Method, device, server and storage medium for processing item information

Abstract

The present application provides a method, device, server and storage medium for processing item information, which belongs to the field of Internet technology. The method includes: obtaining the attribute information and item image of the target item; determining the first semantic feature corresponding to the attribute information and the second semantic feature corresponding to the item image; fusing the first semantic feature and the second semantic feature to obtain the target semantic feature of the target item; and determining the standard item information of the target item based on the target semantic feature. Since the first semantic feature is the semantic feature corresponding to the attribute information of the item, and the second semantic feature is the semantic feature corresponding to the item image, the target semantic feature obtained by fusing the first semantic feature and the second semantic feature fuses multiple semantic features of different modalities. Therefore, determining the standard item information through the target semantic feature can improve the accuracy of the determined standard item information.

Description

Technical Field

[0001] The present application relates to the field of Internet technology, and in particular to a method, device, server, and storage medium for processing item information. Background Art

[0002] With the development of internet technology, more and more users are ordering takeout through online platforms. Users can enter the name of a dish, and the server will recommend multiple dishes that match the dish name based on the dish name and the names of dishes from multiple merchants. To improve search accuracy, the server needs to determine the standard dish name corresponding to the merchant's dish name, and then make dish recommendations based on the standard dish name, thereby improving accuracy.

[0003] In the related art, the server determines the semantic similarity between a dish name and a plurality of standard dish names, and based on the semantic similarity, determines the standard dish name corresponding to the dish name.

[0004] However, due to the high degree of personalization and incomplete information of some merchants' dish names, the accuracy of determining the standard dish name corresponding to the dish name based on the semantic similarity between the dish name and the standard dish name is low. Summary of the Invention

[0005] The embodiments of the present application provide a method, device, server, and storage medium for processing item information, which can improve the accuracy of information processing. The technical solution is as follows:

[0006] According to one aspect of an embodiment of the present application, a method for processing item information is provided, the method comprising:

[0007] Obtaining attribute information and item images of the target item;

[0008] Determining a first semantic feature corresponding to the attribute information and a second semantic feature corresponding to the object image;

[0009] fusing the first semantic feature and the second semantic feature to obtain a target semantic feature of the target object;

[0010] Based on the target semantic feature, standard item information of the target item is determined.

[0011] In one possible implementation, the attribute information includes at least two of an item name, a merchant name, and tag information; and determining the first semantic feature corresponding to the attribute information includes:

[0012] Determine at least two semantic vectors among a first semantic vector, a second semantic vector, and a third semantic vector, wherein the first semantic vector is a semantic vector corresponding to the item name, the second semantic vector is a semantic vector corresponding to the merchant name, and the third semantic vector is a semantic vector corresponding to the tag information;

[0013] The first semantic feature is determined based on the at least two semantic vectors.

[0014] In another possible implementation, determining the first semantic feature based on the at least two semantic vectors includes:

[0015] splicing the at least two semantic vectors to obtain a target semantic vector, or adding category information to each semantic vector and splicing the at least two semantic vectors after adding the category information to obtain a target semantic vector;

[0016] Determine the first semantic feature corresponding to the target semantic vector.

[0017] In another possible implementation, determining the first semantic feature corresponding to the target semantic vector includes:

[0018] The target semantic vector is input into a semantic vector model, and the first semantic feature is output. The semantic vector model is used to convert the semantic vector into a semantic feature.

[0019] In another possible implementation, the process of determining the second semantic feature corresponding to the object image includes:

[0020] determining at least one region of interest in the image of the object;

[0021] For each region of interest, determining a second semantic vector corresponding to the region of interest to obtain at least one second semantic vector;

[0022] The second semantic feature is determined based on the at least one second semantic vector.

[0023] In another possible implementation, fusing the first semantic feature and the second semantic feature to obtain a target semantic feature of the target object includes:

[0024] Converting the first semantic feature and the second semantic feature into semantic features of the same dimension to obtain a first standard semantic feature and a second standard semantic feature;

[0025] Determining a first attention feature corresponding to the first standard semantic feature and a second attention feature corresponding to the second standard semantic feature;

[0026] The first attention feature and the second attention feature are fused to obtain the target semantic feature.

[0027] In another possible implementation, determining a first attention feature corresponding to the first standard semantic feature and a second attention feature corresponding to the second standard semantic feature includes:

[0028] Determine the first query vector, first key vector and first value vector corresponding to the first standard semantic feature, and the second query vector, second key vector and second value vector corresponding to the second standard semantic feature; determine the first attention weight and the second attention weight based on the first query vector, the first key vector, the second query vector and the second key vector; based on the first value vector and the first attention weight, perform weighted processing on the first standard semantic feature to obtain the first attention feature, and based on the second value vector and the second attention weight, perform weighted processing on the second standard semantic feature to obtain the second attention feature.

[0029] In another possible implementation, determining the first attention weight and the second attention weight includes:

[0030] determining the first attention weight based on a first similarity, a second similarity, and the first value vector, wherein the first similarity is the similarity between the first query vector and the first key vector, and the second similarity is the similarity between the second query vector and the first key vector;

[0031] The second attention weight is determined based on a third similarity, a fourth similarity, and the second value vector, wherein the third similarity is the similarity between the first query vector and the second key vector, and the fourth similarity is the similarity between the second query vector and the second key vector.

[0032] In another possible implementation, the first attention weight includes a first weight and a second weight;

[0033] The determining the first attention weight based on the first similarity, the second similarity and the first value vector includes:

[0034] Normalizing the first similarity to obtain the first weight, and normalizing the second similarity to obtain the second weight;

[0035] The first attention weight is determined based on a product between the first weight and the first value vector and a product between the second weight and the first value vector.

[0036] In another possible implementation, the second attention weight includes a third weight and a fourth weight;

[0037] The determining the second attention weight based on the third similarity, the fourth similarity, and the first value vector includes:

[0038] Normalizing the third similarity to obtain the third weight, and normalizing the fourth similarity to obtain the fourth weight;

[0039] The first attention weight is determined based on a product between the third weight and the second value vector and a product between the fourth weight and the second value vector.

[0040] In another possible implementation, fusing the first attention feature and the second attention feature to obtain the target semantic feature includes:

[0041] fusing the first attention feature and the second attention feature to obtain an attention feature;

[0042] The attention features are pooled to obtain the target semantic features.

[0043] In another possible implementation, determining the standard item information of the target item based on the target semantic feature includes:

[0044] Determining similarities between the target object and the plurality of standard object information based on the target semantic features and the semantic features of the plurality of standard object information;

[0045] Based on the similarity between the target item and the plurality of standard item information, the standard item information of the target item is determined from the plurality of standard item information.

[0046] According to another aspect of an embodiment of the present application, a device for processing item information is provided, the device comprising:

[0047] An acquisition module is used to obtain the attribute information and image of the target object;

[0048] a first determining module, configured to determine a first semantic feature corresponding to the attribute information and a second semantic feature corresponding to the object image;

[0049] a fusion module, configured to fuse the first semantic feature and the second semantic feature to obtain a target semantic feature of the target object;

[0050] The second determining module is configured to determine the standard item information of the target item based on the target semantic feature.

[0051] In one possible implementation, the attribute information includes at least two of the following: an item name, a merchant name, and tag information; and the first determining module includes:

[0052] a first determining unit, configured to determine at least two semantic vectors among a first semantic vector, a second semantic vector, and a third semantic vector, wherein the first semantic vector is a semantic vector corresponding to the item name, the second semantic vector is a semantic vector corresponding to the merchant name, and the third semantic vector is a semantic vector corresponding to the tag information;

[0053] The second determining unit is configured to determine the first semantic feature based on the at least two semantic vectors.

[0054] In another possible implementation, the second determination unit is used to splice the at least two semantic vectors to obtain a target semantic vector, or to add category information to each semantic vector and splice the at least two semantic vectors after adding the category information to obtain a target semantic vector; and determine the first semantic feature corresponding to the target semantic vector.

[0055] In another possible implementation, the second determining unit is configured to input the target semantic vector into a semantic vector model and output the first semantic feature, and the semantic vector model is configured to convert the semantic vector into a semantic feature.

[0056] In another possible implementation, the first determining module further includes:

[0057] a third determining unit, configured to determine at least one region of interest of the object image;

[0058] a fourth determining unit, configured to determine, for each region of interest, a second semantic vector corresponding to the region of interest, to obtain at least one second semantic vector;

[0059] A fifth determining unit is configured to determine the second semantic feature based on the at least one second semantic vector.

[0060] In another possible implementation, the fusion module includes:

[0061] a conversion unit, configured to convert the first semantic feature and the second semantic feature into semantic features of the same dimension to obtain a first standard semantic feature and a second standard semantic feature;

[0062] a sixth determining unit, configured to determine a first attention feature corresponding to the first standard semantic feature and a second attention feature corresponding to the second standard semantic feature;

[0063] A fusion unit is used to fuse the first attention feature and the second attention feature to obtain the target semantic feature.

[0064] In another possible implementation, the sixth determination unit is used to determine the first query vector, first key vector and first value vector corresponding to the first standard semantic feature and the second query vector, second key vector and second value vector corresponding to the second standard semantic feature; determine the first attention weight and the second attention weight based on the first query vector, the first key vector, the second query vector and the second key vector; perform weighted processing on the first standard semantic feature based on the first value vector and the first attention weight to obtain the first attention feature, and perform weighted processing on the second standard semantic feature based on the second value vector and the second attention weight to obtain the second attention feature.

[0065] In another possible implementation, the sixth determination unit is used to determine the first attention weight based on a first similarity, a second similarity and the first value vector, where the first similarity is the similarity between the first query vector and the first key vector, and the second similarity is the similarity between the second query vector and the first key vector; and to determine the second attention weight based on a third similarity, a fourth similarity and the second value vector, where the third similarity is the similarity between the first query vector and the second key vector, and the fourth similarity is the similarity between the second query vector and the second key vector.

[0066] In another possible implementation, the first attention weight includes a first weight and a second weight; the sixth determination unit is used to normalize the first similarity to obtain the first weight, and normalize the second similarity to obtain the second weight; and determine the first attention weight based on the product between the first weight and the first value vector and the product between the second weight and the first value vector.

[0067] In another possible implementation, the second attention weight includes a third weight and a fourth weight; the sixth determination unit is used to normalize the third similarity to obtain the third weight, and normalize the fourth similarity to obtain the fourth weight; and determine the second attention weight based on the product between the third weight and the second value vector and the product between the fourth weight and the second value vector.

[0068] In another possible implementation, the fusion unit is used to fuse the first attention feature and the second attention feature to obtain an attention feature; and perform pooling processing on the attention feature to obtain the target semantic feature.

[0069] In another possible implementation, the second determination module is configured to determine the similarity between the target item and the multiple standard item information based on the target semantic features and the semantic features of the multiple standard item information; and determine the standard item information of the target item from the multiple standard item information based on the similarity between the target item and the multiple standard item information.

[0070] According to another aspect of an embodiment of the present application, a server is provided, comprising: a processor and a memory, wherein the memory stores at least one program code, and the at least one program code is loaded and executed by the processor to implement the operations in the method for processing item information described in any of the above possible implementations.

[0071] According to another aspect of an embodiment of the present application, a computer-readable storage medium is provided, in which at least one program code is stored. The at least one program code is loaded by a processor and has the function of implementing the operations performed in the method for processing item information.

[0072] According to another aspect of an embodiment of the present application, a computer program product or computer program is provided, comprising computer program code stored in a computer-readable storage medium. A processor of a computer device reads the computer program code from the computer-readable storage medium and executes the computer program code, causing the computer device to perform the operations performed by the method for determining item information processing in any of the possible implementations described above.

[0073] The beneficial effects of the technical solutions provided by the embodiments of the present application include at least:

[0074] An embodiment of the present application provides a method for processing item information. Since the first semantic feature is the semantic feature corresponding to the attribute information of the item, and the second semantic feature is the semantic feature corresponding to the image of the item, the target semantic feature obtained by fusing the first semantic feature and the second semantic feature fuses multiple semantic features of different modalities. Therefore, determining the standard item information through the target semantic feature can improve the accuracy of the determined standard item information. BRIEF DESCRIPTION OF THE DRAWINGS

[0075] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the following briefly introduces the drawings required for use in the description of the embodiments. 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 any creative work.

[0076] Figure 1 This is a schematic diagram of an implementation environment provided by an embodiment of the present application;

[0077] Figure 2 This is a flowchart of a method for processing item information provided by an embodiment of the present application;

[0078] Figure 3 This is a flowchart of a method for processing item information provided by an embodiment of the present application;

[0079] Figure 4 is a schematic diagram of a method for determining a first semantic feature and a second semantic feature provided by an embodiment of the present application;

[0080] Figure 5 is a schematic diagram of a method for determining a first semantic feature provided by an embodiment of the present application;

[0081] Figure 6 is a schematic diagram of a method for determining target semantic features through a self-attention mechanism provided by an embodiment of the present application;

[0082] Figure 7 is a schematic diagram of a method for processing item information provided in an embodiment of the present application;

[0083] Figure 8 This is a block diagram of an item information processing device provided by an embodiment of the present application;

[0084] Figure 9 This is a block diagram of an item information processing device provided by an embodiment of the present application;

[0085] Figure 10 This is a structural diagram of a server provided in an embodiment of the present application. DETAILED DESCRIPTION

[0086] In order to make the objectives, technical solutions and advantages of this application clearer, the implementation methods of this application will be further described in detail below with reference to the accompanying drawings.

[0087] Figure 1 This is a schematic diagram of an implementation environment provided by an embodiment of the present application. Figure 1 , the implementation environment includes a terminal 101 and a server 102.

[0088] Terminal 101 and server 102 are connected via a wireless or wired network. Furthermore, a client provided by server 102 may be installed on terminal 101. The user of terminal 101 can use this client to communicate with server 102, for example, to perform functions such as data transmission and message exchange. The client may be a client installed on terminal 101 that includes Internet access. For example, the client may be a browser, social networking application, gaming application, or food delivery application.

[0089] The terminal 101 may be a computer, a mobile phone, a tablet computer or other electronic devices. The server 102 may be a single server, a server cluster consisting of several servers, or a cloud computing service center.

[0090] The method for processing item information provided in the embodiments of this application can be applied in multiple scenarios:

[0091] For example, in a search scenario where a user orders takeout, when a merchant uploads dish information to the server, the server uses the item information processing method provided in the embodiments of this application to determine the standard dish name for the dish and associate the dish information with the standard dish name. Subsequently, when the user enters a dish name, the server retrieves a matching standard dish name based on the dish name and the stored standard dish names, and outputs the dish information associated with the standard dish name.

[0092] For another example, it is applied in a recommendation scenario when a user orders takeout: when a merchant uploads the dish information of a dish to the server, the server determines the standard dish name of the dish through the method for processing item information provided in the embodiment of this application, and associates the dish information and the standard dish name of the dish. The server determines the dish name of the user's favorite dish based on the user's historical browsing history or historical order information, obtains the standard dish name that matches the dish name based on the dish name of the dish and the stored standard dish name, and recommends the dish information associated with the standard dish name to the user. For example, based on the user's historical browsing history, the server determines that the dish name of the user's favorite dish is "Spicy Pork Intestine Mixed Noodles", and the server obtains the standard dish name that matches the dish name as "Mixed Noodles", and recommends "Mixed Noodles" to the user.

[0093] For another example, in the scenario of automatic association between the dish names uploaded by merchants and the descriptive information of the dishes: the server stores the correspondence between the standard dish names and the descriptive information. The descriptive information includes information such as the ingredients, taste, and cooking methods of the dish. When the merchant uploads the dish information of the dish to the server, the server determines the standard dish name of the dish through the method for processing item information provided in the embodiment of the present application; based on the correspondence between the stored standard dish names and the descriptive information, the server determines the descriptive information that matches the dish name, and associates the dish name uploaded by the merchant with the descriptive information of the dish.

[0094] For another example, in an operational analysis scenario for the supply and sales of dishes: when a merchant uploads dish information to the server, the server determines the standard dish name of the dish through the item information processing method provided in the embodiments of this application, and associates the dish information, sales information, and standard dish name of the dish. Later, when performing operational analysis on the dish, the server determines the supply and sales volume of the dish corresponding to the standard dish name based on the association between the dish information, sales information, and standard dish name, and performs operational analysis on the dish based on the supply and sales volume.

[0095] Figure 2 This is a flowchart of a method for processing item information provided by an embodiment of the present application. Figure 2 , the method for processing item information includes the following steps:

[0096] 201. Obtain attribute information and an image of a target object.

[0097] 202. Determine a first semantic feature corresponding to the attribute information and a second semantic feature corresponding to the object image.

[0098] 203. Fuse the first semantic feature and the second semantic feature to obtain a target semantic feature of the target object.

[0099] 204. Determine standard item information of the target item based on the target semantic feature.

[0100] In one possible implementation, the attribute information includes at least two of an item name, a merchant name, and tag information; and the process of determining the first semantic feature corresponding to the attribute information includes:

[0101] Determine at least two semantic vectors among a first semantic vector, a second semantic vector, and a third semantic vector, where the first semantic vector is a semantic vector corresponding to the item name, the second semantic vector is a semantic vector corresponding to the merchant name, and the third semantic vector is a semantic vector corresponding to the tag information;

[0102] A first semantic feature is determined based on the at least two semantic vectors.

[0103] In another possible implementation, determining the first semantic feature based on at least two semantic vectors includes:

[0104] splicing at least two semantic vectors to obtain a target semantic vector, or adding category information to each semantic vector and splicing at least two semantic vectors after adding the category information to obtain a target semantic vector;

[0105] Determine a first semantic feature corresponding to the target semantic vector.

[0106] In another possible implementation, determining the first semantic feature corresponding to the target semantic vector includes:

[0107] The target semantic vector is input into the semantic vector model, and the first semantic feature is output. The semantic vector model is used to convert the semantic vector into a semantic feature.

[0108] In another possible implementation, the process of determining the second semantic feature corresponding to the object image includes:

[0109] determining at least one region of interest in the image of the object;

[0110] For each region of interest, determining a second semantic vector corresponding to the region of interest to obtain at least one second semantic vector;

[0111] A second semantic feature is determined based on the at least one second semantic vector.

[0112] In another possible implementation, the first semantic feature and the second semantic feature are fused to obtain a target semantic feature of the target object, including:

[0113] Converting the first semantic feature and the second semantic feature into semantic features of the same dimension to obtain a first standard semantic feature and a second standard semantic feature;

[0114] determining a first attention feature corresponding to the first standard semantic feature and a second attention feature corresponding to the second standard semantic feature;

[0115] The first attention feature and the second attention feature are fused to obtain the target semantic feature.

[0116] In another possible implementation, determining a first attention feature corresponding to the first standard semantic feature and a second attention feature corresponding to the second standard semantic feature includes:

[0117] Determining a first query vector, a first key vector, and a first value vector corresponding to the first standard semantic feature, and a second query vector, a second key vector, and a second value vector corresponding to the second standard semantic feature;

[0118] determining a first attention weight and a second attention weight based on the first query vector, the first key vector, the second query vector, and the second key vector;

[0119] Based on the first value vector and the first attention weight, the first standard semantic feature is weighted to obtain the first attention feature, and based on the second value vector and the second attention weight, the second standard semantic feature is weighted to obtain the second attention feature.

[0120] In another possible implementation, determining the first attention weight and the second attention weight includes:

[0121] determining a first attention weight based on a first similarity and a second similarity, wherein the first similarity is a similarity between the first query vector and the first key vector, and the second similarity is a similarity between the second query vector and the first key vector;

[0122] A second attention weight is determined based on a third similarity and a fourth similarity, where the third similarity is a similarity between the first query vector and the second key vector, and the fourth similarity is a similarity between the second query vector and the second key vector.

[0123] In another possible implementation, the first attention weight includes a first weight and a second weight;

[0124] Determining a first attention weight based on the first similarity and the second similarity includes:

[0125] The first similarity is normalized to obtain a first weight, and the second similarity is normalized to obtain a second weight.

[0126] In another possible implementation, the second attention weight includes a third weight and a fourth weight;

[0127] Determining a second attention weight based on the third similarity and the fourth similarity includes:

[0128] The third similarity is normalized to obtain a third weight, and the fourth similarity is normalized to obtain a fourth weight.

[0129] In another possible implementation, the first attention feature and the second attention feature are fused to obtain the target semantic feature, including:

[0130] Fusing the first attention feature and the second attention feature to obtain an attention feature;

[0131] The attention features are pooled to obtain the target semantic features.

[0132] In another possible implementation, determining the standard item information of the target item based on the target semantic feature includes:

[0133] Determining the similarity between the target object and the plurality of standard object information based on the target semantic feature and the semantic features of the plurality of standard object information;

[0134] Based on the similarity between the target item and the plurality of standard item information, the standard item information of the target item is determined from the plurality of standard item information.

[0135] An embodiment of the present application provides a method for processing item information. Since the first semantic feature is the semantic feature corresponding to the attribute information of the item, and the second semantic feature is the semantic feature corresponding to the image of the item, the target semantic feature obtained by fusing the first semantic feature and the second semantic feature fuses multiple semantic features of different modalities. Therefore, determining the standard item information through the target semantic feature can improve the accuracy of the determined standard item information.

[0136] Figure 3 This is a flowchart of a method for processing item information provided by an embodiment of the present application, which is executed by a server. Figure 3 , the method for processing item information includes the following steps:

[0137] 301. The server obtains attribute information and an image of a target item.

[0138] Optionally, the target item is a product listed by a merchant on an internet platform. This product can be a food item, such as fish-flavored shredded pork, or a clothing item, such as jeans. The attribute information includes at least one of the item name, the merchant name, and tag information. In one possible implementation, the attribute information includes at least the item name. In another possible implementation, the attribute information also includes at least one of the merchant name and tag information. The item image is an image of the target item.

[0139] In one possible implementation, the server receives the item attribute information and item image uploaded by the terminal in real time. Accordingly, this step involves the merchant uploading the item attribute information and item image to the server via the terminal, and the server receiving the uploaded item attribute information and item image to obtain the target item attribute information and item image.

[0140] In the embodiment of the present application, since the server receives the attribute information and item image of an item each time the terminal uploads the attribute information and item image of an item, it is ensured that the server can promptly receive the attribute information and item image of the item uploaded by the merchant, thereby improving the timeliness of obtaining the attribute information and item image of the target item.

[0141] It should be noted that each time the server obtains attribute information and an image of a target item, it can directly execute step 302 to process the attribute information and image of the target item. In another possible implementation, the server batch processes the attribute information and images of multiple target items obtained within a first preset time period. Accordingly, the specific steps are as follows: in response to the reception time period reaching the first preset time period, the server processes the attribute information and image of each received target item using the item information processing method of this application. In this embodiment of the application, the value of the first preset time period is not specifically limited and can be set and modified as needed. For example, the first preset time period is 5 minutes, 10 minutes, 60 minutes, etc.

[0142] In the embodiment of the present application, since the server periodically processes the attribute information and item images of the items uploaded by the terminal, the amount of item information processed each time is increased, thereby improving the processing efficiency of the item information.

[0143] In one possible implementation, merchants upload more item information during the day and less at night. The server can also adjust the processing method based on the time range of the current time. Accordingly, the specific steps are as follows: the server determines the time range of the current time. If the time range is daytime, the server directly executes step 302 to process the attribute information and image of each target item after obtaining the attribute information and image. If the time range is nighttime, in response to the reception duration reaching a first preset duration, the server processes the attribute information and image of each target item received using the item information processing method of this application. In this embodiment of the application, the time ranges of daytime and nighttime are not specifically limited. For example, the daytime time range is: 9:00 AM to 9:00 PM; the nighttime time range is: 9:00 PM to 9:00 AM.

[0144] In an embodiment of the present application, the server flexibly adjusts the method of processing item information according to the current time. When a large amount of item information is received, the item information is directly processed. When a small amount of item information is received, the item information is processed in batches, thereby improving the processing efficiency of the item information.

[0145] 302. The server determines a first semantic feature corresponding to the attribute information.

[0146] In one possible implementation, the attribute information includes at least two items of the item name, the merchant name, and the tag information. Accordingly, this step is as follows: the server determines at least two semantic vectors from the first semantic vector, the second semantic vector, and the third semantic vector, where the first semantic vector is the semantic vector corresponding to the item name, the second semantic vector is the semantic vector corresponding to the merchant name, and the third semantic vector is the semantic vector corresponding to the tag information; and based on the at least two semantic vectors, determines the first semantic feature. Optionally, the tag information includes the category of the item. For example, the item is "leek and egg pie," and the tag information of the item is "signature pie."

[0147] In an embodiment of the present application, since the first semantic feature is determined by multiple semantic vectors, and the multiple semantic vectors correspond to attribute information of different information sources such as item name, merchant name and label information, when determining the first semantic feature, the attribute information of different information sources can be verified with each other, thereby improving the accuracy of the determined first semantic feature.

[0148] In one possible implementation, the server determines the first semantic feature based on the at least two semantic vectors by concatenating the at least two semantic vectors to obtain a target semantic vector, and determining the first semantic feature corresponding to the target semantic vector. In this embodiment of the present application, the server improves the efficiency of determining the target semantic vector by concatenating multiple semantic vectors to obtain the target semantic vector.

[0149] In one possible implementation, the server determines the first semantic feature using a semantic vector model. Accordingly, the server determines the first semantic feature corresponding to the target semantic vector by inputting the target semantic vector into the semantic vector model and outputting the first semantic feature. The semantic vector model is used to convert the semantic vector into the semantic feature.

[0150] In an embodiment of the present application, the server determines the first semantic feature corresponding to the target semantic vector through a semantic vector model, and the semantic vector model is a pre-trained model, thereby improving the accuracy of determining the first semantic feature.

[0151] In another possible implementation, the attribute information includes at least an item name. Accordingly, the server determines a first semantic feature corresponding to the attribute information by: determining a first semantic vector, and determining a first semantic feature based on the first semantic vector; the first semantic vector is a semantic vector corresponding to the item name.

[0152] In the embodiment of the present application, since the name of the item is the main attribute information of the item, the interference of secondary attribute information can be avoided when determining the first semantic feature, thereby improving the accuracy of the determined first semantic feature.

[0153] In another possible implementation, the attribute information also includes at least one of a merchant name and tag information; accordingly, the steps of the process in which the server determines the first semantic feature corresponding to the attribute information are: the server determines a first semantic vector and at least one of the second semantic vector and the third semantic vector, the first semantic vector is the semantic vector corresponding to the item name, the second semantic vector is the semantic vector corresponding to the merchant name, and the third semantic vector is the semantic vector corresponding to the tag information; based on the first semantic vector and at least one of the second semantic vector and the third semantic vector, the first semantic feature is determined.

[0154] In an embodiment of the present application, since the first semantic feature is determined by multiple semantic vectors, and the multiple semantic vectors correspond to attribute information of different information sources such as item name, merchant name and label information, when determining the first semantic feature, the attribute information of different information sources can be verified with each other. Moreover, the multiple semantic vectors include the semantic vector corresponding to the item name, and the item name is the main attribute information of the item. In this way, when determining the first semantic feature, both the main attribute information and the mutual verification of attribute information from different information sources are taken into consideration, thereby further improving the accuracy of the determined first semantic feature.

[0155] Alternatively, see Figure 4 The semantic vector model is a BERT (Bidirectional Encoder Representation from Transformers, a pre-trained language representation) model. The item name is "Pink Lady," the merchant name is "XX Tea," and the label information is "Fruit Tea." The first semantic vector is the semantic vector corresponding to "Pink Lady," the second semantic vector is the semantic vector corresponding to "XX Tea," and the third semantic vector is the semantic vector corresponding to "Fruit Tea." The first, second, and third semantic vectors are concatenated to obtain the target semantic vector. The target semantic vector is input into the BERT model to obtain the first semantic feature.

[0156] In an embodiment of the present application, the server determines the first semantic feature corresponding to the target semantic vector through the BERT model, and the BERT model is a pre-trained model with the most advanced performance, thereby further improving the accuracy of determining the first semantic feature.

[0157] In another possible implementation, to distinguish semantic vectors of different categories, the server needs to add category information to at least two semantic vectors. Accordingly, the server determines the first semantic feature based on the at least two semantic vectors by: adding category information to each semantic vector, concatenating the at least two semantic vectors with added category information to obtain a target semantic vector; and determining the first semantic feature corresponding to the target semantic vector.

[0158] Alternatively, see Figure 5 , the type information includes [CLS] identification bit and number. The first semantic vector is represented by semantic vector i, the second semantic vector is represented by semantic vector j, and the third semantic vector is represented by semantic vector k. Add [CLS] identification bit and number to each semantic vector; where the number of semantic vector i includes ID1 to ID i ; Semantic vector j is numbered from ID1 to ID j ; Semantic vector k is numbered from ID1 to ID k The target semantic vector is concatenated by adding the [CLS] identifier and the number to the three semantic vectors. The target semantic vector is input into the BERT model to obtain the first semantic feature. Optionally, the [CLS] identifier is used to distinguish different target semantic vectors, and the number is used to distinguish the first, second, and third semantic vectors.

[0159] In an embodiment of the present application, the server adds category information to each semantic vector so that the spliced ​​target semantic vector also contains the category information. In this way, when determining the semantic feature through the semantic vector model, the first semantic feature corresponding to the target semantic vector can be determined based on the category information, thereby improving the efficiency of determining the first semantic feature.

[0160] In another possible implementation, the server first concatenates the semantic vectors and then adds the category information. Accordingly, the server obtains the target semantic vector by concatenating at least two semantic vectors to obtain a concatenated semantic vector; and then adding the category information to the concatenated semantic vector to obtain the target semantic vector.

[0161] In an embodiment of the present application, the server directly adds category information to the spliced ​​semantic vector to obtain a target semantic vector. In this way, when determining the semantic feature through the semantic vector model, the first semantic feature corresponding to the target semantic vector can be determined based on the category information, thereby improving the efficiency of determining the first semantic feature.

[0162] In one possible implementation, to ensure that the encoding lengths of target semantic vectors corresponding to different target items are consistent, the server determines the encoding length of the first semantic vector to be a first preset length, the encoding length of the second semantic vector to be a second preset length, and the encoding length of the third semantic vector to be a third preset length. Accordingly, the server determines the first, second, and third semantic vectors in the following steps: the server determines the initial semantic vector corresponding to the item name, pads or truncates the initial semantic vector to obtain a first semantic vector of the first preset length; determines the initial semantic vector corresponding to the merchant name, pads or truncates the initial semantic vector to obtain a second semantic vector of the second preset length; and determines the initial semantic vector corresponding to the tag information, pads or truncates the initial semantic vector to obtain a third semantic vector of the third preset length.

[0163] In the embodiment of the present application, the values ​​of the first preset length, the second preset length, and the third preset length are not specifically limited and can be set and modified as needed. Optionally, the server obtains the first preset length, the second preset length, and the third preset length by pre-training an MLM (Masked Language Model) based on multiple item data on the internet platform.

[0164] In an embodiment of the present application, since the encoding length of each semantic vector is a fixed length, the encoding length of the target semantic vector obtained by splicing multiple semantic vectors is a fixed length. Therefore, before the target semantic vector is input into the semantic vector model, there is no need to adjust the input length of the target semantic vector, and the first semantic feature corresponding to the target semantic vector can be determined, thereby improving the efficiency of determining the first semantic feature.

[0165] 303. The server determines a second semantic feature corresponding to the object image.

[0166] In one possible implementation, the server determines semantic features corresponding to the item image based on regions of interest within the item image. Accordingly, this step involves: the server determines at least one region of interest within the item image; for each region of interest, determines a second semantic vector corresponding to the region of interest, obtaining at least one second semantic vector; and based on the at least one second semantic vector, determines a second semantic feature.

[0167] In an embodiment of the present application, since the region of interest in the object image is a region with a high correlation with the object, the server determines the semantic feature corresponding to the object image based on the region of interest in the object image, thereby improving the correlation between the semantic feature and the object, thereby improving the accuracy of the determined second semantic feature.

[0168] Optionally, continue with Figure 4 The server extracts features from the image information using the Faster R-CNN model to obtain at least one ROI (Region of Interest) corresponding to the image information. In the embodiment of the present application, the Faster R-CNN model is an image model widely used in the image field, and the efficiency and accuracy of feature extraction can be improved by using this image model.

[0169] Optionally, the step of determining the second semantic vector corresponding to the region of interest by the server is as follows: the server performs pooling processing on each region of interest to obtain at least one second semantic vector of the same dimension. The pooling processing may be maximum pooling processing or average pooling processing.

[0170] In one possible implementation, when there is only one second semantic vector of the same dimension, the server directly determines the second semantic vector as the second semantic feature. When there are multiple second semantic vectors of the same dimension, the server adds the multiple second semantic vectors to obtain the second semantic feature.

[0171] In the embodiment of the present application, by performing pooling processing on the region of interest, the amount of data that needs to be processed is reduced, thereby improving the efficiency of determining the second semantic vector.

[0172] In another possible implementation, the server first determines a semantic vector corresponding to the entire item image and then determines a second semantic feature based on the ROI. Accordingly, the server determines the second semantic feature corresponding to the item image by: determining the overall semantic vector corresponding to the item image; identifying at least one ROI in the item image; for each ROI, determining a second semantic vector corresponding to the ROI from the overall semantic vector to obtain at least one second semantic vector; and determining the second semantic feature based on the at least one second semantic vector.

[0173] In an embodiment of the present application, since the region of interest is a region with a high correlation with the object, the second semantic vector corresponding to the region of interest is determined from the overall semantic vector based on the region of interest, which has a high correlation with the object, thereby improving the correlation between the semantic feature determined based on the second semantic vector and the object, thereby improving the accuracy of the determined second semantic feature.

[0174] 304. The server fuses the first semantic feature and the second semantic feature to obtain a target semantic feature of the target object.

[0175] In one possible implementation, the server fuses the two semantic features through a self-attention mechanism. Accordingly, the server determines the target semantic features of the target item in steps (1) to (3):

[0176] (1) The server converts the first semantic feature and the second semantic feature into semantic features of the same dimension to obtain a first standard semantic feature and a second standard semantic feature.

[0177] Optionally, the server maps the first semantic feature and the second semantic feature to a vector space of the same dimension through an MLP (Multilayer Perceptron) to obtain a first standard semantic feature and a second standard semantic feature of the same dimension.

[0178] (2) The server determines a first attention feature corresponding to the first standard semantic feature and a second attention feature corresponding to the second standard semantic feature.

[0179] In one possible implementation, the server determines the first attention feature and the second attention feature based on the attention model. Accordingly, this step includes the following steps s21-s23:

[0180] s21: The server determines a first query vector, a first key vector, and a first value vector corresponding to the first standard semantic feature, and a second query vector, a second key vector, and a second value vector corresponding to the second standard semantic feature.

[0181] Optionally, the attention model is provided with a trained first query weight, first key weight, first value weight, and second query weight, second key weight, and second value weight. Accordingly, this step is as follows: the server determines the first query weight, first key weight, and first value weight corresponding to the first standard semantic feature according to the attention model, determines the first query vector based on the product of the first standard semantic feature and the first query weight, determines the first key vector based on the product of the first standard semantic feature and the first key weight, and determines the first value vector based on the product of the first standard semantic feature and the first value weight; and, determines the second query weight, second key weight, and second value weight corresponding to the second standard semantic feature according to the attention model, determines the second query vector based on the product of the second standard semantic feature and the second query weight, determines the second key vector based on the product of the second standard semantic feature and the second key weight, and determines the second value vector based on the product of the second standard semantic feature and the second value weight.

[0182] For example, see Figure 6 , the attention model is a self-attention model. The first standard semantic feature is x t The second standard semantic feature is represented by x r Indicates that the first query weight is W Q Indicates that the first key weight is W K Indicates that the first value weight is W V The first query vector is represented by qt If q t =W Q x t ; The first key vector is k t If k t =W K x t ; The first value vector is v t If v t =W V x t The second query vector is q r If q r =W Q x r ; The second key vector is k r If k r =W K x r ; The second value vector is v r If v r =W V x r .

[0183] In an embodiment of the present application, the server determines the first attention feature and the second attention feature through an attention model, and the attention model can combine the semantic information of the context, thereby improving the accuracy of the determined first attention feature and the second attention feature.

[0184] s22: The server determines a first attention weight and a second attention weight based on the first query vector, the first key vector, the second query vector, and the second key vector.

[0185] In one possible implementation, the server determines a first attention weight and a second attention weight based on the similarity between the query vector and the key vector. Accordingly, this step involves: the server determines the first attention weight based on the first similarity between the first query vector and the first key vector and the second similarity between the second query vector and the first key vector; and determines the second attention weight based on the third similarity between the first query vector and the second key vector and the fourth similarity between the second query vector and the second key vector.

[0186] In an embodiment of the present application, the server determines the first attention weight and the second attention weight based on the similarity between the query vector and the key vector, and the query vector and the key vector include text features and image features, thereby realizing the fusion of semantic features of multiple modalities, thereby improving the accuracy of the determined first attention weight and the second attention weight.

[0187] In one possible implementation, the first attention weight includes a first weight and a second weight. Accordingly, the server determines the first attention weight by normalizing the first similarity to obtain the first weight, and normalizing the second similarity to obtain the second weight. Optionally, the server normalizes the first similarity and the second similarity using a Softmax function. For example, continue to refer to Figure 6 , the first weight is Indicates that the second weight is Indicates that the first similarity is α tt Indicates that the second similarity is α rt Indicates. Accordingly,

[0188] Optionally, the step of the server determining the first similarity between the first query vector and the first key vector is: the server determines the first similarity according to the first query vector and the first key vector using the following formula 1;

[0189] Formula 1:

[0190] Among them, α tt Indicates the first similarity, k t represents the first bond vector, q t represents the first query vector, Represents the distance between the first query vector and the first key vector.

[0191] Optionally, the step of the server determining the second similarity between the second query vector and the first key vector is: the server determines the second similarity according to the second query vector and the first key vector using the following formula 2;

[0192] Formula 2:

[0193] Among them, α rt Represents the second similarity, k t represents the first bond vector, q r represents the second query vector, represents the distance between the second query vector and the first key vector.

[0194] In one possible implementation, the second attention weight includes a third weight and a fourth weight. Accordingly, the server determines the second attention weight by normalizing the third similarity to obtain a third weight, and normalizing the fourth similarity to obtain a fourth weight. Optionally, the server normalizes the third similarity and the fourth similarity using a Softmax function. For example, continue to refer to Figure 6 , the third weight is Indicates that the fourth weight is Indicates that the third similarity is α tr Indicates that the fourth similarity is α rr Indicates. Accordingly,

[0195] Optionally, the step of the server determining the third similarity between the first query vector and the second key vector is: the server determines the third similarity according to the first query vector and the second key vector using the following formula 3;

[0196] Formula 3:

[0197] Among them, α tr Represents the third similarity, k r represents the second key vector, q t represents the first query vector, represents the distance between the first query vector and the second key vector.

[0198] Optionally, the step of the server determining the fourth similarity between the second query vector and the second key vector is: the server determines the fourth similarity according to the second query vector and the second key vector using the following formula 4;

[0199] Formula 4:

[0200] Among them, α rr Indicates the fourth similarity, k r represents the second key vector, q r represents the second query vector, represents the distance between the second query vector and the second key vector.

[0201] In the embodiment of the present application, the server normalizes the multiple similarities and converts the multiple similarities into the same numerical range, thereby improving the accuracy of the determined first standard semantic features and the second standard semantic features.

[0202] s23: The server performs weighted processing on the first standard semantic feature based on the first value vector and the first attention weight to obtain the first attention feature, and performs weighted processing on the second standard semantic feature based on the second value vector and the second attention weight to obtain the second attention feature.

[0203] In one possible implementation, this step is: the server performs weighted processing on the first standard semantic feature based on the product of the first weight and the first value vector and the product of the second weight and the first value vector to obtain the first attention feature; based on the product of the third weight and the second value vector and the product of the fourth weight and the second value vector, the server performs weighted processing on the second standard semantic feature to obtain the second attention feature.

[0204] Optionally, continue with Figure 6 , the first attention vector is z t The second attention vector is represented by z r Indicates that the first value vector is v t Indicates that the second value vector is v r The server performs weighted processing on the first standard semantic feature to obtain the first attention vector The server performs weighted processing on the second standard semantic feature to obtain the second attention vector

[0205] In an embodiment of the present application, the server performs weighted processing on the standard semantic features according to the value vector and the attention weight to obtain the attention feature, and the value vector and the attention weight both include text features and image features, thereby realizing the fusion of semantic features of multiple modalities, thereby improving the accuracy of the determined first attention vector and the second attention vector.

[0206] (3) The server fuses the first attention feature and the second attention feature to obtain the target semantic feature.

[0207] In one possible implementation, see Figure 6 ,This step is: the server fuses the first attention feature and the second attention feature to obtain the attention feature; and performs pooling processing on the attention feature to obtain the target semantic feature.

[0208] In an embodiment of the present application, the server fuses the features of the two modalities of text and image through the attention mechanism, thereby realizing the supplementation and interaction of multimodal information and ensuring the integrity and robustness of the obtained feature semantic information.

[0209] 305. The server determines the standard product information of the target item based on the target semantic information.

[0210] In one possible implementation, this step is as follows: the server determines the similarity between the target object and the multiple standard item information based on the target semantic features and the semantic features of the multiple standard item information; and determines the standard item information of the target object from the multiple standard item information based on the similarity between the target object and the multiple standard item information.

[0211] In the embodiments of this application, the text information source is enriched by introducing multiple attribute information such as item name, merchant name, and tag name. The inclusion of item images adds a cross-modal information source. The fusion of multiple text information sources and item image sources at the presentation layer overcomes the shortcomings of traditional single-source information sources and improves the accuracy of the determined standard item information. For example, using the same sample data, the multi-source information processing method implemented in this application was compared with the traditional single-source information processing method. The traditional processing method achieved an accuracy rate of 97.24%, while the processing method implemented in this application achieved an accuracy rate of 99.81%.

[0212] In a possible implementation, the plurality of standard item information is information in a standard item information library; the server determines the semantic features of the plurality of standard item information by using a BRET model; and determines the similarity between the target item and the plurality of standard item information by using cosine similarity. Figure 7 For each standard item, the server implements a dual-tower model consisting of three modules: the input layer, the representation layer, and the matching layer. The server inputs the target item's attribute information, the item image, and the standard item information into the input layer. In the representation layer, the server uses the BRET model and the Fast R-CNN model to determine the multimodal semantic features corresponding to the target item and the textual semantic features corresponding to the standard item information. The server then uses cosine similarity to determine the similarity between the target item and the standard item information.

[0213] An embodiment of the present application provides a method for processing item information. Since the first semantic feature is the semantic feature corresponding to the attribute information of the item, and the second semantic feature is the semantic feature corresponding to the image of the item, the target semantic feature obtained by fusing the first semantic feature and the second semantic feature fuses multiple semantic features of different modalities. Therefore, determining the standard item information through the target semantic feature can improve the accuracy of the determined standard item information.

[0214] Figure 8 This is a schematic diagram of the structure of an item information processing device provided in an embodiment of the present application. Figure 8 , the device comprises:

[0215] Acquisition module 801, used to acquire attribute information and item image of target item;

[0216] A first determining module 802 is configured to determine a first semantic feature corresponding to the attribute information and a second semantic feature corresponding to the object image;

[0217] A fusion module 803 is configured to fuse the first semantic feature and the second semantic feature to obtain a target semantic feature of the target object;

[0218] The second determining module 804 is configured to determine standard item information of the target item based on the target semantic feature.

[0219] In one possible implementation, see Figure 9 The attribute information includes at least two items of the item name, the merchant name, and the tag information; the first determination module 802 includes:

[0220] A first determining unit 8021 is configured to determine at least two semantic vectors among a first semantic vector, a second semantic vector, and a third semantic vector, wherein the first semantic vector is a semantic vector corresponding to the item name, the second semantic vector is a semantic vector corresponding to the merchant name, and the third semantic vector is a semantic vector corresponding to the tag information;

[0221] The second determining unit 8022 is configured to determine a first semantic feature based on at least two semantic vectors.

[0222] In another possible implementation, the second determination unit 8022 is used to splice at least two semantic vectors to obtain a target semantic vector, or to add category information to each semantic vector and splice at least two semantic vectors after adding the category information to obtain a target semantic vector; and determine the first semantic feature corresponding to the target semantic vector.

[0223] In another possible implementation, the second determining unit 8022 is configured to input the target semantic vector into a semantic vector model and output a first semantic feature. The semantic vector model is configured to convert the semantic vector into a semantic feature.

[0224] In another possible implementation, see Figure 9 The first determining module 802 further includes:

[0225] The third determining unit 8023 is configured to determine at least one region of interest of the object image;

[0226] The fourth determining unit 8024 is configured to determine, for each region of interest, a second semantic vector corresponding to the region of interest, to obtain at least one second semantic vector;

[0227] The fifth determining unit 8025 is configured to determine a second semantic feature based on at least one second semantic vector.

[0228] In another possible implementation, see Figure 9 , the fusion module 803 includes:

[0229] A conversion unit 8031 ​​is configured to convert the first semantic feature and the second semantic feature into semantic features of the same dimension to obtain a first standard semantic feature and a second standard semantic feature;

[0230] A sixth determining unit 8032 is configured to determine a first attention feature corresponding to the first standard semantic feature and a second attention feature corresponding to the second standard semantic feature;

[0231] The fusion unit 8033 is used to fuse the first attention feature and the second attention feature to obtain the target semantic feature.

[0232] In another possible implementation, the sixth determination unit 8032 is used to determine the first query vector, the first key vector and the first value vector corresponding to the first standard semantic feature and the second query vector, the second key vector and the second value vector corresponding to the second standard semantic feature; determine the first attention weight and the second attention weight based on the first query vector, the first key vector, the second query vector and the second key vector; perform weighted processing on the first standard semantic feature based on the first value vector and the first attention weight to obtain the first attention feature, and perform weighted processing on the second standard semantic feature based on the second value vector and the second attention weight to obtain the second attention feature.

[0233] In another possible implementation, the sixth determination unit 8032 is used to determine a first attention weight based on a first similarity, a second similarity, and a first value vector, where the first similarity is the similarity between the first query vector and the first key vector, and the second similarity is the similarity between the second query vector and the first key vector; and to determine a second attention weight based on a third similarity, a fourth similarity, and a second value vector, where the third similarity is the similarity between the first query vector and the second key vector, and the fourth similarity is the similarity between the second query vector and the second key vector.

[0234] In another possible implementation, the first attention weight includes a first weight and a second weight; the sixth determination unit 8032 is used to normalize the first similarity to obtain the first weight, and normalize the second similarity to obtain the second weight; the first attention weight is determined based on the product between the first weight and the first value vector and the product between the second weight and the first value vector.

[0235] In another possible implementation, the second attention weight includes a third weight and a fourth weight; the sixth determination unit 8032 is used to normalize the third similarity to obtain a third weight, and to normalize the fourth similarity to obtain a fourth weight; the second attention weight is determined based on the product between the third weight and the second value vector and the product between the fourth weight and the second value vector.

[0236] In another possible implementation, the fusion unit 8033 is used to fuse the first attention feature and the second attention feature to obtain an attention feature; and perform pooling processing on the attention feature to obtain a target semantic feature.

[0237] In another possible implementation, the second determination module 804 is configured to determine the similarity between the target item and the multiple standard item information based on the target semantic features and the semantic features of the multiple standard item information; and determine the standard item information of the target item from the multiple standard item information based on the similarity between the target item and the multiple standard item information.

[0238] An embodiment of the present application provides a device for processing item information. Since the first semantic feature is a semantic feature corresponding to the attribute information of the item, and the second semantic feature is a semantic feature corresponding to the image of the item, the target semantic feature obtained by fusing the first semantic feature and the second semantic feature fuses multiple semantic features of different modalities. Therefore, determining the standard item information through the target semantic feature can improve the accuracy of the determined standard item information.

[0239] It should be noted that the aforementioned embodiments of the item information processing device, when processing item information, illustrate the division of the aforementioned functional modules only as an example. In actual applications, the aforementioned functions can be assigned to different functional modules as needed, i.e., the internal structure of the server can be divided into different functional modules to perform all or part of the functions described above. Furthermore, the item information processing device and the item information processing method embodiments provided in the aforementioned embodiments share the same concept. The specific implementation process is detailed in the method embodiments and will not be further elaborated here.

[0240] Figure 10 1 is a schematic diagram of the structure of a server provided in an embodiment of the present application. Server 1000 may vary significantly depending on configuration or performance, and may include one or more processors (central processing units, CPUs) 1001 and one or more memories 1002. Memory 1002 stores at least one program code, which is loaded and executed by processor 1001 to implement the item information processing methods provided in the various method embodiments described above. Of course, the server may also include components such as a wired or wireless network interface, a keyboard, and input / output interfaces for input and output. The server may also include other components for implementing device functions, which are not detailed here.

[0241] An embodiment of the present application further provides a computer-readable storage medium, in which at least one program code is stored. The at least one program code is loaded by a processor and has the function of implementing the operations in the method for processing item information of the above embodiment.

[0242] An embodiment of the present application further provides a computer program product. When the instructions in the computer program product are executed by a processor of a server, the server is enabled to execute the method for processing item information in each of the above method embodiments.

[0243] Those skilled in the art will understand that all or part of the steps of implementing the above embodiments may be accomplished by hardware, or by a program instructing the relevant hardware to accomplish the steps. The program may be stored in a storage medium, and the above-mentioned storage medium may be a read-only memory, a disk, or an optical disk, etc.

[0244] The above description is merely an optional embodiment of the embodiments of the present application and is not intended to limit the embodiments of the present application. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principles of the embodiments of the present application shall be included in the scope of protection of the present application.

Claims

1. A method for processing item information, characterized in that: The method comprises: Obtaining attribute information and item images of the target item; Determining a first semantic feature corresponding to the attribute information and a second semantic feature corresponding to the object image; fusing the first semantic feature and the second semantic feature to obtain a target semantic feature of the target object; Determining standard item information of the target item based on the target semantic feature; The attribute information includes at least two items of an item name, a merchant name, and tag information; and the process of determining a first semantic feature corresponding to the attribute information includes: Determine at least two semantic vectors among a first semantic vector, a second semantic vector, and a third semantic vector, wherein the first semantic vector is a semantic vector corresponding to the item name, the second semantic vector is a semantic vector corresponding to the merchant name, and the third semantic vector is a semantic vector corresponding to the tag information; determining the first semantic feature based on the at least two semantic vectors; The determining the first semantic feature based on the at least two semantic vectors includes: splicing the at least two semantic vectors to obtain a target semantic vector, or adding category information to each semantic vector and splicing the at least two semantic vectors after adding the category information to obtain a target semantic vector; Determining the first semantic feature corresponding to the target semantic vector; The fusing the first semantic feature and the second semantic feature to obtain the target semantic feature of the target object includes: Converting the first semantic feature and the second semantic feature into semantic features of the same dimension to obtain a first standard semantic feature and a second standard semantic feature; determining a first attention feature corresponding to the first standard semantic feature and a second attention feature corresponding to the second standard semantic feature; fusing the first attention feature and the second attention feature to obtain the target semantic feature; The determining of a first attention feature corresponding to the first standard semantic feature and a second attention feature corresponding to the second standard semantic feature includes: Determining a first query vector, a first key vector, and a first value vector corresponding to the first standard semantic feature, and a second query vector, a second key vector, and a second value vector corresponding to the second standard semantic feature; determining a first attention weight and a second attention weight based on the first query vector, the first key vector, the second query vector, and the second key vector; Based on the first value vector and the first attention weight, the first standard semantic feature is weighted to obtain the first attention feature; based on the second value vector and the second attention weight, the second standard semantic feature is weighted to obtain the second attention feature.

2. The method according to claim 1, characterized in that The determining the first semantic feature corresponding to the target semantic vector includes: The target semantic vector is input into a semantic vector model, and the first semantic feature is output. The semantic vector model is used to convert the semantic vector into a semantic feature.

3. The method according to claim 1, characterized in that The process of determining the second semantic feature corresponding to the object image includes: determining at least one region of interest in the image of the object; For each region of interest, determining a second semantic vector corresponding to the region of interest to obtain at least one second semantic vector; The second semantic feature is determined based on the at least one second semantic vector.

4. The method according to claim 1, wherein The determining, based on the first query vector, the first key vector, the second query vector, and the second key vector, a first attention weight and a second attention weight, includes: determining the first attention weight based on a first similarity and a second similarity, wherein the first similarity is a similarity between the first query vector and the first key vector, and the second similarity is a similarity between the second query vector and the first key vector; The second attention weight is determined based on a third similarity and a fourth similarity, wherein the third similarity is the similarity between the first query vector and the second key vector, and the fourth similarity is the similarity between the second query vector and the second key vector.

5. The method according to claim 4, characterized in that The first attention weight includes a first weight and a second weight; The determining the first attention weight based on the first similarity and the second similarity includes: The first similarity is normalized to obtain the first weight, and the second similarity is normalized to obtain the second weight.

6. The method according to claim 4, characterized in that The second attention weight includes a third weight and a fourth weight; The determining the second attention weight based on the third similarity and the fourth similarity includes: The third similarity is normalized to obtain the third weight, and the fourth similarity is normalized to obtain the fourth weight.

7. The method according to claim 1, characterized in that The fusing the first attention feature and the second attention feature to obtain the target semantic feature includes: fusing the first attention feature and the second attention feature to obtain an attention feature; The attention features are pooled to obtain the target semantic features.

8. The method according to any one of claims 1 to 7, characterized in that The determining, based on the target semantic feature, the standard item information of the target item includes: Determining similarities between the target object and the plurality of standard object information based on the target semantic features and the semantic features of the plurality of standard object information; Based on the similarity between the target item and the plurality of standard item information, the standard item information of the target item is determined from the plurality of standard item information.

9. An item information processing device, characterized in that: The device comprises: An acquisition module is used to obtain the attribute information and image of the target object; a first determining module, configured to determine a first semantic feature corresponding to the attribute information and a second semantic feature corresponding to the object image; a fusion module, configured to fuse the first semantic feature and the second semantic feature to obtain a target semantic feature of the target object; a second determining module, configured to determine standard item information of the target item based on the target semantic feature; The acquisition module further includes: the attribute information includes at least two items of the item name, the merchant name, and the tag information; and the process of determining the first semantic feature corresponding to the attribute information includes: Determine at least two semantic vectors among a first semantic vector, a second semantic vector, and a third semantic vector, wherein the first semantic vector is a semantic vector corresponding to the item name, the second semantic vector is a semantic vector corresponding to the merchant name, and the third semantic vector is a semantic vector corresponding to the tag information; determining the first semantic feature based on the at least two semantic vectors; The determining the first semantic feature based on the at least two semantic vectors includes: splicing the at least two semantic vectors to obtain a target semantic vector, or adding category information to each semantic vector and splicing the at least two semantic vectors after adding the category information to obtain a target semantic vector; Determining the first semantic feature corresponding to the target semantic vector; The fusion module further includes: fusing the first semantic feature and the second semantic feature to obtain the target semantic feature of the target object, including: Converting the first semantic feature and the second semantic feature into semantic features of the same dimension to obtain a first standard semantic feature and a second standard semantic feature; determining a first attention feature corresponding to the first standard semantic feature and a second attention feature corresponding to the second standard semantic feature; fusing the first attention feature and the second attention feature to obtain the target semantic feature; The determining of a first attention feature corresponding to the first standard semantic feature and a second attention feature corresponding to the second standard semantic feature includes: Determining a first query vector, a first key vector, and a first value vector corresponding to the first standard semantic feature, and a second query vector, a second key vector, and a second value vector corresponding to the second standard semantic feature; determining a first attention weight and a second attention weight based on the first query vector, the first key vector, the second query vector, and the second key vector; Based on the first value vector and the first attention weight, the first standard semantic feature is weighted to obtain the first attention feature; based on the second value vector and the second attention weight, the second standard semantic feature is weighted to obtain the second attention feature.

10. A server, characterized in that: The server includes: A processor and a memory, wherein the memory stores at least one instruction, and the at least one instruction is loaded and executed by the processor to implement the operation of the method for processing item information according to any one of claims 1 to 7.

11. A computer-readable storage medium, characterized in that The computer-readable storage medium stores at least one instruction, and the at least one instruction is loaded and executed by a processor to implement the operations performed in the method for processing item information according to any one of claims 1 to 7.

Images