Article matching method and device

Abstract

The embodiment of the invention discloses an article matching method and device. A specific embodiment of the method comprises the following steps: acquiring a first article set and a second article set, and respectively selecting articles from the first article set and the second article set to form a to-be-matched article pair; for an article pair in the article pair set, images of two articles in the article pair are obtained respectively, and the positions of the two articles in the corresponding images are determined respectively; determining the similarity of the images of the two articles according to the positions; screening out a first article pair subset from the article pair set according to the similarity of the images; for an article pair in the first article pair subset, respectively obtaining description texts of two articles in the article pair; determining the similarity of the description texts of the two articles; and screening out a second article pair subset from the first article pair subset according to the similarity of the description text. According to the embodiment, the precision of an article matching result can be improved.

Description

Technical Field

[0001] Embodiments of the present disclosure relate to the field of computer technology, and in particular to an item matching method and device. Background Art

[0002] For many e-commerce and logistics companies or platforms, item matching is often one of the necessary basic tasks, which has a great impact on item assignment, introduction and identification of new items, etc. The essence of item matching is generally to establish a mapping relationship between items in two item sets. That is, to determine whether the items in each item pair in the Cartesian product of the two item sets match. The difficulty of item matching is related to various factors such as matching granularity (for example, matching at SKU granularity is usually more difficult than matching at SPU granularity), item set size, sample concentration, etc.

[0003] Existing item matching methods generally include: item matching based on the item's parameter attributes, item matching based on the item's parameter attributes and text information, and item matching based on the item's multimodal information (parameter attributes, text information, images, videos, etc.). Among them, the recall rates of item matching based on the item's parameter attributes and item matching based on the item's parameter attributes and text information are generally low, and some items cannot be matched at the SKU granularity only through parameter attributes and text information. Item matching methods based on the item's multimodal information generally vectorize the multimodal information and match items by calculating the similarity of the vectors. However, this matching method still has room for improvement in matching accuracy when there is a lot of interference information and item details in the image. Summary of the invention

[0004] The embodiments of the present disclosure provide an item matching method and device.

[0005] In a first aspect, an embodiment of the present disclosure provides an item matching method, the method comprising: obtaining a first item set and a second item set, and selecting items from the first item set and the second item set to form item pairs to be matched, so as to obtain a set of item pairs to be matched; for item pairs in the item pair set, respectively obtaining images of the two items in the item pair, and determining the positions of the two items in the corresponding images; determining the similarity of the images of the two items according to the positions; filtering out a first item pair subset from the item pair set according to the similarity of the images, wherein the two items in the item pairs in the first item pair subset belong to the same standardized product unit SPU; for item pairs in the first item pair subset, respectively obtaining description texts of the two items in the item pair; determining the similarity of the description texts of the two items; filtering out a second item pair subset from the first item pair subset according to the similarity of the description texts, wherein the two items in the item pairs in the second item pair subset belong to the same minimum stock keeping unit SKU.

[0006] In a second aspect, an embodiment of the present disclosure provides an item matching device, which includes: an acquisition unit, configured to acquire a first item set and a second item set, and select items from the first item set and the second item set to form item pairs to be matched, so as to obtain a set of item pairs to be matched; a first matching unit, configured to acquire images of two items in the item pair for each item pair in the item pair set, and determine the positions of the two items in the corresponding images; determine the similarity of the images of the two items according to the positions; and filter out a first item pair subset from the item pair set according to the similarity of the images, wherein the two items in the item pairs in the first item pair subset belong to the same standardized product unit SPU; a second matching unit, configured to acquire description texts of the two items in the item pair for each item pair in the first item pair subset; determine the similarity of the description texts of the two items; and filter out a second item pair subset from the first item pair subset according to the similarity of the description texts, wherein the two items in the item pairs in the second item pair subset belong to the same minimum stock keeping unit SKU.

[0007] In a third aspect, an embodiment of the present disclosure provides an electronic device, comprising: one or more processors; a storage device for storing one or more programs; when the one or more programs are executed by one or more processors, the one or more processors implement the method described in any implementation manner in the first aspect.

[0008] In a fourth aspect, an embodiment of the present disclosure provides a computer-readable medium having a computer program stored thereon, which, when executed by a processor, implements the method described in any implementation manner in the first aspect.

[0009] The item matching method and device provided by the embodiments of the present disclosure first determine the similarity of the images of the two items in the item pair to be matched by using the positions of the two items in the corresponding images, thereby avoiding the influence of interference information such as other image content other than the main item in the image on the item matching, and improving the accuracy of the first item pair subset at SPU granularity screened out based on the image similarity. Then, on this basis, the description text of the item is used to further screen out the second item pair subset at SKU granularity from the first item pair subset at SPU granularity, thereby achieving item matching at SKU granularity. BRIEF DESCRIPTION OF THE DRAWINGS

[0010] Other features, objects and advantages of the present disclosure will become more apparent from the detailed description of non-limiting embodiments made with reference to the following drawings:

[0011] Figure 1is an exemplary system architecture diagram in which an embodiment of the present disclosure may be applied;

[0012] Figure 2 is a flow chart of an embodiment of an item matching method according to the present disclosure;

[0013] Figure 3 is a flow chart of another embodiment of the item matching method according to the present disclosure;

[0014] Figure 4 is a schematic diagram of an application scenario of the object matching method according to an embodiment of the present disclosure;

[0015] Figure 5 is a schematic structural diagram of an embodiment of an object matching device according to the present disclosure;

[0016] Figure 6 It is a schematic diagram of the structure of an electronic device suitable for implementing the embodiments of the present disclosure. DETAILED DESCRIPTION

[0017] The present disclosure is further described in detail below in conjunction with the accompanying drawings and embodiments. It is understood that the specific embodiments described herein are only used to explain the relevant invention, rather than to limit the invention. It is also necessary to explain that, for ease of description, only the relevant parts are shown in the accompanying drawings.

[0018] It should be noted that, in the absence of conflict, the embodiments and features in the embodiments of the present disclosure may be combined with each other. The present disclosure will be described in detail below with reference to the accompanying drawings and in combination with the embodiments.

[0019] Figure 1 An exemplary architecture 100 is shown to which an embodiment of the item matching method or item matching apparatus of the present disclosure can be applied.

[0020] like Figure 1 As shown, the system architecture 100 may include terminal devices 101, 102, 103, a network 104 and a server 105. The network 104 is used to provide a medium for communication links between the terminal devices 101, 102, 103 and the server 105. The network 104 may include various connection types, such as wired, wireless communication links or optical fiber cables, etc.

[0021] The terminal devices 101, 102, 103 interact with the server 105 via the network 104 to receive or send messages, etc. Various client applications may be installed on the terminal devices 101, 102, 103, such as browser applications, shopping applications, search applications, data processing applications, etc.

[0022] Terminal devices 101, 102, 103 can be hardware or software. When terminal devices 101, 102, 103 are hardware, they can be various electronic devices, including but not limited to smart phones, tablet computers, e-book readers, laptop portable computers and desktop computers, etc. When terminal devices 101, 102, 103 are software, they can be installed in the electronic devices listed above. It can be implemented as multiple software or software modules (for example, multiple software or software modules for providing distributed services), or it can be implemented as a single software or software module. No specific limitation is made here.

[0023] The server 105 may be a server that provides various services, such as a backend server that provides data support for client applications installed on the terminal devices 101, 102, and 103. The server 105 may obtain the first set of items and the second set of items from the terminal devices 101, 102, and 103, and filter out a set of item pairs to be matched therefrom, and then perform item matching to obtain an item matching result at the SKU granularity.

[0024] It should be noted that the first set of items and the second set of items may also be directly stored locally on the server 105, and the server 105 may directly extract the first set of items and the second set of items stored locally and process them. In this case, the terminal devices 101, 102, 103 and the network 104 may not exist.

[0025] It should be noted that the item matching method provided in the embodiments of the present disclosure is generally executed by the server 105 , and accordingly, the item matching device is generally disposed in the server 105 .

[0026] It should also be noted that data processing applications may also be installed in the terminal devices 101, 102, and 103, and the terminal devices 101, 102, and 103 may also match the first set of items with the second set of items based on the data processing applications. In this case, the item matching method may also be executed by the terminal devices 101, 102, and 103, and accordingly, the item matching device may also be set in the terminal devices 101, 102, and 103. In this case, the exemplary system architecture 100 may not have the server 105 and the network 104.

[0027] It should be noted that the server 105 can be hardware or software. When the server 105 is hardware, it can be implemented as a distributed server cluster consisting of multiple servers, or it can be implemented as a single server. When the server 105 is software, it can be implemented as multiple software or software modules (for example, multiple software or software modules for providing distributed services), or it can be implemented as a single software or software module. No specific limitation is made here.

[0028] It should be understood that Figure 1 The number of terminal devices, networks and servers in the embodiment is only for illustration. Any number of terminal devices, networks and servers may be provided according to implementation requirements.

[0029] Continue to refer Figure 2 , which shows a process 200 of an embodiment of an item matching method according to the present disclosure. The item matching method comprises the following steps:

[0030] Step 201: Obtain a first item set and a second item set, and select items from the first item set and the second item set to form item pairs to be matched, thereby obtaining a set of item pairs to be matched.

[0031] In this embodiment, the item set may be composed of a plurality of items. The items in the item set may be items of any type. For example, for an e-commerce platform, the items may be various commodities. The first item set and the second item set may be item sets from different sources.

[0032] Each pair of items to be matched may include two items, one of which belongs to the first set of items, and the other belongs to the second set of items. Specifically, various methods may be flexibly used to select items from the first set of items and the second set of items to form pairs of items to be matched, so as to obtain a set of pairs of items to be matched.

[0033] For example, the Cartesian product of the first item set and the second item set may be calculated, and then a preset preprocessing operation may be used to remove obviously mismatched item pairs, and the remaining item pairs may be used as item pairs to be matched, so as to form a set of item pairs to be matched.

[0034] Step 202: for an item pair in the item pair set, respectively obtain images of two items in the item pair, determine positions of the two items in the corresponding images; determine similarity of the images of the two items based on the positions; and filter out a first item pair subset from the item pair set based on the image similarity.

[0035] In this embodiment, the image of an object may refer to various images showing an object. For example, taking a commodity as an example, the image of the commodity may include a commodity display picture, etc. For each image of an object, the position of the object in its corresponding image may be determined using various existing object detection and recognition methods. For example, the position of an object in its corresponding image may be determined based on the open source YOLO-V4 model.

[0036] Then, based on the position of each object in its corresponding image, various existing similarity calculation methods can be used to determine the similarity between the images of the two objects in each object pair. Specifically, based on the position of each object in its corresponding image, the similarity between the image regions where the two objects are located can be calculated as the similarity between the two images.

[0037] After obtaining the similarity corresponding to each item pair, the first item pair subset can be screened out from the item pair set according to the similarity. The two items in each item pair in the first item pair subset belong to the same SPU, that is, the item matching result of SPU granularity can be obtained. SPU (Standard Product Unit) is the smallest unit for commodity information aggregation. For example, items with the same brand and model can be regarded as one SPU.

[0038] Specifically, item pairs whose similarity is not less than a preset image similarity threshold can be selected from the above item pair set to form a first item pair subset, wherein the image similarity threshold can be flexibly set by technicians according to actual application scenarios.

[0039] Step 203: for the item pairs in the first item pair subset, respectively obtain the description texts of the two items in the item pair; determine the similarity of the description texts of the two items; and filter out the second item pair subset from the first item pair subset according to the similarity of the description texts.

[0040] In this embodiment, for each item in each item pair in the first item pair subset, the description text of the item may be obtained first. The description text of the item may be various texts related to the item. For example, the description text includes but is not limited to: the title of the item, attribute information, etc. Taking a commodity as an example, the description text may be a display title and marketing information of the commodity, etc.

[0041] Then, for each item pair, various similarity calculation methods can be used to determine the similarity between the description texts of the two items in the item pair. After obtaining the similarity corresponding to each item pair, the second item pair subset can be screened out from the first item pair subset according to the similarity. Among them, the two items in the item pair in the second item pair subset belong to the same SKU, that is, the item matching result of SKU granularity is obtained. SKU (Stock Keeping Unit) is the basic unit for measuring inventory in and out.

[0042] Specifically, item pairs whose similarity is not less than a preset text similarity threshold can be selected from the first item pair subset to form a second item pair subset. The text similarity threshold can be flexibly set by technicians according to actual application scenarios.

[0043] The method provided by the above-mentioned embodiment of the present disclosure can more accurately estimate the similarity of the main parts of the objects in the images by first detecting the position of each object in its corresponding image and then determining the similarity between the images based on the position, thereby avoiding the influence of other content in the object image (such as marketing information, promotional information, etc. in the product image) on the similarity calculation, thereby helping to improve the accuracy of subsequent object matching results based on the pixel degree.

[0044] In some optional implementations of this embodiment, the following steps may be performed to select items from the first item set and the second item set to form item pairs to be matched, thereby obtaining a set of item pairs to be matched:

[0045] Step 1: Obtain target attribute values ​​of each item in the first item set and the second item set respectively.

[0046] In this step, the target attribute value of the item can be the attribute value of various attributes of the item, which can be set by the technician according to the actual application requirements. For example, the target attribute corresponding to the target attribute value of the item can be a label set by the technician for the item in advance, the origin of the item, etc.

[0047] Optionally, the target attribute includes but is not limited to at least one of the following: category, brand, title, and value. The category can be used to indicate the category to which the item belongs. The value can be used to indicate various values ​​of the item, such as the price of the item.

[0048] Step 2: According to the target attribute value, select items from the first item set and the second item set to form item pairs to be matched.

[0049] In this step, for each item pair, the target attribute values ​​of the two items in the item pair can be compared to determine whether the item pair is an item pair to be matched, so as to filter out some item pairs that do not meet the matching requirements. Different comparison methods and determination conditions can be used for different target attribute values.

[0050] For example, if the categories of two items are related (such as the same or belong to parent-child categories, etc.), the item pair formed by the two items can be determined to be an item pair to be matched. Correspondingly, if the categories of two items are unrelated, it can be determined that the item pair formed by the two items does not belong to an item pair to be matched.

[0051] If the brand similarity of the two items is high (such as not less than the first threshold), the item pair consisting of the two items can be determined to be the item pair to be matched. Correspondingly, if the brand similarity of the two items is low (such as less than the first threshold), it can be determined that the item pair consisting of the two items does not belong to the item pair to be matched.

[0052] If the similarity between the titles of the two items is high (such as not less than the second threshold), it can be determined that the item pair formed by the two items is an item pair to be matched. Correspondingly, if the similarity between the titles of the two items is low (such as less than the second threshold), it can be determined that the item pair formed by the two items is not an item pair to be matched.

[0053] If the difference in value between the two items is small (e.g., not greater than the third threshold), then the item pair consisting of the two items can be determined to be the item pair to be matched. Correspondingly, if the difference in value between the two items is large (e.g., greater than the third threshold), then the item pair consisting of the two items can be determined not to be the item pair to be matched.

[0054] Among them, the above-mentioned first threshold, second threshold and third threshold, etc. can be pre-set by technical personnel according to actual application scenarios (such as historical experience or data statistical results, etc.).

[0055] Optionally, taking the case where the target attributes include the category, brand, title, and value of an item as an example, the item pairs to be matched can be determined by the following steps:

[0056] Step 1: According to the categories of the items, items are selected from the first item set and the second item set to form category item pairs, thereby obtaining a category item pair set.

[0057] In this step, according to the correlation between the categories to which the items in the two item sets belong, item pairs can be selected as category item pairs to obtain a category item pair set. The categories to which the two items in each category item pair belong are related. Specifically, the condition of whether the categories are related can be preset by a technician. For example, the correlation between various categories can be preset.

[0058] Step 2: According to the brand of the item, filter out the brand item pair set from the category item pair set.

[0059] In this step, according to the screening results of the above step 1, the brands of the items can be used to further screen item pairs with high brand similarity between the two items in the category item pair set as brand item pairs to obtain a brand item pair set. For example, the brand similarity between the two items in each brand item pair is not less than a first threshold.

[0060] Step 3: According to the title of the item, filter out the title item pair set from the brand item pair set.

[0061] In this step, according to the screening results of the above step 2, the item titles can be used to further screen item pairs with two items having high title similarity from the brand item pair set as title item pairs to obtain a title item pair set. For example, the similarity between the titles of the two items in each title item pair is not less than the second threshold.

[0062] Step 4: According to the value of the items, select the item pair set to be matched from the title item pair set.

[0063] In this step, according to the screening result of the above step 3, the value of the items can be used to further screen item pairs with a low value difference between the two items in the title item pair set as item pairs to be matched, thereby obtaining a set of item pairs to be matched. For example, the value difference between the two items in each set of item pairs to be matched is not greater than the third threshold.

[0064] It should be noted that the attributes used in the above steps 1 to 4 are only exemplary and can be flexibly replaced.

[0065] After obtaining the first set of items and the second set of items, some unmatched item pairs can be filtered out by first using the item's category, brand, title, value and other attribute values, thereby reducing the number of item pairs to be matched later, thereby improving the item matching efficiency. In addition, when using multiple attribute values ​​of items for screening, further screening can be performed on the results of the previous screening in a certain order, which can further save some unnecessary calculation processes, thereby further improving the item matching efficiency.

[0066] In some optional implementations of this embodiment, the similarity between the images of the two objects may be determined according to the positions of the two objects in the object pair in the corresponding images through the following steps:

[0067] Step 1: According to the positions of two objects in the object pair in the corresponding images, the image regions where the two objects are located are determined.

[0068] In this step, various existing cutout methods may be used to obtain the image region where the object is located according to the position of the object in its corresponding object.

[0069] Step 2: Using a pre-trained image matching model, determine the similarity of the image regions where the two objects are located as the similarity of the images of the two objects.

[0070] In this step, the image matching model can be used to determine the similarity between two images. The structure of the image matching model can be preset by the technician. For example, the image matching model can adopt a self-supervised training method and modify the traditional classification Softmax to calculate the loss by adding a margin variable so that the corresponding distances of images of the same type are closer and the corresponding distances of images of different types are farther, thereby achieving learning of image details.

[0071] Similarly, a pre-trained text matching model can be used to determine the similarity between the description texts of two items. Among them, the text matching model can be used to determine the similarity between two texts. The structure of the text matching model can be pre-set by a technician. For example, the text matching model is built based on a traditional binary classification model. At the same time, the traditional classification Softmax can be modified by adding a margin variable to calculate the loss to increase the convergence difficulty of the text matching model, thereby achieving full learning of the positive sample pairs.

[0072] Among them, the margin loss function can be expressed as follows:

[0073] Loss=W]·|X|·cos0

[0074] Among them, "Loss" represents the loss value of the margin loss function. "W" represents the parameter weight of the image matching model or text matching model. "X" represents the feature of the input image or text. "θ" represents the angle between the feature representations of the two input images or texts.

[0075] By adding the margin variable to modify the loss function, we can learn the pixel details in the image, thereby improving the accuracy of image similarity calculation, which in turn helps to improve the subsequent item matching accuracy. At the same time, we can increase the convergence difficulty of the text matching model, thereby achieving full learning of positive sample pairs, improving the accuracy of text similarity calculation, and thus helping to improve the subsequent item matching accuracy.

[0076] Further references Figure 3 , which shows a process 300 of another embodiment of an item matching method. The process 300 of the item matching method includes the following steps:

[0077] Step 301: Obtain a first item set and a second item set, and select items from the first item set and the second item set to form item pairs to be matched, thereby obtaining a set of item pairs to be matched.

[0078] Step 302: for an item pair in the item pair set, respectively obtain images of two items in the item pair, determine positions of the two items in the corresponding images; determine similarity of the images of the two items based on the positions; and filter out a first item pair subset from the item pair set based on the image similarity.

[0079] Step 303: for the item pairs in the first item pair subset, respectively obtain the description texts of the two items in the item pair; determine the similarity of the description texts of the two items; and filter out the second item pair subset from the first item pair subset according to the similarity of the description texts.

[0080] Step 304: Perform target processing on the second item pair subset to obtain a processing result.

[0081] In this embodiment, the target processing may include at least one of the following: item pair correction and item pair aggregation. Item pair correction may be used to further correct each item pair to determine whether the item pair belongs to the same SKU. Item pair aggregation may be used to aggregate the SKUs corresponding to each item pair, so that item pairs with the same SKU can be aggregated to obtain an item subset with a corresponding SKU granularity.

[0082] For example, a preset object knowledge base may be used to perform object pair calibration, wherein the object knowledge base may include various attribute information (such as color, size, model, etc.) of the object that is not used in the above-mentioned object matching process.

[0083] For another example, item "A" and item "B" in one item pair have the same SKU. Item "A" and item "C" in another item pair have the same SKU. In this case, the two item pairs can be aggregated, and items "A", "B" and "C" can be aggregated together for subsequent use and storage.

[0084] Continue to see Figure 4 , Figure 4 This is an illustrative application scenario of the object matching method according to this embodiment. Figure 4 In the application scenario, you can first obtain the product pools "A" and "B". Then, through steps "M1"-"M4", matching pairs are constructed to screen out the product pairs to be matched as the candidate matching pair set. Specifically, step "M1" is category matching, and some product pairs with mismatched categories can be filtered out based on categories. Step "M2" is similar brand filtering, and some product pairs with mismatched brands can be filtered out based on brands. Step "M3" is similar product filtering, and some product pairs with mismatched titles can be filtered out based on titles. Step "M4" is matching pair price difference filtering, and some product pairs with too large price differences can be filtered out based on product prices.

[0085] Then, for the matching pairs in the candidate matching pair set, the product matching at the SPU granularity can be achieved through steps "M5" and "M6" based on the pictures of each product in the matching pair. Specifically, step "M5" is product detection in the picture, which can detect the main position of the product in the product picture. Step "M6" is product picture metric learning, which can calculate the similarity between the pictures of the two products in the matching pair.

[0086] Then, based on the title of each product in the matching pair, SKU-granular product matching can be achieved through steps "M7" and "M8". Specifically, step "M7" is product title metric learning, which can calculate the similarity between the titles of two products in the matching pair. Step "M8" is SKU matching post-processing, and the SKU-granular product pairs obtained in step "M7" can be further sorted through steps such as correction and aggregation to obtain the final SKU-granular product pairs for output.

[0087] After obtaining the item matching results at the SKU granularity, you can further use various other attributes of the item to make corrections, and further organize the item matching results through operations such as aggregation to obtain the final item matching results for subsequent use.

[0088] Further references Figure 5 As an implementation of the methods shown in the above figures, the present disclosure provides an embodiment of a product matching device. Figure 2 Corresponding to the method embodiment shown, the device can be specifically applied to various electronic devices.

[0089] like Figure 5As shown, the object matching device 500 provided in this embodiment includes an acquisition unit 501 , a first matching unit 502 , and a second matching unit 503 . The acquisition unit 501 is configured to acquire a first item set and a second item set, and select items from the first item set and the second item set to form item pairs to be matched, so as to obtain the item pair set to be matched; the first matching unit 502 is configured to acquire images of two items in the item pair for each item pair in the item pair set, and determine the positions of the two items in the corresponding images; determine the similarity of the images of the two items according to the positions; and filter out a first item pair subset from the item pair set according to the similarity of the images, wherein the two items in the item pairs in the first item pair subset belong to the same standardized product unit SPU; the second matching unit 503 is configured to acquire description texts of the two items in the item pair for each item pair in the first item pair subset; determine the similarity of the description texts of the two items; and filter out a second item pair subset from the first item pair subset according to the similarity of the description texts, wherein the two items in the item pairs in the second item pair subset belong to the same minimum stock keeping unit SKU.

[0090] In this embodiment, in the object matching device 500, the specific processing of the acquisition unit 501, the first matching unit 502 and the second matching unit 503 and the technical effects thereof can be referred to in Figure 2 The relevant descriptions of step 201, step 202 and step 203 in the corresponding embodiment are not repeated here.

[0091] In some optional implementations of this embodiment, the acquisition unit 501 is further configured to: respectively acquire target attribute values ​​of each item in the first item set and the second item set; and select items from the first item set and the second item set to form item pairs to be matched according to the target attribute values.

[0092] In some optional implementations of this embodiment, the target attribute corresponding to the above target attribute value includes at least one of the following: category, brand, title, and value.

[0093] In some optional implementations of the present embodiment, the acquisition unit 501 is further configured to: select items from the first item set and the second item set to form category item pairs according to the categories of the items, and obtain a category item pair set, wherein the categories to which the two items in the category item pairs belong are related; select a brand item pair set from the category item pair set according to the brands of the items, wherein the brand similarity of the two items in the brand item pairs in the brand item pair set is not less than a first threshold; select a title item pair set from the brand item pair set according to the titles of the items, wherein the title similarity of the two items in the title item pairs in the title item pair set is not less than a second threshold; select a to-be-matched item pair set from the title item pair set according to the values ​​of the items, wherein the value difference of the two items in the item pairs in the to-be-matched item pair set is not greater than a third threshold.

[0094] In some optional implementations of the present embodiment, the first matching unit 502 is further matched to: determine the image regions where the two objects are respectively located based on the above-mentioned positions; and use a pre-trained image matching model to determine the similarity of the image regions where the two objects are respectively located as the similarity of the images of the two objects, wherein the loss function of the image matching model is constructed based on the margin loss function.

[0095] In some optional implementations of this embodiment, the second matching unit 503 is further matched to: determine the similarity between the description texts of two objects using a pre-trained text matching model, wherein the loss function of the text matching model is constructed based on the margin loss function.

[0096] In some optional implementations of this embodiment, the second matching unit 503 is further matched to: perform target processing on the second item pair subset to obtain a processing result, wherein the target processing includes at least one of the following: item pair correction and item pair aggregation.

[0097] The device provided by the above-mentioned embodiment of the present disclosure obtains a first item set and a second item set through an acquisition unit, and selects items from the first item set and the second item set to form item pairs to be matched, thereby obtaining an item pair set to be matched; the first matching unit obtains images of two items in the item pair for the item pairs in the item pair set, and determines the positions of the two items in the corresponding images respectively; determines the similarity of the images of the two items according to the positions; and selects a first item pair subset from the item pair set according to the similarity of the images, wherein the two items in the item pairs in the first item pair subset belong to the same standardized product unit SPU; the second matching unit obtains description texts of the two items in the item pair for the item pairs in the first item pair subset; determines the similarity of the description texts of the two items; and selects a second item pair subset from the first item pair subset according to the similarity of the description texts, wherein the two items in the item pairs in the second item pair subset belong to the same minimum stock keeping unit SKU, thereby avoiding the influence of interference information such as other image content other than the main item in the image on item matching, thereby improving the accuracy of the item matching result obtained based on image matching.

[0098] Reference below Figure 6 , which shows an electronic device (eg, Figure 1 The terminal device in the embodiment of the present disclosure may include but is not limited to mobile terminals such as mobile phones, notebook computers, digital broadcast receivers, PDAs (personal digital assistants), PADs (tablet computers), PMPs (portable multimedia players), vehicle-mounted terminals (such as vehicle-mounted navigation terminals), etc., and fixed terminals such as digital TVs, desktop computers, etc. Figure 6 The server shown is only an example and should not bring any limitation to the functions and scope of use of the embodiments of the present disclosure.

[0099] like Figure 6 As shown, the electronic device 600 may include a processing device (e.g., a central processing unit, a graphics processing unit, etc.) 601, which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM) 602 or a program loaded from a storage device 608 into a random access memory (RAM) 603. In the RAM 603, various programs and data required for the operation of the electronic device 600 are also stored. The processing device 601, the ROM 602, and the RAM 603 are connected to each other via a bus 604. An input / output (I / O) interface 605 is also connected to the bus 604.

[0100] Typically, the following devices may be connected to the I / O interface 605: an input device 606 including, for example, a touch screen, a touch pad, a keyboard, a mouse, a camera, a microphone, an accelerometer, a gyroscope, etc.; an output device 607 including, for example, a liquid crystal display (LCD), a speaker, a vibrator, etc.; a storage device 608 including, for example, a magnetic tape, a hard disk, etc.; and a communication device 609. The communication device 609 may allow the electronic device 600 to communicate with other devices wirelessly or by wire to exchange data. Although Figure 6 The electronic device 600 is shown with various devices, but it should be understood that it is not required to implement or possess all the devices shown. More or fewer devices may be implemented or possessed instead. Figure 6 Each block shown in the figure may represent one device, or may represent multiple devices as required.

[0101] In particular, according to an embodiment of the present disclosure, the process described above with reference to the flowchart can be implemented as a computer software program. For example, an embodiment of the present disclosure includes a computer program product, which includes a computer program carried on a computer-readable medium, and the computer program contains program code for executing the method shown in the flowchart. In such an embodiment, the computer program can be downloaded and installed from a network through a communication device 609, or installed from a storage device 608, or installed from a ROM 602. When the computer program is executed by the processing device 601, the above-mentioned functions defined in the method of the embodiment of the present disclosure are executed.

[0102] It should be noted that the computer-readable medium described in the embodiments of the present disclosure may be a computer-readable signal medium or a computer-readable storage medium or any combination of the above two. The computer-readable storage medium may be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, device or device, or any combination of the above. More specific examples of computer-readable storage media may include, but are not limited to: an electrical connection with one or more wires, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above. In the embodiments of the present disclosure, the computer-readable storage medium may be any tangible medium containing or storing a program, which may be used by or in combination with an instruction execution system, device or device. In the embodiments of the present disclosure, the computer-readable signal medium may include a data signal propagated in a baseband or as part of a carrier wave, in which a computer-readable program code is carried. This propagated data signal may take a variety of forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination of the above. The computer readable signal medium may also be any computer readable medium other than a computer readable storage medium, which may send, propagate or transmit a program for use by or in conjunction with an instruction execution system, apparatus or device. The program code contained on the computer readable medium may be transmitted using any suitable medium, including but not limited to: wires, optical cables, RF (radio frequency), etc., or any suitable combination of the above.

[0103] The computer-readable medium may be included in the electronic device; or it may exist independently without being installed in the electronic device. The computer-readable medium carries one or more programs. When the one or more programs are executed by the electronic device, the electronic device: obtains a first set of items and a second set of items, and selects items from the first set of items and the second set of items to form pairs of items to be matched, thereby obtaining a set of item pairs to be matched; for item pairs in the set of item pairs, obtains images of the two items in the item pairs, and determines the positions of the two items in the corresponding images; determines the similarity of the images of the two items according to the positions; selects a first subset of item pairs from the set of item pairs according to the similarity of the images, wherein the two items in the item pairs in the first subset of item pairs belong to the same standardized product unit SPU; for item pairs in the first subset of item pairs, obtains description texts of the two items in the item pairs, determines the similarity of the description texts of the two items; selects a second subset of item pairs from the first subset of item pairs according to the similarity of the description texts, wherein the two items in the item pairs in the second subset of item pairs belong to the same minimum stock keeping unit SKU.

[0104] Computer program code for performing the operations of embodiments of the present disclosure may be written in one or more programming languages ​​or a combination thereof, including object-oriented programming languages, such as Java, Smalltalk, C++, and conventional procedural programming languages, such as "C" or similar programming languages. The program code may be executed entirely on a user's computer, partially on a user's computer, as a separate software package, partially on a user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving a remote computer, the remote computer may be connected to the user's computer via any type of network, including a local area network (LAN) or a wide area network (WAN), or may be connected to an external computer (e.g., via the Internet using an Internet service provider).

[0105] The flow chart and block diagram in the accompanying drawings illustrate the possible architecture, function and operation of the system, method and computer program product according to various embodiments of the present disclosure. In this regard, each square box in the flow chart or block diagram can represent a module, a program segment or a part of a code, and the module, the program segment or a part of the code contains one or more executable instructions for realizing the specified logical function. It should also be noted that in some implementations as replacements, the functions marked in the square box can also occur in a sequence different from that marked in the accompanying drawings. For example, two square boxes represented in succession can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each square box in the block diagram and / or flow chart, and the combination of the square boxes in the block diagram and / or flow chart can be implemented with a dedicated hardware-based system that performs the specified function or operation, or can be implemented with a combination of dedicated hardware and computer instructions.

[0106] The units involved in the embodiments described in the present disclosure may be implemented by software or hardware. The units described may also be set in a processor, for example, may be described as: a processor includes an acquisition unit, a first matching unit, and a second matching unit. The names of these units do not constitute limitations on the units themselves in certain cases. For example, the acquisition unit may also be described as "a unit for acquiring a first set of items and a second set of items, and selecting items from the first set of items and the second set of items to form pairs of items to be matched, thereby obtaining a set of pairs of items to be matched."

[0107] The above description is only a preferred embodiment of the present disclosure and an explanation of the technical principles used. Those skilled in the art should understand that the scope of the invention involved in the embodiments of the present disclosure is not limited to the technical solutions formed by a specific combination of the above technical features, but should also cover other technical solutions formed by any combination of the above technical features or their equivalent features without departing from the concept. For example, the above features are replaced with (but not limited to) technical features with similar functions disclosed in the embodiments of the present disclosure to form a technical solution.

Claims

1. An item matching method, comprising: Acquire a first item set and a second item set, and select items from the first item set and the second item set to form item pairs to be matched, to obtain a set of item pairs to be matched; For each item pair in the item pair set, images of two items in the item pair are obtained respectively, and positions of the two items in the corresponding images are determined; similarities of the images of the two items are determined according to the positions; and a first item pair subset is selected from the item pair set according to the image similarities, wherein the two items in the item pair in the first item pair subset belong to the same standardized product unit SPU; For the item pairs in the first item pair subset, respectively obtain description texts of two items in the item pair; determine the similarity of the description texts of the two items; and filter out a second item pair subset from the first item pair subset based on the similarity of the description texts, wherein the two items in the item pairs in the second item pair subset belong to the same minimum stock keeping unit SKU.

2. The method according to claim 1, wherein: The selecting items from the first item set and the second item set to form item pairs to be matched includes: Obtain target attribute values ​​of each item in the first item set and the second item set respectively; According to the target attribute value, items are selected from the first item set and the second item set to form item pairs to be matched.

3. The method according to claim 2, wherein: The target attribute corresponding to the target attribute value includes at least one of the following: category, brand, title, and value.

4. The method according to claim 3, wherein: The selecting items from the first item set and the second item set to form item pairs to be matched according to the target attribute value includes: According to the categories of the items, items are selected from the first item set and the second item set to form category item pairs, thereby obtaining a category item pair set, wherein the two items in the category item pairs belong to related categories; According to the brands of the items, a brand item pair set is selected from the category item pair set, wherein the brand similarity of two items in the brand item pair in the brand item pair set is not less than a first threshold; Filtering a title item pair set from the brand item pair set according to the item titles, wherein the similarity between the titles of two items in the title item pairs in the title item pair set is not less than a second threshold; The set of item pairs to be matched is screened out from the set of title item pairs according to the value of the items, wherein the difference in value between two items in the item pairs in the set of item pairs to be matched is not greater than a third threshold.

5. The method according to claim 1, wherein: Determining the similarity of the images of the two objects according to the positions includes: According to the positions, determining the image regions where the two objects are respectively located; Using a pre-trained image matching model, similarities of the image regions where the two objects are located are determined as similarities of the images of the two objects, wherein a loss function of the image matching model is constructed based on a margin loss function.

6. The method according to claim 1, wherein: Determining the similarity between the description texts of the two items includes: The similarity between the description texts of the two items is determined by using a pre-trained text matching model, wherein a loss function of the text matching model is constructed based on a margin loss function.

7. The method according to any one of claims 1 to 6, wherein: The method further comprises: Perform target processing on the second subset of item pairs to obtain a processing result, wherein the target processing includes at least one of the following: item pair correction and item pair aggregation.

8. An object matching device, wherein: The device comprises: an acquisition unit, configured to acquire a first set of items and a second set of items, and select items from the first set of items and the second set of items to form pairs of items to be matched, to obtain a set of pairs of items to be matched; The first matching unit is configured to obtain images of two items in the item pair in the item pair set, respectively, and determine positions of the two items in the corresponding images; determine similarities of the images of the two items according to the positions; and select a first item pair subset from the item pair set according to the image similarities, wherein the two items in the item pair in the first item pair subset belong to the same standardized product unit SPU; The second matching unit is configured to obtain, for each item pair in the first item pair subset, description texts of two items in the item pair; determine the similarity of the description texts of the two items; and filter out a second item pair subset from the first item pair subset based on the similarity of the description texts, wherein the two items in the item pairs in the second item pair subset belong to the same minimum stock keeping unit SKU.

9. An electronic device, comprising: one or more processors; a storage device having one or more programs stored thereon; When the one or more programs are executed by the one or more processors, the one or more processors implement the method according to any one of claims 1 to 7.

10. A computer readable medium having a computer program stored thereon, wherein: When the program is executed by a processor, the method according to any one of claims 1 to 7 is implemented.

Images