comparison service calculates the semantic similarity with the image feature data of each product picture in the preset product picture library, and then determines the product pictures with the semantic similarity exceeding the preset threshold as the infringed pictures. Correspondingly, the target picture becomes an infringing picture. When the semantic similarity obtained for each product picture in the product picture library is lower than the preset threshold, it indicates that there is no product picture in the product picture library that is substantially similar to the target picture, so the target picture can pass the audit.

[0084] In one embodiment, for the target picture, the picture comparison service can return the corresponding determination result to the risk control service. When the target picture is an infringing picture, the determination result can be encapsulated to include the download addresses of each infringed picture; when the target picture is not an infringing picture, the determination result can be encapsulated as a null value. Thus, it is convenient for the risk control service to make subsequent processing according to the content of the determination result.

[0085] In one embodiment, when there are multiple product images in the product image library whose semantic similarity exceeds the preset threshold, only the product image with the highest semantic similarity can be used as the infringed image, and then the download address of the product image can be obtained separately and encapsulated in the determination result. That is, only the download address of the infringed image that is most similar to the content of the target image is provided for the risk control service.

[0086] According to the above embodiments, for the target image that needs to be identified for infringement in the risk control service of the present application, a pointer identifier is generated using its storage address and the version identifier of the image encoder. First, it is checked whether there is corresponding image feature data for the pointer identifier in the cache list. When there is no image feature data, the image encoding service calls the image encoder to generate the image feature data of the target image, and the generated image feature data is associated with the pointer identifier of the target image and stored in the cache list for subsequent reuse. After obtaining the image feature data of the target image, the risk control service performs infringement identification on the target image. Thus, adapting to the characteristics of high-frequency dynamic changes of product-related images in the e-commerce platform, it is possible to avoid, to the greatest extent, frequently calling the image encoder to obtain the image feature data of the image, and the image encoder is precisely the reason for the large system overhead. By quickly and conveniently calling the image feature data of the image from the cache list through the present application, the infringement analysis efficiency of the image can be greatly improved, and the response speed of the infringement analysis can be significantly enhanced, thereby improving the user experience.

[0087] To further improve the review efficiency, based on any embodiment of the present application, the risk control service transmits the returned image feature data to the image comparison service. After the image comparison service determines whether the target image constitutes an infringed image of the product image in the product image library according to the returned image feature data, it includes:

[0088] Step S2100: The risk control service stores the determination result indicating whether the target image is an infringed image in a mapping relationship with the pointer identifier in a temporary storage pool. The determination result includes the download address of the infringed product image.

[0089] Based on the determination result obtained by the risk control service, it can be confirmed whether the target image constitutes an infringed image. That is, when the determination result includes at least one download address, it indicates that the target image constitutes an infringed image. When the determination result does not include any download address, it can be confirmed that the target image is a non-infringed image.

[0090] For the convenience of subsequent calls, the risk control service can construct the pointer identifier of the target image and its corresponding determination result into mapping relationship data, and then store it in a preset temporary storage pool. This temporary storage pool can be a cache space in the memory, which is convenient for subsequent calls to the determination result of the same target image from the temporary storage pool.

[0091] According to the above embodiments, the determination result obtained through the cached image comparison service enables the reuse of the determination result. When performing risk control audits on the same target image subsequently, the response speed of the risk control audit can be improved by querying the corresponding determination result from the staging pool, avoiding frequent calls to the image comparison service and saving the overall system overhead.

[0092] Based on any embodiment of the present application, refer to Figure 4 , before the risk control service transmits the returned image feature data to the image comparison service and the image comparison service determines whether the target image constitutes an infringing image of the product image in the product image library according to the returned image feature data, it includes:

[0093] Step S3100: The risk control service calculates the semantic similarity between the returned image feature data and each piece of image feature data in the cache list;

[0094] Since the cache list stores the image feature data of the historically audited target images, for the target image that needs to be audited currently, after the risk control service obtains the returned image feature data, the semantic similarity can be determined based on the image feature data of the target image that needs to be audited currently and each piece of image feature data in the cache list, so as to determine whether there is a historical target image similar to the target image that needs to be audited currently according to the semantic similarity. In this way, using the pointer identifier of the historical target image that is highly similar to the target image that needs to be audited currently to obtain the determination result in the staging pool can not only save the step of calling the image comparison service but also improve the audit efficiency.

[0095] When calculating the semantic similarity between the image feature data of the target image that needs to be audited currently, that is, the returned image feature data, and each piece of image feature data in the cache list, any data distance algorithm can be used for implementation, including but not limited to the cosine similarity algorithm, Euclidean distance algorithm, Pearson correlation coefficient algorithm, Jaccard coefficient algorithm, etc. As long as the final data distance result is normalized to the same similarity metric dimension, the smaller the value, the more similar it means.

[0096] For the embodiment in which the cache list stores the image feature data as compressed data, when performing semantic similarity calculation, the corresponding decompression algorithm can be applied to decompress the compressed data to restore the original image feature data and then perform the calculation.

[0097] Step S3200: The risk control service determines the pointer identifiers of some image feature data whose semantic similarity exceeds the preset threshold as candidate objects;

[0098] After determining the semantic similarity of the image feature data corresponding to each historical target image in the cache list, a preset threshold can be used for screening. The preset threshold can be an empirical threshold or a measured threshold, and is used to measure the degree to which a corresponding historical target image is similar to the target image to be audited currently. For the image feature data with a semantic similarity higher than the preset threshold, the corresponding pointer identifier can be obtained, and these pointer identifiers are used as candidate objects. It can be seen that the candidate objects are actually some historical target images.

[0099] Step S3300: The risk control service queries the determination result corresponding to the candidate object from the staging pool, and uses it as the determination result of the target image to end the subsequent processing.

[0100] Since in some embodiments of the present application, a mapping relationship data between the determination result of the historical target image and the pointer identifier of the historical target image is stored in a staging pool, and in the previous step, multiple candidate objects corresponding to the target image to be audited currently are obtained, and these candidate objects are represented by pointer identifiers. Accordingly, further using these candidate objects, that is, pointer identifiers, to query one or more determination results in the staging pool. After obtaining these determination results, the subsequent processing can be terminated, for example, step S1500 is no longer executed. Subsequently, the risk control service can return the corresponding audit result to the user according to these determination results.

[0101] Through the above embodiments, on the basis of using the image feature data in the cache list, further using the association relationship between the pointer identifiers, matching the pointer identifiers indicating some historical target images highly similar to the target image to be audited currently from the cache list, and then retrieving the corresponding determination results in the staging pool where the determination results are pre-stored in advance according to these pointer identifiers, not only reusing the image feature data, but also reusing the determination results, avoiding frequent calls to the image encoding service and the image comparison service, and being able to quickly respond to the user's audit request by using the high response rate of the cache mechanism, improving the risk control audit efficiency, making the security audit process of the product pictures faster, and thus improving the user experience when the product pictures are published on the e-commerce platform.

[0102] Based on any embodiment of the present application, please refer to Figure 5 , and the picture comparison service determines whether the target picture constitutes an infringing picture of the product pictures in the product picture library according to the returned image feature data, including:

[0103] Step S1510: The picture comparison service calculates the data distance between the returned image feature data and the image feature data of each product picture in the product picture library as the semantic similarity, and the image feature data of each product picture in the product picture library is pre-stored therein;

[0104] After the image comparison service obtains the image feature data transmitted by the risk control service, it calculates the semantic similarity using the image feature data and the image feature data of each product image in the preset product image library. Thus, for each product image in the product image library, the data distance between it and the target image can be obtained and converted into semantic similarity. The method for calculating semantic similarity is the same as that described above and will not be elaborated here.

[0105] As described above, a large number of product images are pre-collected and stored in the product image library. The product images usually contain image content and / or text content related to the product, such as the appearance diagram of the product or the registered trademark of the product. It is not difficult to understand that if the unauthorized target image contains content similar to the appearance diagram of the product image in the product image library, it may easily constitute an infringement of the design patent. If there is content similar to the registered product in the product image in the product image library, it may easily constitute a trademark infringement. Thus, by performing semantic matching with the product images in the product image library, it is possible to make a prediction with a certain degree of accuracy as to whether the target image is an infringing image.

[0106] Step S1520: The image comparison service determines whether there is at least one product image in the product image library whose semantic similarity is higher than a preset threshold. When there are such product images, these product images constitute the infringed images, and the target image constitutes an infringing image.

[0107] Furthermore, the image comparison service uses a preset threshold to compare the semantic similarity obtained for each product image in the product image library with this preset threshold. If it is higher than the preset threshold, the corresponding product image is regarded as the infringed image, which also means that the target image constitutes an infringement of this product image.

[0108] Similarly, the preset threshold can be an empirical threshold or a measured threshold, which is used to measure the degree of similarity between a corresponding historical target image and the target image that needs to be reviewed currently.

[0109] For the same target image, there may be multiple infringed images in the product image library. When there are multiple infringed images, they can also be determined together as all the infringed images corresponding to the target image. In one embodiment, it is also possible to only determine the infringed image with the highest semantic similarity as the unique infringed image actually corresponding to the target image. In this regard, those skilled in the art can implement it flexibly.

[0110] Step S1530: The picture comparison service returns the download address of the infringed picture as the determination result to the risk control service, or returns a null value as the determination result when there is no infringed picture.

[0111] The picture comparison service can be responsible for encapsulating the determination result and returning it to the risk control service. In one embodiment, when encapsulating the determination result, the picture comparison service encapsulates the download address of the finally determined infringed picture as the determination result. When there is no infringed picture, a null value is encapsulated as the determination result, and then this determination result is returned to the risk control service.

[0112] In another embodiment, the picture comparison service can further download the binary data of the infringed picture, and then encapsulate these binary data in the determination result and return it to the risk control service. Thus, the risk control service can save the time for separately downloading the binary data.

[0113] As can be seen from the above embodiments, the picture comparison service identifies whether the target picture is similar to the product pictures in the product picture library based on the semantic similarity between the image feature data. Since the image feature data is obtained by the image encoder extracting the deep semantic information of the picture, the deep semantic information can highly abstract the semantic features of the corresponding picture, and the semantic features can effectively determine whether two pictures are similar. Therefore, the picture comparison service can accurately determine the infringed picture for the target picture, thereby realizing an effective picture security audit mechanism, laying a reliable technical foundation for the picture information security audit of the e-commerce platform, and ensuring the picture information security of the e-commerce platform.

[0114] Based on any embodiment of the present application, please refer to Figure 6 , after the picture comparison service determines whether the target picture constitutes an infringed picture of the product picture in the product picture library according to the returned image feature data, it includes:

[0115] Step S4100: When the target picture constitutes an infringed picture, the risk control service sends a picture deletion instruction to the distributed storage server pointed to by the storage address to delete the target picture;

[0116] After the image comparison service returns the determination result of the target image to the risk control service, the risk control service can identify whether the target image is an infringing image based on whether the determination result is a null value. When the target image is not an infringing image, the risk control service can pass the review and allow the user to finally publish the target image to the desired location. When the target image is an infringing image, the risk control service can generate an image deletion instruction and send it to the distributed storage server storing the target image, that is, the distributed storage server pointed to by the storage address, so as to execute the deletion operation of the target image through the distributed storage server and purify the image information of the e-commerce platform.

[0117] Step S4200: The risk control service generates a bad record of the user who uploaded the target image and statistically updates the total amount of the user's bad records.

[0118] The e-commerce platform can set a credit management mechanism for infringing images. Accordingly, when the target image is identified as an infringing image, the risk control service generates a corresponding bad record for the user who uploaded the target image, and statistically calculates the total amount of all bad records generated by this user in history. It is not difficult to understand that the higher this total amount is, the lower the credit of this user.

[0119] Step S4300: The risk control service detects whether the total amount of the user's bad records exceeds a preset threshold. When it exceeds the preset threshold, it sends a permission modification instruction to the distributed storage server to restrict the user's image upload permission.

[0120] For the need of credit management, a preset threshold can be set corresponding to the bad record. This preset threshold is used to measure whether the total amount of a user's bad records is sufficient to trigger the punishment mechanism. Accordingly, compare the total amount of the bad records of the user who uploaded the target image with this preset threshold. When the total amount of bad records exceeds the preset threshold, it means that the punishment mechanism is triggered. Accordingly, a permission modification instruction for this user can be sent to the distributed storage server. This permission modification instruction instructs the distributed storage server to modify the user's image upload permission, such as prohibiting the user from uploading new images within a certain time range, or reducing the total number of images that the user can upload, etc., so as to effectively punish this user.

[0121] According to the above embodiments, it can be seen that in the process of implementing efficient security review of product images in this application, the user punishment mechanism can be automatically triggered, the user's image upload permission can be timely controlled according to the total amount of the user's bad records, the information security management mechanism of the e-commerce platform can be automatically maintained, the information security of the e-commerce platform can be ensured, excessive manual intervention can be avoided, and the information security maintenance cost can be effectively saved.

[0122] Please refer toFigure 7 , a risk control service data scheduling device provided according to an aspect of the present application, includes an address acquisition module 1100, a pointer generation module 1200, a cache call module 1300, an encoding processing module 1400, and an infringement analysis module 1500, wherein: the address acquisition module 1100 is configured to acquire the storage address of the target picture from the risk control service and transmit it to the image encoding service; the pointer generation module 1200 is configured to generate a pointer identifier by the image encoding service according to the version identifier of the image encoder it references and the storage address; the cache call module 1300 is configured to detect whether the pointer identifier exists in the cache list by the image encoding service, and return the image feature data pointed to by the pointer identifier when it exists; the encoding processing module 1400 is configured to, when the pointer identifier does not exist in the cache list, input the picture data of the target picture into the image encoder by the image encoding service to obtain the corresponding image feature data and return it, and map and store the obtained image feature data and the pointer identifier in the cache list; the infringement analysis module 1500 is configured to transmit the returned image feature data from the risk control service to the picture comparison service, and the picture comparison service determines whether the target picture is an infringing picture of the commodity pictures in the commodity picture library according to the returned image feature data.

[0123] Based on any embodiment of the present application, the pointer generation module 1200 includes: a mapping processing unit configured to perform a preset mapping operation on the storage address by the image encoding service to convert the storage address into a unique feature identifier with a unified character length; an identifier concatenation unit configured to concatenate the version identifier of the image encoder and the unique feature identifier into a pointer identifier by the image encoding service.

[0124] Based on any embodiment of the present application, after the operation of the infringement analysis module 1500, it includes: a result temporary storage module configured to map and store the determination result indicating whether the target picture is an infringing picture and the pointer identifier in a temporary storage pool by the risk control service, and the determination result includes the download address of the infringed commodity picture.

[0125] Based on any embodiment of the present application, before the operation of the infringement analysis module 1500, it includes: a semantic matching module configured to calculate the semantic similarity between the returned image feature data and each image feature data in the cache list by the risk control service; an object determination module configured to determine the pointer identifiers of some image feature data whose semantic similarity exceeds a preset threshold as candidate objects by the risk control service; a temporary storage query module configured to query the determination result corresponding to the candidate object from the temporary storage pool by the risk control service, and use it as the determination result of the target picture and end the subsequent processing.

[0126] Based on any embodiment of the present application, the infringement analysis module 1500 includes: a similarity calculation unit configured to calculate the data distance between the image feature data returned by the image comparison service and the image feature data of each product image in the product image library as the semantic similarity, wherein the image feature data of each product image in the product image library is pre-stored; an infringement determination unit configured to determine by the image comparison service whether there is at least one product image in the product image library with a semantic similarity higher than a preset threshold. When there are such product images, these product images constitute the infringed images, and the target image constitutes an infringing image; a result output unit configured to return the download address of the infringed image as the determination result by the image comparison service or a null value as the determination result when there is no infringed image to the risk control service.

[0127] Based on any embodiment of the present application, the encoding processing module 1400 includes: a data compression unit configured to compress the obtained image feature data by the image encoding service using a preset compression algorithm to obtain compressed data; a data storage unit configured to map and store the compressed data and the pointer identifier in the cache list by the image encoding service.

[0128] Based on any embodiment of the present application, the subsequent module after the infringement analysis module 1500 includes: a picture deletion module configured to, when the target picture constitutes an infringing picture, send a picture deletion instruction by the risk control service to the distributed storage server pointed to by the storage address to delete the target picture; a statistics update module configured to generate a bad record of the user who uploaded the target picture by the risk control service and statistically update the total amount of the user's bad records; a permission control module configured to detect by the risk control service whether the total amount of the user's bad records exceeds a preset threshold. When it exceeds the preset threshold, send a permission modification instruction to the distributed storage server to limit the picture upload permission of the user.

[0129] Another embodiment of the present application further provides a risk control service data scheduling device. As Figure 8 shown, it is a schematic internal structure diagram of the risk control service data scheduling device. The risk control service data scheduling device includes a processor, a computer-readable storage medium, a memory, and a network interface connected through a system bus. Among them, the computer-readable non-volatile storage medium of the risk control service data scheduling device stores an operating system, a database, and computer-readable instructions. Information sequences can be stored in the database. When the computer-readable instructions are executed by the processor, the processor can implement a risk control service data scheduling method.

[0130] The processor of the risk control service data scheduling device is used to provide computing and control capabilities to support the operation of the entire risk control service data scheduling device. Computer-readable instructions can be stored in the memory of the risk control service data scheduling device. When the computer-readable instructions are executed by the processor, the processor can execute the risk control service data scheduling method of this application. The network interface of the risk control service data scheduling device is used to connect and communicate with the terminal.

[0131] Those skilled in the art can understand that Figure 8 the structure shown in is only a block diagram of some structures related to the solution of this application, and does not constitute a limitation on the risk control service data scheduling device to which the solution of this application is applied. The specific risk control service data scheduling device may include more or fewer components than those shown in the figure, or combine some components, or have different component arrangements.

[0132] In this embodiment, the processor is used to execute Figure 7 the specific functions of each module in . The memory stores the program codes and various types of data required to execute the above modules or sub-modules. The network interface is used to implement data transmission between the user terminal or the server. In this embodiment, the non-volatile readable storage medium stores the program codes and data required to execute all modules in the risk control service data scheduling device of this application. The server can call the program codes and data of the server to execute the functions of all modules.

[0133] This application also provides a non-volatile readable storage medium storing computer-readable instructions. When the computer-readable instructions are executed by one or more processors, one or more processors are caused to execute the steps of the risk control service data scheduling method of any embodiment of this application.

[0134] This application also provides a computer program product, including computer programs / instructions. When the computer programs / instructions are executed by one or more processors, the steps of the method described in any embodiment of this application are implemented.

[0135] Those of ordinary skill in the art can understand that to implement all or part of the processes in the methods of the above embodiments of this application, it can be completed by instructing relevant hardware through a computer program. The computer program can be stored in a non-volatile readable storage medium. When the program is executed, it can include the processes of the embodiments of the above methods. Among them, the aforementioned storage medium can be a computer-readable storage medium such as a magnetic disk, an optical disk, a Read-Only Memory (ROM), or a Random Access Memory (RAM), etc.

[0136] In summary, through this application, the e-commerce platform can quickly and conveniently call the image feature data of pictures from the cache list, which can greatly improve the infringement analysis efficiency of pictures, significantly enhance the response speed of infringement analysis, and thus improve the user experience.

Claims

1. A risk control service data scheduling method, characterized in that, Including: The risk control service obtains the storage address of the target picture and transmits it to the image encoding service; The image encoding service generates a pointer identifier according to the version identifier of the image encoder it references and the storage address; The image encoding service detects whether the pointer identifier exists in the cache list. When it exists, it returns the image feature data pointed to by the pointer identifier; When the pointer identifier does not exist in the cache list, the image encoding service inputs the picture data of the target picture into the image encoder to obtain the corresponding image feature data and returns it, and maps the obtained image feature data to the pointer identifier and stores it in the cache list; The risk control service transmits the returned image feature data to the picture comparison service, and the picture comparison service determines whether the target picture constitutes an infringing picture of the product pictures in the product picture library according to the returned image feature data.

2. The risk control service data scheduling method according to claim 1, wherein The image encoding service generates a pointer identifier according to the version identifier of the image encoder it references and the storage address, including: The image encoding service performs a preset mapping operation on the storage address to convert the storage address into a unique feature identifier with a unified character length; The image encoding service concatenates the version identifier of the image encoder and the unique feature identifier into a pointer identifier.

3. The risk control service data scheduling method according to claim 1, wherein After the risk control service transmits the returned image feature data to the picture comparison service, and the picture comparison service determines whether the target picture constitutes an infringing picture of the product pictures in the product picture library according to the returned image feature data, including: The risk control service maps and stores the determination result indicating whether the target picture is an infringing picture to the pointer identifier in the temporary storage pool, and the determination result includes the download address of the infringed product picture.

4. The risk control service data scheduling method according to claim 3, wherein Before the risk control service transmits the returned image feature data to the picture comparison service, and the picture comparison service determines whether the target picture constitutes an infringing picture of the product pictures in the product picture library according to the returned image feature data, including: The risk control service calculates the semantic similarity between the returned image feature data and each image feature data in the cache list; The risk control service determines the pointer identifiers of some of the image feature data whose semantic similarity exceeds a preset threshold as candidate objects; The risk control service queries the determination result corresponding to the candidate object from the temporary storage pool as the determination result of the target picture and ends the subsequent processing.

5. The risk control service data scheduling method according to claim 1, wherein The picture comparison service determines whether the target picture constitutes an infringing picture of the product pictures in the product picture library according to the returned image feature data, including: The picture comparison service calculates the data distance between the returned image feature data and the image feature data of each product picture in the product picture library as the semantic similarity, and the image feature data of each product picture is pre-stored in the product picture library; The picture comparison service determines whether there is at least one product picture in the product picture library whose semantic similarity is higher than a preset threshold. When there are such product pictures, these product pictures constitute the infringed pictures, and the target picture constitutes an infringing picture; The picture comparison service returns the download address of the infringed picture as the determination result or a null value as the determination result when there is no infringed picture to the risk control service.

6. The risk control service data scheduling method according to any one of claims 1 to 5, characterized in that Mapping and storing the obtained image feature data to the pointer identifier in the cache list, including: The image encoding service compresses the obtained image feature data using a preset compression algorithm to obtain compressed data; The image encoding service maps and stores the compressed data to the pointer identifier in the cache list.

7. The risk control service data scheduling method according to any one of claims 1 to 5, characterized in that After the picture comparison service determines whether the target picture constitutes an infringing picture of the product pictures in the product picture library according to the returned image feature data, including: When the target picture constitutes an infringing picture, the risk control service sends a picture deletion instruction to the distributed storage server pointed to by the storage address to delete the target picture; The risk control service generates a bad record of the user who uploaded the target picture and statistically updates the total amount of the user's bad records; The risk control service detects whether the total amount of the user's bad records exceeds a preset threshold. When it exceeds the preset threshold, it sends a permission modification instruction to the distributed storage server to restrict the user's picture upload permission.

8. A risk control service data scheduling device, characterized in that, Including: An address acquisition module, configured to obtain the storage address of the target picture by the risk control service and transmit it to the image encoding service; A pointer generation module, configured to generate a pointer identifier by the image encoding service according to the version identifier of the image encoder it references and the storage address; A cache call module, configured to detect whether the pointer identifier exists in the cache list by the image encoding service. When it exists, return the image feature data pointed to by the pointer identifier; An encoding processing module, configured to, when the pointer identifier does not exist in the cache list, input the picture data of the target picture to the image encoder by the image encoding service to obtain corresponding image feature data and return it, and map and store the obtained image feature data to the pointer identifier in the cache list; An infringement analysis module, configured to transmit the returned image feature data from the risk control service to the picture comparison service, and the picture comparison service determines whether the target picture constitutes an infringing picture of the product pictures in the product picture library according to the returned image feature data.

9. A risk control service data scheduling device, comprising a central processing unit and a memory, characterized in that, The central processing unit is used to call and run the computer program stored in the memory to execute the steps of the method according to any one of claims 1 to 7.

10. A non-volatile readable storage medium, characterized in that, It stores in the form of computer-readable instructions a computer program implemented according to the method according to any one of claims 1 to 7. When the computer program is called and run by a computer, it executes the steps included in the corresponding method.

Images