Commodity recall method and device, computer equipment, medium and product

Abstract

The invention discloses a commodity recall method and device, computer equipment, a medium and a product, and the method comprises the steps: obtaining a plurality of commodity clusters obtained by clustering commodities in a commodity database, and obtaining the clustering features of each commodity cluster in the plurality of commodity clusters; obtaining an index image associated with the target object, and determining at least one commodity cluster matched with the index image from the plurality of commodity clusters according to the image features of the index image and the clustering features of each commodity cluster; and determining recalled commodities of the target object from the commodities included in the at least one commodity cluster. According to the method, the commodities are recalled through the commodity clusters, the commodity recall capability can be improved, and the commodity recall speed can also be improved.

Description

Technical Field

[0001] The present application relates to the field of Internet technology, and specifically to a product recall method, apparatus, computer equipment, medium, and product. Background Art

[0002] With the development of Internet technology, various applications have gradually introduced item trading functions, that is, the application introduces merchants and allows each merchant to display products in the corresponding module of the application so that users can purchase. Before purchasing, users always use the search function to search for other products that match the given pictures and texts to meet their own purchasing needs.

[0003] Currently, image features are determined based on the image given by the user. Then, the image features of all products are obtained from the product library, and products are recalled from the entire product library based on the image features corresponding to the image given by the user, so as to obtain products that match the image given by the user. However, this method has limited recall capabilities and cannot meet the purchasing needs of users; moreover, it is necessary to filter the text of each product from all products, which is time-consuming.

[0004] Therefore, how to recall sufficient goods accurately and efficiently becomes a technical problem that needs to be solved urgently. Summary of the invention

[0005] The embodiments of the present application provide a product recall method, apparatus, computer equipment, medium, and product, which can not only improve the product recall capability but also improve the product recall speed by clustering products.

[0006] The present application discloses a product recall method, which includes:

[0007] Acquire multiple commodity clusters obtained by clustering commodities in a commodity database, and acquire clustering features of each commodity cluster in the multiple commodity clusters;

[0008] Acquire an index image associated with the target object, and determine at least one commodity cluster matching the index image from the plurality of commodity clusters according to image features of the index image and cluster features of each commodity cluster;

[0009] The recalled products of the target object are determined from the products included in the at least one product cluster.

[0010] The present application discloses a product recall device, which includes:

[0011] An acquisition unit, configured to acquire a plurality of commodity clusters obtained by clustering commodities in a commodity database, and to acquire a clustering feature of each commodity cluster in the plurality of commodity clusters;

[0012] The acquisition unit is further configured to acquire an index image associated with the target object, and determine at least one commodity cluster matching the index image from the plurality of commodity clusters according to image features of the index image and cluster features of each commodity cluster;

[0013] A determination unit is used to determine the recalled commodity of the target object from the commodities included in the at least one commodity cluster.

[0014] On the one hand, an embodiment of the present application discloses a computer device, which includes a processor, a storage device and a communication interface, wherein the processor, the storage device and the communication interface are interconnected, the storage device is used to store a computer program, and the computer program includes program instructions; the processor is used to call the program instructions to execute the above-mentioned product recall method.

[0015] In one aspect, the present application discloses a computer-readable storage medium storing one or more computer programs. The computer program includes program instructions. When the program instructions are executed by a processor, they are used to execute the above-mentioned product recall method.

[0016] On the one hand, an embodiment of the present application discloses a computer program product, which includes a computer program, and the computer program is suitable for being loaded by a processor and executing the above-mentioned product recall method.

[0017] In the embodiment of the present application, multiple commodity clusters are obtained by clustering commodities in a commodity database, and cluster features of each commodity cluster in the multiple commodity clusters are obtained; an index image associated with a target object is obtained, and at least one commodity cluster matching the index image is determined from the multiple commodity clusters based on the image features of the index image and the cluster features of each commodity cluster; and the recalled commodities of the target object are determined from the commodities included in the at least one commodity cluster. In this method, on the one hand, since the cluster features represent the features of a class of commodities, the recall rate of commodities recalled by cluster features is higher than that of a single commodity; on the other hand, since the number of cluster features is less than that of a single commodity, the recall rate of commodities recalled by cluster features is faster. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0019] Figure 1 It is a schematic diagram of the network architecture of a product recall system disclosed in an embodiment of the present application;

[0020] Figure 2 It is a flowchart of a product recall method disclosed in an embodiment of the present application;

[0021] Figure 3 is a schematic diagram of a commodity cluster disclosed in an embodiment of the present application;

[0022] Figure 4 It is a flowchart of another product recall method disclosed in an embodiment of the present application;

[0023] Figure 5 It is a framework diagram for determining commodity clusters disclosed in an embodiment of the present application;

[0024] Figure 6 It is a framework diagram for product recall disclosed in an embodiment of the present application;

[0025] Figure 7 It is a structural schematic diagram of a product recall device disclosed in an embodiment of the present application;

[0026] Figure 8 It is a structural schematic diagram of a computer device disclosed in an embodiment of the present application. DETAILED DESCRIPTION

[0027] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0028] With the development of Internet technology, various applications have gradually introduced item trading functions, that is, the application introduces merchants and allows each merchant to display goods in the corresponding module of the application so that users can purchase. Before purchasing, users always use the search function to search for other products that match the given pictures and texts to meet their own purchasing needs. At present, the image features are determined based on the image given by the user. Then the image features of all products are obtained from the product library, and the products are recalled from the entire product based on the image features of the image given by the user, so as to obtain the products that match the image given by the user. However, the recall ability of this method is limited and cannot meet the purchasing needs of users; furthermore, it is necessary to filter the text of each product from all products, which takes a long time.

[0029] Therefore, based on this, an embodiment of the present application provides a commodity recall method, which includes: ① obtaining multiple commodity clusters obtained by clustering commodities in a commodity database, and obtaining cluster features of each commodity cluster in the multiple commodity clusters; ② obtaining an index image associated with a target object, and determining at least one commodity cluster that matches the index image from multiple commodity clusters based on the image features of the index image and the cluster features of each commodity cluster. ③ determining the recalled commodity of the target object from the commodities included in at least one commodity cluster. In this method, cluster features are used to recall commodities, with a high recall rate and a fast recall speed. Because cluster features represent the features of a class of commodities, compared with the features of a single commodity, the recall rate of commodities recalled through cluster features is higher; in addition, because the number of cluster features is less than the number of features of a single commodity, the recall rate of commodities recalled through cluster features is faster.

[0030] See also Figure 1 , is a network architecture diagram of a product recall system disclosed in an embodiment of the present application, and the product recall system includes a terminal device 101, a server 102, and a database 103. The terminal device 101 and the server 102 are connected via a network, and the server 102 and the database 103 are connected via a network, for example, via a local area network, a wide area network, a mobile Internet, etc. It should be noted that the number of terminal devices 101 can be multiple, and the interaction between a terminal device 101 and a server 102 is taken as an example in the subsequent embodiments.

[0031] Among them, the terminal device 101 is a terminal device corresponding to the target object (i.e., user) who has a product recall demand. Usually, one terminal device corresponds to one target object. The terminal device 101 mainly sends index images to the server 102 and receives the recalled products returned by the server 102. The terminal device 101 in the embodiment of the present application includes but is not limited to: smart phones, tablet computers, smart wearable devices, smart voice interaction devices, smart home appliances, personal computers, vehicle-mounted terminals, smart cameras, virtual reality devices (such as AR (Augmented Reality, augmented reality) devices) and other devices, and this application does not limit this.

[0032] The server 102 is mainly used to determine the commodity clustering cluster, determine the clustering features of the commodity clustering cluster, and recall the commodity according to the image features of the index image and the clustering features of the commodity clustering cluster. Optionally, the commodity clustering cluster and the clustering features of each commodity clustering cluster in the multiple commodity clustering clusters can also be directly obtained from the database 103. After the server 102 recalls the commodity and obtains the recalled commodity, it returns the recalled commodity of the target object to the terminal device 101. Optionally, the server 102 can also obtain the index image associated with the target object from the database 103, and the index image is different for different target objects. The server 102 or database 103 in the embodiment of the present application can be an independent physical server, or a server cluster or distributed system composed of multiple physical servers, or a cloud server that provides cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communications, middleware services, domain name services, security services, CDN (Content Delivery Network), and basic cloud computing services such as big data and artificial intelligence platforms, which are not limited in the embodiment of the present application. Optionally, the database 103 may also be deployed in the server 102 .

[0033] Optionally, Figure 1 This is just an exemplary network architecture diagram of the product recall system and is not intended to be limiting. Figure 1 The server 102 in the blockchain network can be deployed as a node in the blockchain network, or the server 102 can be connected to the blockchain network, so that the server 102 can upload the recalled products of the target object to the blockchain network for storage to prevent the internal data from being tampered with, thereby ensuring data security.

[0034] In the embodiment of the present application, the product recall method provided in the embodiment of the present application is described in combination with the product recall system: multiple product clusters obtained by clustering the products in the product database, and the clustering features of each product cluster in the multiple product clusters are obtained; an index image associated with the target object is obtained, and at least one product cluster matching the index image is determined from the multiple product clusters according to the image features of the index image and the clustering features of each product cluster; the recalled products of the target object are determined from the products included in the at least one product cluster. In this method, since the clustering features represent the features of a class of products, the recall rate of products recalled through clustering features is higher than that of a single product; in addition, since the number of clustering features is less than the number of features of a single product, the recall rate of products recalled through clustering features is faster.

[0035] In a possible implementation, based on the product recall system, the product recall method provided in the embodiment of the present application can be applied to the following scenarios:

[0036] 1: User purchase intention recognition scenario. Specifically, the user enters search information in the search interface of the corresponding application, and the terminal device generates a product recall request based on the input search information; the terminal device sends the product recall request to the server, and the server obtains the index image matching it from the image database according to the search information, and then the server determines the recalled product corresponding to the index image according to the index image and the clustering features of the obtained product clusters, and returns the recalled product to the terminal device to meet the user's purchase needs.

[0037] 2: Product distribution scenario. Specifically, the server actively (including timing or real-time) obtains the index image of the interaction of the client corresponding to the user in the historical time period. The index image here is determined based on the note information that the user is interested in, wherein the note information that the user is interested in includes the note information collected by the user, the note information published by the user, and the note information with the most likes by the user. The note information that the user is interested in includes at least an image, and the included image is used as the index image. The server then determines the recalled product corresponding to the index image based on the index image and the clustering features of the obtained product clustering cluster, and returns it to the terminal device. When the user opens the product display interface through the terminal device, these recalled products will be displayed first to enhance the user experience.

[0038] 3. Popular product push scenario: Specifically, after recalling products based on index images of different target objects, popular products within a period of time are counted and sent to different merchants, so that merchants can purchase products based on the popular products.

[0039] It should be noted that the collection and processing of relevant data in this application should be strictly in accordance with the requirements of relevant laws and regulations when applied in examples, and the informed consent or separate consent of the personal information subject should be obtained, and subsequent data use and processing should be carried out within the scope of authorization of laws and regulations and the personal information subject.

[0040] The product recall method proposed in the embodiment of the present application will be introduced in detail below.

[0041] See also Figure 2 , is a flow chart of a product recall method disclosed in an embodiment of the present application. The product recall method is described with a server as the execution subject, which may be Figure 1 The server 102 in the embodiment should, correspondingly, the product recall method may include but is not limited to the following steps:

[0042] S201: Acquire multiple commodity clusters obtained by clustering commodities in a commodity database, and a clustering feature of each commodity cluster.

[0043] In a possible implementation, multiple commodity clusters are obtained by clustering commodities in a commodity database, each commodity cluster includes one or more commodities, and the commodity attribute information corresponding to the commodities included in each commodity cluster is the same. The commodity attribute information includes commodity type information and graphic information, wherein the commodity type information includes one or both of commodity brand and commodity category. The commodity brand indicates the brand of the commodity, such as brand A, brand B, etc.; the commodity category indicates which category the commodity belongs to, and the commodity categories include but are not limited to beauty, clothing, toys, digital, and home appliances, etc. The graphic information includes the commodity description information and image information of the commodity, wherein the graphic information of each commodity can be determined based on the graphic information. The graphic features of the commodities included in the same commodity cluster are the same, and the commodity type information is the same.

[0044] For each of the multiple commodity clusters, it is necessary to construct corresponding cluster features. Taking the construction of the cluster features of the kth commodity cluster in the multiple commodity clusters as an example, the process may include: obtaining the commodity image features corresponding to each commodity included in the kth commodity cluster; wherein, the commodity image features corresponding to the commodity may be obtained after the visual neural network model extracts the features of the commodity image. In the embodiment of the present application, the visual neural network model is a deep learning model with the function of extracting the features of the image, and the visual neural network model may be a pre-trained model. When determining the commodity cluster, the commodity image features corresponding to each commodity have been determined based on the visual neural network model, so here the corresponding commodity image features can be directly obtained from the database according to the commodity image. According to the commodity image features corresponding to each commodity included in the kth commodity cluster, the cluster features of the kth commodity cluster are determined; wherein, k is a positive integer. Assuming there are x commodity clusters, correspondingly, k is less than or equal to x (i.e., k=1,2,…,x), in the embodiment of the present application, x is an integer greater than 1. Since the number of features of commodities in the commodity database is much larger than the number of clustering features of commodity clusters, constructing a recall base database based on the clustering features of commodity clusters consumes fewer resources and can save data processing resources.

[0045] Further, assuming that the kth commodity cluster includes N commodities, N is a positive integer, correspondingly, according to the commodity image features corresponding to each commodity included in the kth commodity cluster, determining the clustering features of the kth commodity cluster may include the following methods:

[0046] Method 1: Determine the distance between the product image feature corresponding to the i-th product among N products and the product image features corresponding to each product except the i-th product among N products; i is a positive integer less than or equal to N; add the distance between the product image feature corresponding to the i-th product and the product image features corresponding to each other product to obtain the distance feature value of the i-th product; use the product image feature corresponding to the i-th product with the smallest distance feature value among N products as the cluster feature of the k-th product cluster. Since the distance between image features actually represents the similarity between two images, it is most accurate to select the image feature with the smallest distance feature value.

[0047] For example, Figure 3 is a schematic diagram of a commodity cluster disclosed in an embodiment of the present application. The commodity cluster includes three commodities, and the commodity image features of the three commodities are commodity image feature A, commodity image feature B, and commodity image feature C respectively. Correspondingly, the clustering feature determination process of the product cluster is as follows: for product image feature A, the distance between product image feature A and product image feature B, and the distance between product image feature A and product image feature C are calculated, which are AB and AC respectively, and then AB and AC are added to obtain the distance feature value of product image feature A; for product image feature B, the distance between product image feature B and product image feature A, and the distance between product image feature B and product image feature C are calculated, which are BA and BC respectively, and then BA and BC are added to obtain the distance feature value of product image feature B; for product image feature C, the distance between product image feature C and product image feature A, and the distance between product image feature C and product image feature B are calculated, which are CA and CB respectively, and then CA and CB are added to obtain the distance feature value of product image feature C; finally, the image feature corresponding to the smallest distance feature value is selected from the three distance feature values ​​as the clustering feature of the product cluster.

[0048] Method 2: Determine the average image feature value of the product image features of N products in the kth product cluster, and use the average image feature value as the cluster feature of the kth product cluster. The image feature average value here can be the image feature average value or the image feature weighted average value. The image feature average value is calculated by adding the N product image features and then dividing by N; the image feature weighted average value is calculated by multiplying each product image feature in the N product image features by the corresponding weight, adding them up and dividing by N. The weight corresponding to each product image feature can be customized or calculated based on each product image feature.

[0049] Assuming that the product image features of N products are {x1, x2, ..., xN}, the calculation formula for the average value of the image features is as follows (1):

[0050] Image feature average value = (x1+x2+...+xN) / N (1)

[0051] Assuming that the product image features of N products are {x1, x2, ..., xN}, and the weights corresponding to each product image feature are {k1, k2, ..., kN}, the calculation formula for the weighted average of the image features is as follows (2):

[0052] Weighted average of image features = (x1*k1+x2*k2+...+xN*kN) / N (2)

[0053] Furthermore, after determining the clustering features of each commodity cluster, the clustering features of each commodity cluster can be combined into a clustering feature base library, and the clustering feature base library is stored in the commodity database. When a commodity recall is required, the clustering feature base library can be directly obtained from the commodity database. In addition, the clustering feature base library can also be updated regularly, that is, the clustering feature determination process of the above commodity cluster is executed regularly, so as to achieve the purpose of regularly updating the clustering feature base library. Since images are the most intuitive form of expression when purchasing commodities, the accuracy of recall can be improved by constructing clustering features through commodity image features.

[0054] S202: Acquire an index image associated with the target object, and determine at least one commodity cluster matching the index image from a plurality of commodity clusters according to image features of the index image and cluster features of each commodity cluster.

[0055] The target object refers to any user who uses a target application, and the target application includes dating applications, shopping applications, and audio and video applications, etc. The index image is used to instruct the server to perform a search based on the index image.

[0056] In a possible implementation, the index image can be determined according to the product recall request sent by the client corresponding to the target object. The specific process includes: the server receives the product recall request sent by the client corresponding to the target object, wherein the product recall request carries search information, the search information is input by the target object in the input interface of the client, and the search information includes but is not limited to one or more of text information, image information, and audio and video information, that is, the search information can be one of text information, image information, and audio and video information, or it can be a combination of two or three of text information, image information, and audio and video information. Then, after receiving the product recall request, the server searches the image database for an index image that matches the search information carried by the product recall request. In this case, the index image is actively sent by the target object and can be applied to the user's purchase intention recognition scenario. Since the embodiment of the present application recalls the product through clustering features, in this scenario, the search engine has a faster processing speed and can return the recalled product to the user more quickly.

[0057] In another possible implementation, the index image may also be an image of the client corresponding to the target object interacting in a historical time period. The process of determining the index image includes: the server periodically or in real time obtains the interactive images of the client corresponding to the target object interacting in a historical time period. These interactive images are all directly interacting with the target object. For example, the interactive images may be images posted by the target object, images collected by the target object, or images liked by the target object. The index image is selected from the interactive images. A certain number of interactive images can be selected from the interactive images in the time sequence of time interaction as index images; or the interactive images corresponding to the type with the most interactions can be selected as index images according to the type of interactive images. It should be noted that, in addition to images, the interactive objects of the target object may also have videos. In this case, the interactive image is the cover image of the video. In this case, the index image is actively obtained by the server and can be applied to commodity distribution scenarios. The server can actively recommend corresponding commodities to users by recalling commodities, which is in line with the user's taste, and can adjust the recommended commodities in real time to improve the user's usage experience.

[0058] In a possible implementation, after obtaining an index image associated with a target object, an image feature extraction is performed on the index image using a visual neural network model to obtain the image feature of the index image. Then, further, according to the image feature of the index image and the cluster feature of each commodity cluster, at least one commodity cluster matching the index image is determined from multiple commodity clusters. The process may specifically include: the server may first determine the similarity between the image feature of the index image and the cluster feature of each commodity cluster. If the number of commodity clusters is X, there are X similarities. The commodity cluster corresponding to the cluster feature whose similarity with the image feature is greater than a first similarity threshold is used as at least one commodity cluster matching the index image. In this manner, the number of determined commodity clusters is related to the first similarity threshold. The larger the first similarity threshold, the closer the determined commodity cluster is to the index image. The first similarity threshold can be flexibly adjusted to improve the accuracy of recalling commodities. If a more accurate commodity cluster is to be obtained, the first similarity threshold is adjusted higher.

[0059] In another possible implementation, at least one commodity cluster matching the index image can also be selected according to a set value. If, for example, there are X commodity clusters, X similarities can be determined. Then, the process of determining at least one commodity cluster matching the index image can also include: arranging the X similarities in descending order, and selecting the commodity clusters corresponding to the clustering features corresponding to the preset number of similarities that are ranked at the front as the at least one commodity cluster matching the index image. The preset number can be flexibly adjusted, for example, it can be set to 10 or 20, etc. The smaller the number is, the faster the processing speed and the faster the recall speed are.

[0060] S203: Determine the recalled products of the target object from the products included in at least one product cluster.

[0061] In order to improve the accuracy of recalled products, after determining at least one product cluster matching the index image, text information associated with the index image can be obtained, and then the at least one product cluster can be screened using the text information to obtain a more accurate product cluster.

[0062] In a possible implementation, if the index image is determined based on the search information carried by the product recall request, the process of obtaining the text information associated with the index image is as follows: first, check whether the search information contains text information. If the search information contains text information, use the text information as the text information associated with the index image; if the search information does not contain text information, check whether the obtained index image is configured with text information. If the index image is configured with text information, use the text information configured with the index image as the text information associated with the index image. If the search information contains text information and the index image is also configured with text information, use the text information in the search information as the text information associated with the index image. Since the search information is input by the target object itself and has a stronger correlation with the index image, the text information in the search information is preferred. Optionally, the text information in the search information can also be combined as the text information associated with the index image to obtain the text information associated with the index image. If the search information does not contain text information and the index image is not configured with text information, the step of directly determining the recalled product of the target object from the products included in at least one product cluster is performed.

[0063] In a possible implementation, if the index image is determined based on an interactive image of the client corresponding to the target object interacting in a historical time period, the process of obtaining the text information associated with the index image is as follows: first determine the interactive image corresponding to the index image, then determine whether the interactive image has text information, and if the interactive image has text information, use the text information corresponding to the interactive image as the text information associated with the index image. If the interactive image does not have text information, directly perform the step of determining the recalled product of the target object from the products included in at least one product cluster.

[0064] After obtaining the text information associated with the index image, the product description information in the text information can be identified, and the product type can be identified for the text information. Then, at least one product cluster is screened by the product description information and / or product type information. The product description information is used to describe the style, material, color, etc. of the product, and the product type information is used to describe the brand and / or category of the product. The product type information includes one or both of the product brand and the product category.

[0065] In a possible implementation, if the text information contains both commodity description information corresponding to the index image and commodity type information corresponding to the index image, the process of screening at least one commodity clustering cluster through the text information includes: extracting text features from the commodity description information to obtain the first text features of the index image; obtaining the second text features of each commodity clustering cluster in the at least one commodity clustering cluster; determining the similarity between the first text feature and each second text feature, and taking the commodity clustering cluster corresponding to the second text feature with a similarity greater than a second similarity threshold as the first commodity clustering cluster; obtaining the commodity type information of each first commodity clustering cluster, and then comparing the commodity type information corresponding to the index image with the commodity type information of each first commodity clustering cluster to determine the second commodity clustering cluster whose commodity type information in the first commodity clustering cluster matches the commodity type information corresponding to the index image. The commodity type information includes one or both of commodity brand information and commodity category information. The commodity type information in the first commodity clustering cluster matches the commodity type information corresponding to the index image, which means that the commodity brand information and the commodity category information completely match. If the commodity type information corresponding to the index image includes commodity brand information, then matching means that the commodity brand information of the first commodity cluster is the same as the commodity brand information of the index image; if the commodity type information corresponding to the index image includes commodity category information, then matching means that the commodity category information of the first commodity cluster is the same as the commodity category information of the index image; if the commodity type information corresponding to the index image includes commodity brand information and commodity category information, then matching means that the commodity brand information and commodity category information of the first commodity cluster are the same as the commodity brand information and commodity category information of the index image. After screening, the recalled commodities of the target object are determined from the commodities included in the second commodity cluster, specifically, the commodities included in each second commodity cluster are obtained, and then the commodities included in each second commodity cluster are compared with the index image to obtain the recalled commodities of the target object.

[0066] In another possible implementation, if the text information contains both product description information corresponding to the index image and product type information corresponding to the index image, the process of screening at least one product clustering cluster through text information may also include: obtaining product type information of each product clustering cluster in the at least one product clustering cluster, and then comparing the product type information corresponding to the index image with the product type information of each product clustering cluster in the at least one product clustering cluster, and determining a third product clustering cluster in which the product type information in the product clustering cluster matches the product type information corresponding to the index image. Then, text feature extraction is performed on the product description information corresponding to the index image to obtain the first text feature of the index image; and the third text feature of each third product cluster in the third product cluster is obtained; the similarity between the first text feature and each third text feature is determined, and the third product cluster corresponding to the third text feature whose similarity is greater than the second similarity threshold is used as the fourth product cluster; after screening, the recalled products of the target object are determined from the products included in the fourth product cluster, specifically, the products included in each fourth product cluster are obtained, and then the products included in each fourth product cluster are compared with the index image to obtain the recalled products of the target object.

[0067] In a possible implementation, if there is commodity description information corresponding to the index image in the text information, but there is no commodity type information corresponding to the index image, then the process of screening at least one commodity clustering cluster by using the text information includes: extracting text features from the commodity description information corresponding to the index image to obtain the first text feature of the index image; obtaining the second text feature of each commodity clustering cluster in the at least one commodity clustering cluster; determining the similarity between the first text feature and each second text feature, and taking the commodity clustering cluster corresponding to the second text feature whose similarity is greater than the second similarity threshold as the target commodity clustering cluster. Since there is no commodity type information corresponding to the index image in the text information, the commodity description information corresponding to the index image in the text information can be directly used to screen at least one commodity clustering cluster. After screening, the recalled commodities of the target object are determined from the commodities included in the target commodity clustering cluster. Specifically, the commodities included in each target commodity clustering cluster can be obtained in a timely manner, and then the commodities included in each target commodity clustering cluster and the index image are compared to obtain the recalled commodities of the target object.

[0068] In a possible implementation, if the text information does not contain the commodity description information corresponding to the index image, but contains the commodity type information corresponding to the index image, the process of screening at least one commodity clustering cluster by using the text information includes: obtaining the commodity type information of each commodity clustering cluster in the at least one commodity clustering cluster, and then comparing the commodity type information corresponding to the index image with the commodity type information of each commodity clustering cluster in the at least one commodity clustering cluster, and determining the target commodity clustering cluster whose commodity type information in the commodity clustering cluster matches the commodity type information corresponding to the index image. Since the text information does not contain the commodity description information corresponding to the index image, the at least one commodity clustering cluster can be screened directly by using the commodity type information corresponding to the index image in the text information. After screening, the recalled commodities of the target object are determined from the commodities included in the target commodity clustering cluster.

[0069] In a possible implementation, if the text information does not contain the product description information corresponding to the index image, nor does it contain the product type information corresponding to the index image, then there is no need to further screen the at least one product cluster, and the recalled products of the target object can be directly determined from the products included in the at least one product cluster. The specific process of determining the recalled products of the target object from the products included in the at least one product cluster can be: obtaining the products included in each product cluster in the at least one product cluster, and then comparing the products included in each product cluster with the index image to obtain the recalled products of the target object. Directly compare the images to filter out some obviously different products.

[0070] The above describes how to filter at least one commodity cluster according to text information. After filtering at least one commodity cluster according to the commodity description information and / or commodity type information corresponding to the index image, a commodity cluster that is more relevant (closer) to the index image can be obtained, thereby improving the accuracy of commodity recall. In the above process, commodity clusters are recalled through clustering features. Compared with directly recalling commodities through commodity features, recalling commodity clusters through clustering features and then recalling commodities under the same recall quantity has a stronger recall capability. For example, if the recall quantity is set to 10, only 10 commodities can be recalled directly through commodity features, but the recalled commodities obtained by recalling commodity clusters through clustering features are far more than 10.

[0071] In the above process, the second text feature of the commodity cluster is also involved. The following describes how to obtain the second text feature of each commodity cluster in at least one commodity cluster. Taking the zth commodity cluster in at least one commodity cluster as an example, assuming that the zth commodity cluster includes N commodities, N is a positive integer, correspondingly, obtaining the second text feature of the zth commodity cluster may include the following methods:

[0072] Method 1: Determine the distance between the product text feature corresponding to the ith product among N products and the product text features corresponding to each of the N products except the ith product; i is a positive integer less than or equal to N; add the distances between the product text feature corresponding to the ith product and the product text features corresponding to each of the other products to obtain the text distance feature value of the ith product; use the product text feature corresponding to the ith product with the smallest text distance feature value among the N products as the second text feature of the zth product cluster.

[0073] Method 2: Determine the average value of the text features of the N products in the zth product cluster, and use the average value of the text features as the second text feature of the zth product cluster. The text feature average value here can be the text feature average value or the text feature weighted average value. The text feature average value is calculated by adding the N product text features and then dividing by N; the text feature weighted average value is calculated by multiplying each product text feature in the N product text features by the corresponding weight, adding them up and then dividing by N. The weight corresponding to each product text feature can be customized or calculated based on each product text feature.

[0074] Furthermore, the present embodiment can be described in combination with different application scenarios:

[0075] 1. User purchase intention recognition scenario: Receive a product recall request sent by the client corresponding to the target object; the product recall request carries search information, search for an index image matching the search information in the image database, and extract image features of the index image to obtain image features of the index image; obtain multiple product clusters, and obtain cluster features of each product cluster in the multiple product clusters; determine at least one product cluster matching the index image from the multiple product clusters based on the image features of the index image and the cluster features of each product cluster; determine the recalled product of the target object from the products included in at least one product cluster. Finally, return the recalled product to the client to meet the purchase needs of the target object.

[0076] 2. Product distribution scenario: The server obtains an index image of the interaction of the client corresponding to the target object within a historical time period, and extracts image features of the index image to obtain image features of the index image; obtains multiple product clusters, and obtains cluster features of each product cluster in the multiple product clusters; determines at least one product cluster that matches the index image from the multiple product clusters based on the image features of the index image and the cluster features of each product cluster; determines the recalled products of the target object from the products included in at least one product cluster, and finally returns the recalled products to the client. When the user opens the product display interface through the client, these recalled products will be displayed first to enhance the user experience.

[0077] 3. Popular product push scenario: According to the set cycle, the recalled products determined based on the above two scenarios are counted, and the recalled products are sorted according to the number of recalls of each recalled product. The top y products are selected, and the product type information of each of the y products is determined. The y products are sent to the merchants corresponding to the product type information, so that the merchants can purchase products according to the popular products.

[0078] In the embodiments of the present application, it is mainly explained how to determine the clustering features of each commodity clustering cluster, and how to recall commodities based on the image features of the index image and the clustering features of each commodity clustering cluster. In the process of commodity recall, since the clustering features represent the features of a category of commodities, the recall rate of commodities through the clustering features is higher than that of a single commodity. In addition, since the number of clustering features is less than the number of features of a single commodity, the recall rate of commodities through the clustering features is faster.

[0079] Based on the above embodiments, the present embodiment further combines the construction of commodity clusters to explain the commodity recall method of the present embodiment, see Figure 4 , is a flow chart of another product recall method disclosed in an embodiment of the present application. The product recall method is described with a server as the execution subject, and can be specifically Figure 1 In the server 102, correspondingly, the product recall method may include but is not limited to the following steps:

[0080] S401: Obtain product information of M products in a product database, where M is an integer greater than 1.

[0081] Among them, the commodity database corresponds to the application program, and different applications correspond to different commodity databases. The commodity database includes a huge number of commodities, and M commodities and commodity information of M commodities can be determined by selecting a large number of commodities. Usually, the filtered commodities are defective or sold through informal channels. Among them, the commodity information includes commodity images and commodity description information, and the commodity description information can specifically include the commodity title and the commodity introduction. After obtaining M commodity images, the size of the commodity images can be adjusted, and the commodity description text can be filtered. The size adjustment includes cropping, padding, etc., and its purpose is to unify the size of each commodity image. Text filtering is mainly to delete useless information in the commodity description information, specifically to determine the part of speech of the commodity description information, which includes verbs, nouns, adjectives and adverbs, and retain the text part of the commodity description information whose part of speech is noun.

[0082] S402: extracting features from the product description information of the M products to obtain product features of the M products, where the product features include one or more of image features, text features, and product type features.

[0083] In a possible implementation, the server may perform image feature extraction processing on the processed product images through a visual neural network model to obtain image features of each product image. Correspondingly, the server may also perform text feature extraction processing on the processed product description information through a text neural network model to obtain text features of each product. Correspondingly, the server may also perform attribute feature extraction processing on the processed product description information through a named entity model to obtain product type features of each product. Furthermore, the obtained image features, text features, and product type features of each product are collectively referred to as product features of the product.

[0084] Among them, the visual neural network model, the text neural network model and the named entity model can be pre-trained models. Optionally, the visual neural network model, the text neural network model and the named entity model can also be currently commonly used network models, such as neural network models. For example, the named entity model can be a model based on a deep neural network architecture (Biomedical named entity recognition based on deep neutral network, Bio-NER), referred to as a NER model.

[0085] S403: Clustering the M commodities based on their characteristics to obtain a plurality of commodity clusters.

[0086] Determine the similarity between the product features of any two products among the M products, including: determining a first similarity between the image features of any two products, and determining a second similarity between the text features of any two products. Based on the first similarity and the second similarity, determine whether any two products are of the same category of products. If the first similarity is greater than a first value, and the second similarity is greater than a second value, then determine that the two products are similar. Further, determine whether the brands and / or brands of any two products are the same according to the product type information of the two products. If they are the same, then determine the two products as a product cluster.

[0087] For example, there are two commodities, commodity A and commodity B. The first similarity between the image features of commodity A and the image features of commodity B is determined, and the second similarity between the text features of commodity A and the text features of commodity B is determined. If the first similarity is greater than the first value, and the second similarity is greater than the second value, it is preliminarily determined that commodity A and commodity B are similar, and then the commodity type of commodity A is compared with the commodity type of commodity B to see if they are consistent. If they are consistent, commodity A and commodity B are added to the same commodity cluster.

[0088] In a possible implementation, an adjacency graph may be constructed based on image features corresponding to the M products, text features corresponding to the M products, and product type features corresponding to the M products. The adjacency graph includes connecting edges, and two products connected by the connecting edges have an association relationship. Then, the first similarity and the second similarity of any two products are determined based on the adjacency graph.

[0089] Among them, steps S401-S403 describe the process of determining the commodity clusters, which can also be summarized as follows: Figure 5 The framework diagram shown, Figure 5 This is a framework diagram for determining commodity clusters disclosed in the embodiment of the present application, including a commodity selection part, an algorithm recognition part, a clustering part, an image segmentation part, and an application part. The commodity selection part mainly selects some compliant commodities, such as Figure 5As shown, it includes the product image, product description information and directly marked product type information (brand, category, some products will directly mark this information) of each product. The algorithm recognition part mainly shows multiple models, namely the visual neural network model, the text neural network model and the named entity model. Through these models, the image features, text features and type features of each product can be determined. The image features and text features are collectively referred to as image and text features. The clustering part is clustered by determining the similarity of each feature. The products included in each category have the same image and text features, the same brand features and the same category features. Among them, in the clustering part, the two products connected by the dotted line belong to different products (including different image and text features, different brands or different categories), and the two connected products belong to the same product (including the same image and text features, the same brand and the same category). The image segmentation part cuts out the images in each product cluster, and distinguishes each product cluster by the product type information. The application part is to apply each product cluster to the user purchase intention recognition scenario, product distribution scenario and popular product push scenario after determining each product cluster. Through Figure 5 , it can be clearly understood how the commodity clusters are determined and the application of each commodity cluster.

[0090] S404: Constructing a clustering feature of each commodity cluster in the plurality of commodity clusters.

[0091] Based on each commodity in each commodity cluster, first determine the commodity image of each commodity, then use the visual neural network model to extract the image features of each commodity image, and then determine the clustering features of each commodity cluster based on the image features. The specific clustering feature determination process can be referred to step S201, which will not be repeated here.

[0092] S405: Acquire an index image associated with the target object.

[0093] S406: Determine at least one commodity cluster matching the index image from the plurality of commodity clusters according to the image features of the index image and the cluster features of each commodity cluster.

[0094] S407: Acquire text information associated with the index image, and use the text information to filter at least one commodity cluster to obtain a target commodity cluster.

[0095] S408: Determine the recalled products of the target object from the products included in the target product cluster.

[0096] Steps S404-S407 describe how to determine the clustering characteristics of the commodity clusters, and how to determine the process of recalling commodities based on the clustering characteristics of the commodity clusters and the image characteristics of the index image. The details of the process can be found in steps S202 and S203, and will not be repeated here.

[0097] In a possible implementation, the overall process of steps S404-S407 can be found in Figure 6 , is a framework diagram of a product recall disclosed in an embodiment of the present application, which is specifically divided into two parts, the first part is the construction of clustering features, and the second part is product recall. Among them, the process of constructing clustering features includes: determining the product image of each product from the product clustering cluster, then using the visual neural network model to extract image features of the product image of each product to obtain the image features of each product, and finally determining the clustering features of the corresponding product clustering cluster according to the image features of each product. The product recall process includes: first using the visual neural network model to extract image features of the index image to obtain the image features of the index image, recalling the product clustering cluster according to the image features of the index image and the clustering features of each product clustering cluster, and obtaining at least one recalled product clustering cluster; then using the text neural network model to extract text features of the text to obtain the text features of the index image, and at the same time, using the named entity model to identify the product type of the text information to obtain the corresponding product type information, and then screening at least one product clustering cluster according to the text features and the product type information to obtain the target product clustering cluster, and finally determining the recalled product corresponding to the index image according to the target product clustering cluster.

[0098] In the present application embodiment, combined Figure 5 and Figure 6 The framework diagram shown illustrates the product recall method of the embodiment of the present application. When the product is recalled through this method, on the one hand, since the clustering features represent the characteristics of a category of products, compared with the characteristics of a single product, the recall rate of products recalled through clustering features is higher; on the other hand, since the number of clustering features is less than the number of features of a single product, the recall rate of products recalled through clustering features is faster.

[0099] Based on the above method embodiment, the present application embodiment also provides a product recall device. Figure 7 , is a schematic diagram of the structure of a product recall device provided in an embodiment of the present application, Figure 7 The product recall device shown can operate the following units:

[0100] An acquisition unit 701 is used to acquire a plurality of commodity clusters obtained by clustering commodities in a commodity database, and to acquire a clustering feature of each commodity cluster in the plurality of commodity clusters;

[0101] The acquisition unit 701 is further configured to acquire an index image associated with the target object, and determine at least one commodity cluster matching the index image from the plurality of commodity clusters according to image features of the index image and cluster features of each commodity cluster;

[0102] The determining unit 702 is configured to determine the recalled product of the target object from the products included in the at least one product cluster.

[0103] In a possible implementation, a method for obtaining a clustering feature of a kth commodity cluster among multiple commodity clusters includes:

[0104] Obtaining a product image feature corresponding to each product included in the kth product cluster;

[0105] The clustering feature of the kth commodity cluster is determined according to the commodity image feature corresponding to each commodity included in the kth commodity cluster; wherein k is a positive integer.

[0106] In a possible implementation, the kth commodity cluster includes N commodities, where N is a positive integer. When the determination unit 702 determines the clustering feature of the kth commodity cluster according to the commodity image feature corresponding to each commodity included in the kth commodity cluster, it is specifically configured to perform the following steps:

[0107] Determine respectively the distance between the product image feature corresponding to the i-th product among the N products and the product image features corresponding to each of the N products except the i-th product; i is a positive integer less than or equal to N;

[0108] Adding the distances between the product image feature corresponding to the i-th product and the product image features corresponding to each of the other products to obtain a distance feature value of the i-th product;

[0109] The product image feature corresponding to the i-th product with the smallest distance feature value among the N products is used as the cluster feature of the k-th product cluster.

[0110] In a possible implementation, when the determining unit 702 determines at least one commodity cluster matching the index image from the multiple commodity clusters according to the image feature of the index image and the cluster feature of each commodity cluster, the determining unit 702 is specifically configured to perform the following steps:

[0111] Determining the similarity between the image feature of the index image and the cluster feature of each commodity cluster;

[0112] The commodity cluster corresponding to the cluster feature whose similarity with the image feature is greater than a first similarity threshold is used as at least one commodity cluster matching the index image.

[0113] In a possible implementation, the acquisition unit 701 is further configured to acquire text information associated with the index image;

[0114] The processing unit 703 is used to perform the following steps: identify the commodity description information in the text information to obtain the commodity description information corresponding to the index image; extract text features from the commodity description information to obtain the first text feature of the index image;

[0115] The acquisition unit 701 is further configured to acquire a second text feature of each commodity cluster in the at least one commodity cluster;

[0116] The determination unit 702 is further configured to determine the similarity between the first text feature and each second text feature, and to use the commodity cluster corresponding to the second text feature having a similarity greater than a second similarity threshold as the first commodity cluster;

[0117] When the determination unit 702 determines the recalled product of the target object from the products included in the at least one product cluster, it is specifically configured to perform the following steps:

[0118] The recalled products of the target object are determined from the products included in the first product cluster.

[0119] In a possible implementation, the processing unit 703 is further configured to perform commodity type recognition processing on the text information to obtain commodity type information corresponding to the index image;

[0120] The determination unit 702 is further configured to compare the commodity type information corresponding to the index image with the commodity type information of each first commodity cluster, and determine a second commodity cluster whose commodity type information in the first commodity cluster matches the commodity type information corresponding to the index image;

[0121] When the determination unit 702 determines the recalled product of the target object from the products included in the first product cluster, it is specifically configured to perform the following steps:

[0122] The recalled products of the target object are determined from the products included in the second product cluster.

[0123] In a possible implementation, when the acquisition unit 701 acquires the index image associated with the target object, it is specifically configured to perform the following steps:

[0124] Receiving a product recall request sent by a client corresponding to the target object; wherein the product recall request carries search information;

[0125] Searching the image database for an index image that matches the search information.

[0126] In a possible implementation, when the acquisition unit 701 acquires the index image associated with the target object, it is specifically configured to perform the following steps:

[0127] An index image of the interaction performed by the client corresponding to the target object within a historical time period is obtained.

[0128] It can be understood that the functions of each functional unit of the product recall device provided in the embodiment of the present application can be specifically implemented according to the method in the above method embodiment, and its specific implementation process can refer to the relevant description in the above method embodiment, which will not be repeated here.

[0129] In a feasible embodiment, the product recall device provided in the embodiment of the present application can be implemented in software. The product recall device can be stored in a memory. It can be software in the form of programs and plug-ins, and includes a series of units, including an acquisition unit, a determination unit, and a processing unit; wherein the acquisition unit, the determination unit, and the processing unit are used to implement the product recall method provided in the embodiment of the present application.

[0130] In other feasible embodiments, the product recall device provided in the embodiments of the present application may also be implemented in a combination of software and hardware. As an example, the product recall device provided in the embodiments of the present application may be a processor in the form of a hardware decoding processor, which is programmed to execute the product recall method provided in the embodiments of the present application. For example, the processor in the form of a hardware decoding processor may adopt one or more application-specific integrated circuits (ASICs), DSPs, programmable logic devices (PLDs), complex programmable logic devices (CPLDs), field-programmable gate arrays (FPGAs), or other electronic components.

[0131] In the embodiment of the present application, in combination with the acquisition unit, the determination unit and the processing unit, the product recall method includes: the acquisition unit acquires multiple product clusters obtained by clustering the products in the product database, and acquires the clustering features of each product cluster in the multiple product clusters; acquires an index image associated with the target object; the determination unit determines at least one product cluster that matches the index image from the multiple product clusters according to the image features of the index image and the clustering features of each product cluster; and determines the recalled products of the target object from the products included in the at least one product cluster. In this method, on the one hand, since the clustering features represent the features of a category of products, the recall rate of products recalled through clustering features is higher than that of a single product; on the other hand, since the number of clustering features is less than the number of features of a single product, the recall rate of products recalled through clustering features is faster.

[0132] See also Figure 8 , Figure 8 801, a communication interface 802, and a storage device 803. The processor 801, the communication interface 802, and the storage device 803 may be connected via a bus or other means. The embodiment of the present application takes the bus connection as an example. The computer device may be Figure 1 Server 102 is shown.

[0133] Among them, the processor 801 (or CPU (Central Processing Unit)) is the computing core and control core of the computer device, which can parse various instructions in the computer device and process various data of the computer device. For example, the CPU can be used to parse the power on and off instructions sent by the user to the computer device, and control the computer device to perform power on and off operations; for another example, the CPU can transmit various interactive data between the internal structures of the computer device, and so on. The communication interface 802 can optionally include a standard wired interface, a wireless interface (such as Wi-Fi, a mobile communication interface, etc.), which is controlled by the processor 801 to send and receive data. The storage device 803 (Memory) is a memory device in the computer device for storing programs and data. It can be understood that the storage device 803 here can include both the built-in memory of the computer device and the extended memory supported by the computer device. The storage device 803 provides storage space, which stores the operating system of the computer device, including but not limited to: Android system, iOS system, Windows Phone system, etc., and this application does not limit this.

[0134] In the embodiment of the present application, the processor 801 executes the following operations by running the executable program code in the storage device 803:

[0135] Acquire multiple commodity clusters obtained by clustering commodities in a commodity database, and acquire clustering features of each commodity cluster in the multiple commodity clusters;

[0136] Acquire an index image associated with the target object, and determine at least one commodity cluster matching the index image from the plurality of commodity clusters according to image features of the index image and cluster features of each commodity cluster;

[0137] The recalled products of the target object are determined from the products included in the at least one product cluster.

[0138] In a possible implementation, a method for obtaining a clustering feature of a kth commodity cluster among multiple commodity clusters includes:

[0139] Obtaining a product image feature corresponding to each product included in the kth product cluster;

[0140] The clustering feature of the kth commodity cluster is determined according to the commodity image feature corresponding to each commodity included in the kth commodity cluster; wherein k is a positive integer.

[0141] In a possible implementation, the kth commodity cluster includes N commodities, where N is a positive integer. When the processor 801 determines the clustering feature of the kth commodity cluster according to the commodity image feature corresponding to each commodity included in the kth commodity cluster by running the executable program code in the storage device 803, the processor 801 is specifically used to perform the following steps:

[0142] Determine respectively the distance between the product image feature corresponding to the i-th product among the N products and the product image features corresponding to each of the N products except the i-th product; i is a positive integer less than or equal to N;

[0143] Adding the distances between the product image feature corresponding to the i-th product and the product image features corresponding to each of the other products to obtain a distance feature value of the i-th product;

[0144] The product image feature corresponding to the i-th product with the smallest distance feature value among the N products is used as the cluster feature of the k-th product cluster.

[0145] In a possible implementation, the processor 801 is configured to execute the following steps when determining at least one commodity cluster matching the index image from the plurality of commodity clusters according to the image features of the index image and the cluster features of each commodity cluster by running the executable program code in the storage device 803:

[0146] Determining the similarity between the image feature of the index image and the cluster feature of each commodity cluster;

[0147] The commodity cluster corresponding to the cluster feature whose similarity with the image feature is greater than a first similarity threshold is used as at least one commodity cluster matching the index image.

[0148] In a possible implementation, the processor 801 is further configured to execute the following steps by running the executable program code in the storage device 803:

[0149] Acquiring text information associated with the index image;

[0150] Identify the commodity description information in the text information to obtain the commodity description information corresponding to the index image; extract text features from the commodity description information to obtain a first text feature of the index image;

[0151] Acquire a second text feature of each commodity cluster in the at least one commodity cluster;

[0152] Determine the similarity between the first text feature and each second text feature, and use the commodity cluster corresponding to the second text feature whose similarity is greater than a second similarity threshold as the first commodity cluster;

[0153] When the processor 801 determines the recalled product of the target object from the products included in the at least one product cluster by running the executable program code in the storage device 803, it is specifically configured to perform the following steps:

[0154] The recalled products of the target object are determined from the products included in the first product cluster.

[0155] In a possible implementation, the processor 801 is further configured to execute the following steps by running the executable program code in the storage device 803:

[0156] Perform commodity type recognition processing on the text information to obtain commodity type information corresponding to the index image;

[0157] Comparing the commodity type information corresponding to the index image with the commodity type information of each first commodity cluster, and determining a second commodity cluster whose commodity type information in the first commodity cluster matches the commodity type information corresponding to the index image;

[0158] When the processor 801 determines the recalled product of the target object from the products included in the first product cluster by running the executable program code in the storage device 803, it is specifically configured to perform the following steps:

[0159] The recalled products of the target object are determined from the products included in the second product cluster.

[0160] In a possible implementation, when the processor 801 executes the executable program code in the storage device 803 to obtain the index image associated with the target object, it is specifically configured to perform the following steps:

[0161] Receiving a product recall request sent by a client corresponding to the target object; wherein the product recall request carries search information;

[0162] Searching the image database for an index image that matches the search information.

[0163] In a possible implementation, when the processor 801 executes the executable program code in the storage device 803 to obtain the index image associated with the target object, it is specifically configured to perform the following steps:

[0164] An index image of the interaction performed by the client corresponding to the target object within a historical time period is obtained.

[0165] In the embodiment of the present application, the processor 801 executes the following by running the executable program code in the storage device 803: obtaining multiple commodity clusters obtained by clustering commodities in the commodity database, and obtaining the clustering features of each commodity cluster in the multiple commodity clusters; obtaining an index image associated with the target object; determining at least one commodity cluster that matches the index image from the multiple commodity clusters according to the image features of the index image and the clustering features of each commodity cluster; and determining the recalled commodities of the target object from the commodities included in the at least one commodity cluster. In this method, on the one hand, since the clustering features represent the features of a class of commodities, the recall rate of commodities through the clustering features is higher than that of a single commodity; on the other hand, since the number of clustering features is less than the number of features of a single commodity, the recall rate of commodities through the clustering features is faster.

[0166] According to one aspect of the present application, a computer program product is provided, the computer program product comprising a computer program, the computer program being stored in a computer-readable storage medium. A processor of a computer device reads the computer program from the computer-readable storage medium, and the processor executes the computer program, so that the computer device executes the product recall method provided in the above-mentioned various optional modes.

[0167] It should be noted that, for the above-mentioned method embodiments, for the sake of simplicity, they are all expressed as a series of action combinations, but those skilled in the art should be aware that the present application is not limited by the order of the actions described, because according to the present application, some steps may be performed in other orders or simultaneously. Secondly, those skilled in the art should also be aware that the embodiments described in the specification are all preferred embodiments, and the actions and modules involved are not necessarily required by the present application.

[0168] In the several embodiments provided in this application, it should be understood that the disclosed device can be implemented in other ways. For example, the device embodiments described above are only schematic, and the division of the modules described above is only a logical function division. There may be other division methods in actual implementation, such as multiple modules or components can be combined or integrated into another system, or some features can be ignored or not executed.

[0169] The above is only a specific implementation of the present application, but the protection scope of the present application is not limited thereto. Any person skilled in the art who is familiar with the present technical field can easily think of changes or substitutions within the technical scope disclosed in the present application, which should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be based on the protection scope of the claims.

Claims

1. A product recall method, characterized in that: include: Acquire multiple commodity clusters obtained by clustering commodities in a commodity database, and acquire clustering features of each commodity cluster in the multiple commodity clusters; Acquire an index image associated with the target object, and determine at least one commodity cluster matching the index image from the plurality of commodity clusters according to image features of the index image and cluster features of each commodity cluster; The recalled products of the target object are determined from the products included in the at least one product cluster.

2. The method according to claim 1, characterized in that The method for obtaining the clustering feature of the kth commodity cluster among the multiple commodity clusters includes: Obtaining a product image feature corresponding to each product included in the kth product cluster; The clustering feature of the kth commodity cluster is determined according to the commodity image feature corresponding to each commodity included in the kth commodity cluster; wherein k is a positive integer.

3. The method according to claim 2, characterized in that The k-th commodity cluster includes N commodities, where N is a positive integer, and the determining of the clustering features of the k-th commodity cluster according to the commodity image features corresponding to each commodity included in the k-th commodity cluster includes: Determine respectively the distance between the product image feature corresponding to the i-th product among the N products and the product image features corresponding to each of the N products except the i-th product; i is a positive integer less than or equal to N; Adding the distances between the product image feature corresponding to the i-th product and the product image features corresponding to each of the other products to obtain a distance feature value of the i-th product; The product image feature corresponding to the i-th product with the smallest distance feature value among the N products is used as the cluster feature of the k-th product cluster.

4. The method according to claim 1, characterized in that The determining, based on the image features of the index image and the cluster features of each of the commodity clusters, at least one commodity cluster matching the index image from the plurality of commodity clusters comprises: Determining the similarity between the image feature of the index image and the cluster feature of each commodity cluster; The commodity cluster corresponding to the cluster feature whose similarity with the image feature is greater than a first similarity threshold is used as at least one commodity cluster matching the index image.

5. The method according to any one of claims 1 to 4, characterized in that: The method further comprises: Acquiring text information associated with the index image; Identify the commodity description information in the text information to obtain the commodity description information corresponding to the index image; Performing text feature extraction on the commodity description information to obtain a first text feature of the index image; Acquire a second text feature of each commodity cluster in the at least one commodity cluster; Determine the similarity between the first text feature and each second text feature, and use the commodity cluster corresponding to the second text feature whose similarity is greater than a second similarity threshold as the first commodity cluster; The step of determining the recalled product of the target object from the products included in the at least one product cluster includes: The recalled products of the target object are determined from the products included in the first product cluster.

6. The method according to claim 5, characterized in that The method further comprises: Perform commodity type recognition processing on the text information to obtain commodity type information corresponding to the index image; Comparing the commodity type information corresponding to the index image with the commodity type information of each first commodity cluster, and determining a second commodity cluster whose commodity type information in the first commodity cluster matches the commodity type information corresponding to the index image; The step of determining the recalled product of the target object from the products included in the first product cluster includes: The recalled products of the target object are determined from the products included in the second product cluster.

7. The method according to claim 1, characterized in that The step of acquiring the index image associated with the target object comprises: Receiving a product recall request sent by a client corresponding to the target object; wherein the product recall request carries search information; Searching the image database for an index image that matches the search information.

8. The method according to claim 1, characterized in that The step of acquiring the index image associated with the target object comprises: An index image of the interaction performed by the client corresponding to the target object within a historical time period is obtained.

9. A product recall device, characterized in that: The device comprises: An acquisition unit, configured to acquire a plurality of commodity clusters obtained by clustering commodities in a commodity database, and to acquire a clustering feature of each commodity cluster in the plurality of commodity clusters; The acquisition unit is further configured to acquire an index image associated with the target object, and determine at least one commodity cluster matching the index image from the plurality of commodity clusters according to image features of the index image and cluster features of each commodity cluster; A determination unit is used to determine the recalled commodity of the target object from the commodities included in the at least one commodity cluster.

10. A computer device, characterized in that: The computer device comprises a processor, a storage device and a communication interface, wherein the processor, the storage device and the communication interface are connected to each other, wherein: The storage device is used to store a computer program, wherein the computer program includes program instructions; The processor is used to call the program instructions to execute the product recall method according to any one of claims 1 to 8.

11. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, wherein the computer program includes program instructions. When the program instructions are executed by a processor, they are used to execute the product recall method according to any one of claims 1 to 8.

12. A computer program product, characterized in that The computer program product comprises a computer program, and the computer program is suitable for being loaded by a processor and executing the product recall method according to any one of claims 1 to 8.

Images