Article recommendation method and device, computer equipment, storage medium and program product

Abstract

The invention relates to an article recommendation method and device, computer equipment, a storage medium and a program product. The method comprises the steps that subject detection is conducted on a plurality of target images in published content, subject detection results are obtained, and the target images comprise at least one article; determining a plurality of candidate articles from a preset article library according to the main body detection result, and sorting the plurality of candidate articles to obtain an article sorting result; and determining a target article according to the article sorting result and a preset article set, and recommending the target article. By adopting the method, the searching efficiency can be improved, and the recommendation effect can be improved.

Description

Technical Field

[0001] The present application relates to the field of Internet technology, and in particular to an item recommendation method, device, computer equipment, storage medium and program product. Background Art

[0002] With the development of Internet technology, many applications have integrated multiple functions such as instant messaging, content publishing, shopping, and live broadcasting. In some scenarios, users browse the product content published by merchants in the application and enter the merchant's live broadcast room based on the product content; however, after entering the live broadcast room, users will not directly see the products in the product content, but need to find the same product in the product content among the products sold in the live broadcast room, and perform operations such as viewing product details and purchasing products.

[0003] During the above search process, the user may need to compare repeatedly or browse a large number of products before finding the same product, which results in low search efficiency. Summary of the invention

[0004] Based on this, it is necessary to provide an item recommendation method, device, computer equipment, storage medium and program product that can improve search efficiency and recommendation effect in response to the above technical problems.

[0005] In a first aspect, the present application provides an item recommendation method, the method comprising:

[0006] Perform subject detection on multiple target images in the published content respectively to obtain subject detection results, where the target images include at least one object;

[0007] Determine multiple candidate items from a preset item library according to the subject detection result, and sort the multiple candidate items to obtain an item sorting result;

[0008] Determine the target item based on the item sorting results and the preset item set, and recommend the target item.

[0009] In some embodiments, the subject detection result includes an object subject in the target image and a subject level of the object subject, the subject level being used to characterize the degree to which the object subject represents the target image, and multiple candidate objects are determined from a preset object library according to the subject detection result, and the multiple candidate objects are sorted to obtain an object sorting result, including:

[0010] Find candidate items corresponding to each item body from the preset item library;

[0011] Determine the detection score of the candidate items corresponding to each object subject based on the object subject level and the similarity between the object subject and each candidate item;

[0012] The items are sorted according to the detection scores of the candidate items corresponding to each item body to obtain the item sorting result.

[0013] In some embodiments, based on the subject level of the object subject and the similarity between the object subject and each candidate object, determining the detection score of the candidate object corresponding to each object subject includes:

[0014] Determine the subject score of each candidate item according to the subject level of the item subject;

[0015] Determine the similarity score of each candidate item according to the similarity between the item and each candidate item;

[0016] For the candidate items corresponding to each item subject, the detection score of the candidate item is obtained according to the subject score and similarity score of the candidate item.

[0017] In some embodiments, the items are sorted according to the detection scores of the candidate items corresponding to each item body to obtain the item sorting result, including:

[0018] For the first target image, the first candidate items are sorted according to the detection scores of the first candidate items corresponding to the object bodies to obtain the first image sorting result;

[0019] For other target images, the second candidate objects are sorted according to the detection scores of the second candidate objects corresponding to the object bodies, and the sorting results of other images are obtained;

[0020] The first candidate item and the second candidate item are sorted based on the first image sorting result and the other image sorting results to obtain an item sorting result.

[0021] In some embodiments, the first candidate item and the second candidate item are sorted based on the first image sorting result and the other image sorting results to obtain the item sorting result, including:

[0022] For the candidate items of each subject score, according to the ranking of each candidate item in the first picture ranking result and the ranking of each candidate item in the other picture ranking results, the first candidate item and the second candidate item are selected in turn for ranking to obtain the initial ranking result of each subject score;

[0023] According to the size of the subject score, the initial sorting results of multiple subject scores are spliced ​​to obtain the item sorting result.

[0024] In some embodiments, searching for candidate items corresponding to each item body from a preset item library includes:

[0025] For each item body, the item body is matched with each item in the preset item library to obtain the probability of whether each item and the item body are the same item;

[0026] According to the probability of whether each item and the item body are the same item and the preset probability threshold, the candidate items corresponding to the item body are found.

[0027] In some embodiments, subject detection is performed on multiple target images in the published content to obtain subject detection results, including:

[0028] For each target image, the target image is input into a pre-established subject detection model to obtain the subject position of each object subject in the target image;

[0029] The subject level of each object subject is determined based on the subject position.

[0030] In some embodiments, determining the target item according to the item sorting result and the preset item set includes:

[0031] Determine the search item from multiple candidate items according to the item ranking result;

[0032] Find matching items in the preset item collection according to the search item;

[0033] If a matching object is found, the matching object is determined as the target object.

[0034] In some embodiments, it also includes:

[0035] If no matching item is found, the search item is re-determined from the plurality of candidate items according to the item ranking result, and the step of searching for a matching item in a preset item set according to the search item is performed.

[0036] In a second aspect, the present application also provides an item recommendation device, comprising:

[0037] A subject detection module, used to perform subject detection on multiple target images in the published content respectively to obtain subject detection results, wherein the target image includes at least one object;

[0038] A sorting module, used to determine a plurality of candidate items from a preset item library according to the subject detection result, and sort the plurality of candidate items to obtain an item sorting result;

[0039] The recommendation module is used to determine the target item based on the item sorting results and the preset item set, and recommend the target item.

[0040] In a third aspect, the present application further provides a computer device, including a memory, a processor and a display unit, wherein the memory stores a computer program, and when the processor executes the computer program, the steps of any one of the first aspect are implemented.

[0041] In a fourth aspect, the present application further provides a computer-readable storage medium having a computer program stored thereon, which implements any of the steps in the first aspect when the computer program is executed by a processor.

[0042] In a fifth aspect, the present application further provides a computer program product, comprising a computer program, which implements any step in the first aspect when executed by a processor.

[0043] The above-mentioned method, device, computer equipment, storage medium and program product for recommending items respectively perform subject detection on multiple target images in the published content to obtain subject detection results; determine multiple candidate items from a preset item library based on the subject detection results, and sort the multiple candidate items to obtain item sorting results; determine the target item based on the item sorting results and the preset item set, and recommend the target item. The embodiment of the present application can automatically detect images in the published content and automatically recommend items based on the detection results, without the need for users to search, compare and browse, which can not only improve the search efficiency, but also improve the search accuracy and conversion rate. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the related technologies, the drawings required for use in the embodiments or the related technical descriptions are briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0045] Figure 1 A diagram of an application environment of an item recommendation method in an embodiment;

[0046] Figure 2 A schematic diagram of a flow chart of an item recommendation method in one embodiment;

[0047] Figure 3 A schematic diagram of a flow chart of a step of sorting multiple candidate items in one embodiment;

[0048] Figure 4 A schematic diagram of a flow chart of a step of searching for candidate items in one embodiment;

[0049] Figure 5 A schematic diagram of a flow chart of a step of determining a detection score of a candidate object in one embodiment;

[0050] Figure 6 A schematic diagram of a flow chart of a step of determining similarity scores of candidate items in one embodiment;

[0051] Figure 7is a flow chart of a step of sorting multiple candidate objects according to detection scores in one embodiment;

[0052] Figure 8 A schematic diagram of a process of sorting multiple images in one embodiment;

[0053] Fig. 9 It is a flowchart of the steps of sorting the first picture and other pictures in one embodiment;

[0054] Fig.10 is a flow chart of a subject detection step in one embodiment;

[0055] Fig.11 A schematic diagram of a process for determining a target object in one embodiment;

[0056] Fig.12 is a structural block diagram of an item recommendation device in one embodiment;

[0057] Fig.13 FIG. 4 is a diagram showing the internal structure of a computer device in one embodiment. DETAILED DESCRIPTION

[0058] In order to make the purpose, technical solution and advantages of the present application more clearly understood, the present application is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0059] First of all, before specifically introducing the technical solution of the embodiment of the present application, the technical background or technical evolution context on which the embodiment of the present application is based is introduced. With the development of Internet technology, many applications integrate multiple functions such as instant messaging, content publishing, shopping, and live broadcasting. In some scenarios, users browse the product content published by merchants in the application and enter the merchant's live broadcast room based on the product content; however, after entering the live broadcast room, users will not directly see the products in the product content, but need to search for the same products in the product content among the products sold in the live broadcast room, and view product details, purchase products, and other operations. In the above search process, users may need to compare repeatedly, or browse a large number of products to find the same product, and the search efficiency is low.

[0060] In view of the above problems, the embodiment of the present application provides an item recommendation scheme, which first performs subject detection on multiple target images in the published content to obtain subject detection results; then, multiple candidate items are determined from the preset item library based on the subject detection results, and the multiple candidate items are sorted to obtain item sorting results; finally, the target item is determined based on the item sorting results and the preset item set, and the target item is recommended. The embodiment of the present application can automatically detect images in the published content and automatically recommend items based on the detection results, without the need for users to search, compare and browse, which can not only improve the search efficiency, but also improve the search accuracy and conversion rate.

[0061] The item recommendation method provided in the embodiment of the present application can be applied to Figure 1 In the application environment shown. The application environment includes a terminal 102 and a server 104, and the terminal 102 communicates with the server 104 through a network. The data storage system can store data that the server 104 needs to process. The data storage system can be integrated on the server 104, or it can be placed on the cloud or other network servers. The above-mentioned terminal 102 can be, but is not limited to, various personal computers, laptops, smart phones, tablet computers, Internet of Things devices and portable wearable devices. The Internet of Things devices can be smart TVs, smart car-mounted devices, etc. Portable wearable devices can be head-mounted devices, etc. The server 104 can be implemented with an independent server or a server cluster consisting of multiple servers.

[0062] In an exemplary embodiment, Figure 2 As shown, a method for recommending items is provided, which is applied to Figure 1 Taking the server in as an example, the following steps may be included:

[0063] Step 201 , subject detection is performed on multiple target images in the published content to obtain subject detection results.

[0064] Usually, there are one or more target images in the published content, and each target image includes at least one item. For example, target image 1 includes items A and B, mainly showing item A, and item B is located behind item A; target image 2 includes items C, D, and E, mainly showing item C, and items D and E are located behind item C; target image 3 also includes items F and G, mainly showing item F, and item G is located behind item F.

[0065] After detecting that the user enters the live broadcast room based on the published content, the terminal sends a recommendation request to the server, which may carry the identifier of the published content. The server receives the recommendation request, searches for the published content according to the identifier of the published content carried in the recommendation request, and determines multiple target images in the published content. Afterwards, the server performs subject detection on the multiple target images respectively, and detects at least one object in each target image.

[0066] The subject detection may be performed in the following manner: the server pre-sets a subject detection algorithm, calculates pixel values ​​of all pixels of each target image using the subject detection algorithm, and detects at least one object in the target image based on the calculation result.

[0067] It should be noted that the subject detection method is not limited to the above examples and can be set according to actual conditions.

[0068] Step 202 , determining a plurality of candidate items from a preset item library according to the subject detection result, and sorting the plurality of candidate items to obtain an item sorting result.

[0069] Among them, the preset item library includes all items sold by the merchant.

[0070] After obtaining the subject detection result, if the subject detection result includes an item, the server matches the item with the items in the preset item library, and uses the matched items in the preset item library as candidate items. It can be understood that there may be multiple matching items in the preset item library, and multiple matching items are used as candidate items.

[0071] If the subject detection result includes multiple objects, the server matches each object with objects in a preset object library, and uses multiple objects matched from the preset object library as candidate objects.

[0072] Furthermore, a search is performed for the subject detection result of each target image, so that a plurality of candidate objects can be found.

[0073] The server sorts the multiple candidate items found according to the preset sorting rules to obtain the item sorting results. For example, when matching, the similarity between the items in the subject detection result and the items in the preset item library is calculated; after finding multiple candidate items, each candidate item has a corresponding similarity; when sorting, the multiple candidate items can be sorted in order from high to low similarity.

[0074] It should be noted that the sorting method is not limited to the above examples. In practical applications, other sorting methods can also be used.

[0075] Step 203: determine the target item according to the item sorting result and the preset item set, and recommend the target item.

[0076] The preset item set includes items sold by merchants in the live broadcast room. It should be noted that some items in the preset item library may not be sold in the live broadcast room. Therefore, the items in the preset item library are usually more than the items in the preset item set.

[0077] After obtaining the item ranking result, the item ranked first can be determined as the item that the user wants to find in the live broadcast room. In this case, the server matches the item ranked first with the items in the preset item set, and determines the matched item as the target item.

[0078] After determining the target item, the server sends the item information of the target item to the terminal. The terminal receives the item information and displays the target item in the live broadcast room according to the item information. There are many ways to display it, such as displaying the target item in a pop-up window, or displaying the target item at the top of the shopping cart or shopping bag in the live broadcast room. The embodiment of the present application does not limit the display method, and it can be set according to actual conditions.

[0079] In the above embodiment, subject detection is performed on multiple target images in the published content to obtain subject detection results; multiple candidate items are determined from the preset item library based on the subject detection results, and the multiple candidate items are sorted to obtain item sorting results; the target item is determined based on the item sorting results and the preset item set, and the target item is recommended. The embodiment of the present application can automatically detect images in the published content and automatically recommend items based on the detection results, without the need for users to search, compare and browse, which can not only improve search efficiency, but also improve search accuracy and conversion rate.

[0080] In an exemplary embodiment, the subject detection result includes the object subject in the target image and the subject level of the object subject, and the subject level is used to characterize the degree to which the object subject represents the target image. For example, the target image 1 mainly shows the object A, and the object B is located behind the object A; the subject detection result includes the two object subjects, the object A and the object B, and it is determined that the subject level of the object A is the highest, and the subject level of the object B is the second highest. The subject level can be represented by a number, for example, 0 represents the highest subject level, 1 represents the second highest subject level, and so on. In this way, it can be determined that the subject level of the object A is 0, and the subject level of the object B is 1.

[0081] like Figure 3 As shown, in the above embodiment, “determining multiple candidate items from a preset item library according to the subject detection result, and sorting the multiple candidate items to obtain an item sorting result” may include the following steps:

[0082] Step 301, searching for candidate items corresponding to each item body from a preset item library.

[0083] For example, for target image 1, candidate items 1 and 2 corresponding to item A, and candidate items 3 and 4 corresponding to item B are searched from the preset item library. Similarly, the same search is performed on other target images to obtain candidate items corresponding to each object body in each target image.

[0084] Step 302: Determine the detection score of the candidate items corresponding to each item body based on the item body level and the similarity between the item body and each candidate item.

[0085] For each candidate item, the server may determine the corresponding weight of the candidate item based on the subject level of the item subject corresponding to the candidate item; and calculate the detection score of the candidate item according to the similarity between the item subject and the candidate item and the corresponding weight of the candidate item.

[0086] For example, for candidate item 1, candidate item 1 corresponds to item A, and the similarity between candidate item 1 and item A is a%. According to the main level of item A being 0, the weight corresponding to candidate item 1 is determined to be M, and the detection score of candidate item 1 is calculated as F1=a%*M. For candidate item 3, candidate item 3 corresponds to item B, and the similarity between candidate item 3 and item B is c%. According to the main level of item B being 1, the weight corresponding to candidate item 3 is determined to be N, and the detection score of candidate item 3 is calculated as F3=c%*N. By analogy, the detection scores of other candidate items can be calculated.

[0087] It should be noted that the weight corresponding to each subject level can be set according to actual conditions, and the higher the subject level, the higher the corresponding weight. Optionally, M is 0.9 and N is 0.7; or, M is 0.5 and N is 0.3.

[0088] Step 303: sort the items according to the detection scores of the candidate items corresponding to each item body to obtain an item sorting result.

[0089] In one of the sorting methods, for each item body, the candidate items corresponding to the item body are sorted according to the detection score to obtain a first sorting result; then, the first ones of multiple item bodies are spliced ​​in descending order of the subject level; and then the spliced ​​sorting results are deduplicated to obtain the item sorting result.

[0090] For example, for item A, candidate item 1 and candidate item 2 are sorted according to the detection score; for item B, candidate item 3 and candidate item 4 are sorted according to the detection score. After that, the two sortings are spliced ​​in the order of the subject level from high to low; finally, the spliced ​​sorting results are deduplicated to obtain the item sorting result.

[0091] In another sorting method, all candidate items are sorted according to the detection scores to obtain a second sorting result; then, the second sorting result is deduplicated to obtain an item sorting result.

[0092] For example, candidate items 1, 2, 3, and 4 are ranked according to the detection scores; then, duplicate removal is performed to obtain the item ranking result.

[0093] It should be noted that the sorting method is not limited to the above examples and can be set according to actual conditions.

[0094] In the above embodiment, candidate items corresponding to each item body are found from the preset item library; based on the item body level and the similarity between the item body and each candidate item, the detection score of the candidate item corresponding to each item body is determined; the items are sorted according to the detection score of the candidate items corresponding to each item body to obtain the item sorting result. The embodiment of the present application finds candidate items from the preset item library and then sorts the candidate items according to the detection score, which can improve the sorting accuracy, thereby improving the accuracy of the search, and then improving the accuracy of the recommendation, and improving the user experience.

[0095] In an exemplary embodiment, Figure 4 As shown, in the above embodiment, “searching for candidate items corresponding to each item body from the preset item library” may include the following steps:

[0096] Step 401 : For each object body, the object body is matched with each object in a preset object library to obtain the probability of whether each object and the object body are the same object.

[0097] During the matching process, for each object subject, a preset similarity algorithm is used to calculate the similarity between the object subject and each object in the preset object library. The similarity represents the probability of whether the object subject and each object in the preset object library are the same object.

[0098] For example, the similarity between item A and items 1, 2, ... n in the preset item library is calculated using a preset similarity algorithm, and the probability of whether item A and items 1, 2, ... n are the same item is obtained. The similarity between item B and items 1, 2, ... n in the preset item library is calculated using a preset similarity algorithm, and the probability of whether item B and items 1, 2, ... n are the same item is obtained.

[0099] Step 402, according to the probability of whether each item and the item body are the same item and a preset probability threshold, find out the candidate items corresponding to the item body.

[0100] If the probability that the item in the preset item library is the same item as the item subject is greater than or equal to the preset probability threshold, it indicates that the probability that the item is the same item as the item subject is relatively high, and the item is determined as the candidate item corresponding to the item subject. If the probability that the item in the preset item library is the same item as the item subject is less than the preset probability threshold, it indicates that the probability that the item is the same item as the item subject is relatively low, and the item is ignored.

[0101] In the above embodiment, for each item body, the item body is matched with each item in the preset item library to obtain the probability of whether each item and the item body are the same item; according to the probability of whether each item and the item body are the same item and the preset probability threshold, the candidate item corresponding to the item body is found. The embodiment of the present application uses a similarity algorithm for matching, which can obtain more accurate candidate items, provide a basis for subsequent sorting and recommendation, and improve sorting accuracy and recommendation accuracy.

[0102] In an exemplary embodiment, Figure 5 As shown, in the above embodiment, “determining the detection score of the candidate items corresponding to each item body based on the item body level and the similarity between the item body and each candidate item” may include the following steps:

[0103] Step 501, determining the subject score of each candidate object according to the subject level of the object subject.

[0104] For each candidate item, the subject score of the candidate item can be determined according to the subject level of the subject of the item to which the candidate item corresponds. For example, if candidate item 1 corresponds to item A, and the subject level of item A is 0, then the subject score of candidate item 1 is determined to be 0; if candidate item 3 corresponds to item B, and the subject level of item B is 1, then the subject score of candidate item 3 is determined to be 1.

[0105] It should be noted that the sorting priority of the subject score 0 is higher than the sorting priority of the subject score 1, that is, when sorting, the candidate items with the subject score 0 are prioritized to be placed in the front position.

[0106] Step 502: Determine the similarity score of each candidate item according to the similarity between the item body and each candidate item.

[0107] For each candidate item, the similarity score of the candidate item is determined based on the similarity between the candidate item and the main item. For example, if the similarity between candidate item 1 and item A is a%, a is determined as the similarity score of candidate item 1; if the similarity between candidate item 3 and item B is c%, c is determined as the similarity score of candidate item 3.

[0108] Step 503: For each candidate item corresponding to the main body of each item, the detection score of the candidate item is obtained according to the main body score and similarity score of the candidate item.

[0109] For each candidate item, a connection symbol is used to connect the subject score and the similarity score of the candidate item to obtain the detection score of the candidate item.

[0110] For example, if the connection symbol is an underscore, the main score of candidate item 1 is 0, and the similarity score is a, then the detection score of candidate item 1 is 0_a; the main score of candidate item 3 is 1, and the similarity score is c, then the detection score of candidate item 3 is 1_c.

[0111] It should be noted that the method for determining the detection score is not limited to the connection method in the above example, and in practical applications, other methods may also be used.

[0112] In the above embodiment, the subject score of each candidate item is determined according to the subject level of the object subject; the similarity score of each candidate item is determined according to the similarity between the object subject and each candidate item; for the candidate items corresponding to each object subject, the detection score of the candidate item is obtained according to the subject score and similarity score of the candidate item. When scoring the candidate items, the embodiment of the present application takes into account the influence of each target image on the selection of the object subject, avoiding the problem of subject selection bias caused by a target image having a much higher similarity score than other images due to its high image quality.

[0113] In an exemplary embodiment, Figure 6 As shown, in the above embodiment, “determining the similarity score of each candidate item according to the similarity between the item body and each candidate item” may include the following steps:

[0114] Step 5021, sorting multiple candidate items according to the similarity between the main object and each candidate item to obtain a similarity sorting result.

[0115] For each item subject, according to the similarity between the item subject and each candidate item, multiple candidate items are sorted in descending order of similarity to obtain a similarity sorting result corresponding to each item subject.

[0116] For example, for item A, according to the similarity between item A and candidate item 1, and the similarity between item A and candidate item 2, candidate item 1 and candidate item 2 are sorted to obtain the similarity sorting result corresponding to item A. For item B, according to the similarity between item B and candidate item 3, and the similarity between item B and candidate item 4, candidate item 3 and candidate item 4 are sorted to obtain the similarity sorting result corresponding to item B.

[0117] Step 5022: Determine the similarity score of each candidate item based on the similarity ranking result.

[0118] For each object, according to the similarity sorting result, each candidate object is assigned a value in descending order to obtain the similarity score of each candidate object.

[0119] For example, 0 is the highest similarity score, and 1 is the second highest similarity score; for item A, the similarity ranking results are candidate item 1 and candidate item 2, then candidate item 1 is assigned a similarity score of 0, and candidate item 2 is assigned a similarity score of 1. For item B, the similarity ranking results are candidate item 3 and candidate item 4, then candidate item 3 is assigned a similarity score of 0, and candidate item 4 is assigned a similarity score of 1. And so on, the similarity scores of the candidate items corresponding to each item subject are obtained.

[0120] In the above embodiment, multiple candidate items are sorted according to the similarity between the main body of the item and each candidate item to obtain a similarity sorting result; and the similarity score of each candidate item is determined according to the similarity sorting result. The embodiment of the present application determines the similarity score according to the similarity sorting, so that the similarity score is normalized, which is convenient for subsequent sorting of the candidate items and can improve the sorting efficiency.

[0121] In an exemplary embodiment, among the multiple target images included in the published content, usually, the first target image is the image in the published content that can better reflect the publishing intention, so the first target image and other target images are sorted separately.

[0122] like Figure 7 As shown, in the above embodiment, “ranking the items according to the detection scores of the candidate items corresponding to each item body to obtain the item ranking result” may include the following steps:

[0123] Step 601 , for the first target image, the first candidate items are sorted according to the detection scores of the first candidate items corresponding to the object bodies to obtain the first image sorting result.

[0124] For the first target image, sort the first candidate items corresponding to each object body in descending order according to the detection score to obtain the first image sorting result. For example, for candidate items 1 and 2 corresponding to object A in target image 1, and candidate items 3 and 4 corresponding to object B, sort them in descending order according to the detection score. Figure 8 As shown, item A in the first image is marked with detection frame 0, candidate item 1 corresponding to item A is marked with first image product 0_0, and candidate item 2 is marked with first image product 0_1; item B in the first image is marked with detection frame 1, candidate item 3 corresponding to item B is marked with first image product 1_0, and candidate item 4 is marked with first image product 1_1. Similarly, item C in image 2 is marked with detection frame 0, item D is marked with detection frame 1, and item E is marked with detection frame 2; item F in image 3 is marked with detection frame 0, and item G is marked with detection frame 1.

[0125] It can be understood that, since the subject level of item A is higher than that of item B, the first candidate item corresponding to item A is ranked higher than the first candidate item corresponding to item B.

[0126] Step 602 , for other target images, the second candidate objects are sorted according to the detection scores of the second candidate objects corresponding to the object bodies to obtain other image sorting results.

[0127] For other target images, the second candidate items corresponding to each object body are sorted in descending order according to the detection score to obtain the sorting results of other images. For example, the candidate items corresponding to object C, object D, and object E in target image 2, and the candidate items corresponding to object G and object H in target image 3 are sorted in descending order according to the detection score. Figure 8 As shown, multiple candidate items with a subject score of 0 are sorted first, and then multiple candidate items with a subject score of 1 are sorted, and so on.

[0128] It should be noted that when the subject scores are the same, they are sorted from high to low according to the similarity scores; when the subject scores and similarity scores are the same, they are sorted from high to low according to the similarity.

[0129] Step 603, sorting the first candidate items and the second candidate items based on the first image sorting result and the other image sorting results to obtain an item sorting result.

[0130] After obtaining the ranking result of the first image and the ranking results of other images, the second candidate items in the other target images may be ranked behind the first candidate item in the first target image; thereafter, deduplication processing is performed to obtain the item ranking result.

[0131] In the above embodiment, for the first target image, the first candidate items are sorted according to the detection scores of the first candidate items corresponding to each object body, and the first image sorting result is obtained; for other target images, the second candidate items are sorted according to the detection scores of the second candidate items corresponding to each object body, and the other image sorting results are obtained; based on the first image sorting result and the other image sorting results, the first candidate items and the second candidate items are sorted to obtain the item sorting result. The embodiment of the present application improves the sorting priority of the first image, which can simulate the user's image sorting thinking, improve the sorting effect, and further improve the recommendation effect.

[0132] In an exemplary embodiment, Fig. 9 As shown, in the above embodiment, “ranking the first candidate item and the second candidate item based on the first image ranking result and the other image ranking results to obtain an item ranking result” may include the following steps:

[0133] Step 6031, for the candidate items of each subject score, according to the ranking of each candidate item in the first picture ranking result and the other picture ranking results, the first candidate item and the second candidate item are selected in turn for ranking to obtain the initial ranking result of each subject score.

[0134] Exemplarily, for candidate items with a subject score of 0, the first candidate item with a ranking of 1 in the first image ranking result is ranked first, the second candidate item with a ranking of 1 in the other image ranking results is ranked second, the first candidate item with a ranking of 2 in the first image ranking result is ranked third, and the second candidate item with a ranking of 2 in the other image ranking results is ranked fourth. This is done by analogy to obtain an initial ranking result with a subject score of 0. Then, for candidate items with a subject score of 1, the ranking is performed in the above manner to obtain an initial ranking result with a subject score of 1.

[0135] It should be noted that, for each subject score, after the first candidate items in the first target image are ranked, the second candidate items corresponding to the subject score in other target images are assigned ranks according to the original order.

[0136] Step 6032, according to the size of the subject score, the initial sorting results of multiple subject scores are spliced ​​to obtain the item sorting result.

[0137] Exemplarily, the initial sorting result of the subject score 1 is spliced ​​behind the initial sorting result of the subject score 0, and the initial sorting result of the subject score 2 is spliced ​​behind the initial sorting result of the subject score 1. And so on, the item sorting result is obtained.

[0138] In the above embodiment, for the candidate items of each subject score, according to the ranking position of each candidate item in the ranking result of the first picture and the ranking results of other pictures, the first candidate item and the second candidate item are selected in turn for ranking to obtain the initial ranking result of each subject score; according to the size of the subject score, the initial ranking results of multiple subject scores are spliced ​​to obtain the item ranking result. The embodiment of the present application adopts group sorting for the first picture and other pictures, which can take into account the primary and secondary, avoid the problem of a large ranking tilt when the candidate item selected according to the first picture is not a representative item in the release, reduce the risk of sorting errors, improve sorting accuracy and recommendation accuracy, and thus improve the recommendation effect and enhance the user experience.

[0139] In an exemplary embodiment, Fig.10 As shown, in the above embodiment, “performing subject detection on multiple target images in the published content respectively to obtain subject detection results” may include the following steps:

[0140] Step 701 : for each target image, input the target image into a pre-established subject detection model to obtain the subject position of each object subject in the target image.

[0141] A plurality of training images are obtained in advance, and the position annotation of each object body in each training image is obtained. Then, model training is performed according to the training images and the position annotation to obtain a subject detection model.

[0142] For each target image in the published content, the target image is input into the subject detection model, the subject detection model performs subject detection, and outputs the subject position corresponding to each object subject in the target image. Optionally, the subject position can be identified by a detection box.

[0143] Step 702, determining the subject level of each object subject according to the subject position.

[0144] In some of these methods, the distance relationship between the object body and the image acquisition device is determined according to the subject position, and a level is assigned to each object body in order from near to far to obtain the subject level of each object body.

[0145] For example, according to the subject position, it is determined that object A is closer to the image acquisition device and object B is farther from the image acquisition device. Then, in order from near to far, the subject level of object A is 0 and the subject level of object B is 1.

[0146] In some other methods, the area ratio of the object body in the target image can be determined according to the main body position, and the level of each object body is assigned in descending order according to the area ratio to obtain the main body level of each object body.

[0147] For example, according to the subject position, it is determined that the area of ​​object A in the target image accounts for a larger proportion, and the area of ​​object B in the target image accounts for a smaller proportion. Then, in descending order of proportion, the subject level of object A can be obtained as 0, and the subject level of object B can be obtained as 1.

[0148] It should be noted that the method of determining the subject level according to the subject position is not limited to the above example, and in practical applications, it can be set according to actual conditions.

[0149] In the above embodiment, for each target image, the target image is input into a pre-established subject detection model to obtain the subject position of each object subject in the target image; and the subject level of each object subject is determined according to the subject position. The object subject extraction in the embodiment of the present application can take into account the influence of each target image on the object subject selection, avoid the subject selection bias caused by a target image having a much higher similarity score than other target images due to its high image quality, and improve the sorting accuracy.

[0150] In an exemplary embodiment, Fig.11 As shown, in the above embodiment, “determining the target item according to the item sorting result and the preset item set” may include the following steps:

[0151] Step 801: determine a search item from a plurality of candidate items according to the item ranking result.

[0152] According to the item sorting results, the candidate item ranked first is determined as the search item.

[0153] Step 802: searching for matching items in a preset item set according to the search item.

[0154] For each item in the preset item set, the probability that the search item is the same as the item is calculated; if the calculated probability is greater than or equal to the preset probability threshold, it indicates that the probability that the search item is the same as the item is high, and the item is determined as a matching item. If the calculated probability is less than the preset probability threshold, it indicates that the probability that the search item is the same as the item is low, and the item is ignored.

[0155] Step 803: If a matching object is found, the matching object is determined as the target object.

[0156] If a matching item is found from the preset item set, the matching item is used as the target item to be recommended.

[0157] Step 804: if no matching item is found, the search item is re-determined from the plurality of candidate items according to the item ranking result, and a step of searching for a matching item in a preset item set according to the search item is performed.

[0158] If no matching item is found in the preset item set, the second candidate item is determined as the search item according to the item sorting result. Then, a matching item is searched in the preset item set according to the search item. This process is repeated until the target item is determined.

[0159] In the above embodiment, the search item is determined from multiple candidate items according to the item ranking result; a matching item is searched in a preset item set according to the search item; if a matching item is found, the matching item is determined as the target item; if a matching item is not found, the search item is re-determined from multiple candidate items according to the item ranking result, and the step of searching for a matching item in the preset item set according to the search item is performed. The embodiment of the present application determines the target item to be recommended according to the item ranking result, which can improve the recommendation accuracy, thereby improving the recommendation effect, and improving the conversion rate.

[0160] In an exemplary embodiment, a method for recommending items is provided, wherein the method is applied to Figure 1 Taking the server in as an example, the following steps may be included:

[0161] Step 1: For each target image in the published content, the target image is input into a pre-established subject detection model to obtain the subject position of each object subject in the target image.

[0162] Step 2: Determine the subject level of each object subject according to the subject position.

[0163] Step 3: Determine candidate items corresponding to each item body from the preset item library.

[0164] Among them, for each item body, the item body is matched with each item in the preset item library to obtain the probability of whether each item and the item body are the same item; based on the probability of whether each item and the item body are the same item and the preset probability threshold, the candidate item corresponding to the item body is found.

[0165] Step 4: Determine the subject score of each candidate item according to the subject level of the item subject.

[0166] Step 5: Determine the similarity score of each candidate item based on the similarity between the item and each candidate item.

[0167] Among them, according to the similarity between the main body of the object and each candidate object, multiple candidate objects are sorted to obtain a similarity sorting result; according to the similarity sorting result, the similarity score of each candidate object is determined.

[0168] Step 6: For the candidate items corresponding to each object subject, the detection score of the candidate items is obtained according to the subject score and similarity score of the candidate items.

[0169] Step 7: for the first target image, the first candidate items are sorted according to the detection scores of the first candidate items corresponding to the object bodies to obtain the first image sorting result.

[0170] Step 8, for other target images, sort the second candidate objects according to the detection scores of the second candidate objects corresponding to the object bodies, and obtain sorting results of other images;

[0171] Step 9, sorting the first candidate items and the second candidate items based on the sorting result of the first image and the sorting results of the other images to obtain an item sorting result.

[0172] Among them, for the candidate items of each main score, according to the ranking position of each candidate item in the first picture sorting result and the other picture sorting results, the first candidate item and the second candidate item are selected in turn for sorting to obtain the initial sorting result of each main score; according to the size of the main score, the initial sorting results of multiple main scores are spliced ​​to obtain the item sorting result.

[0173] Step 10: Determine the search item from multiple candidate items according to the item ranking result.

[0174] Step 11: Search for matching items in a preset item set according to the search item.

[0175] Step 12: If a matching item is found, the matching item is determined as a target item and the target item is recommended.

[0176] Step 13: If no matching item is found, the search item is re-determined from the multiple candidate items according to the item sorting result, and a step of searching for a matching item in a preset item set according to the search item is performed.

[0177] The embodiment of the present application is aimed at the scenario of multi-image detection, taking into account factors such as image position, image quality, and single-image subject score. It mines the semantics of multiple images while extracting multi-target subjects, and designs a priority queue sorting method to ensure the accuracy of sorting multiple image subjects, which can improve the recommendation effect.

[0178] In an exemplary embodiment, taking the commodity content published by a merchant in an application as "notes" as an example, the interaction between the terminal and the server may include the following steps:

[0179] 1) The terminal displays the notes published by the merchant, which contain a link to the live broadcast room; the terminal obtains the user's trigger operation on the link, such as a click operation, and sends a recommendation request to the server. The recommendation request can carry the note identifier, the link to the live broadcast room, etc.

[0180] 2) The server receives the recommendation request and determines the note content according to the note identifier carried in the recommendation request, that is, determines multiple target images in the note.

[0181] 3) For target image 1, the server inputs target image 1 into a pre-established subject detection model, and obtains that target image 1 mainly shows object A, and object B is located behind object A.

[0182] 4) The server determines that item A has the highest level of the subject according to the subject position, and item B has the second highest level of the subject; and finds candidate items 1 and 2 corresponding to item A, and candidate items 3 and 4 corresponding to item B from the merchant's available item library.

[0183] 5) The server determines the subject scores of candidate items 1, 2, 3, and 4 according to the subject levels of item A and item B; determines the similarity scores of candidate items 1 and 2 according to the similarity between item A and candidate items 1 and 2; and determines the similarity scores of candidate items 3 and 4 according to the similarity between item B and candidate items 3 and 4.

[0184] 6) For candidate items 1 and 2 corresponding to item subject A, and candidate items 3 and 4 corresponding to item B, the detection score of candidate item 1 is obtained according to the subject score and similarity score of candidate item 1. Similarly, the detection scores of candidate items 2, 3, and 4 are obtained.

[0185] 7) The server processes other target images in sequence according to steps 3-6.

[0186] 8) For candidate items 1 and 2 corresponding to item A in target image 1, and candidate items 3 and 4 corresponding to item B, sort them in descending order according to the detection scores to obtain the first image sorting result.

[0187] 9) The server obtains the sorting results of other graphs according to step 8.

[0188] 10) Sort the items based on the sorting results of the first image and other images to obtain the sorting results of all items.

[0189] 11) The server determines the search item based on the item sorting results and selects the candidate item ranked first.

[0190] 12) The server searches for matching items in the shopping cart of the live broadcast room based on the searched items.

[0191] 13) If a matching item is found, the server determines the matching item as the target item and sends the item information of the target item and the screen of the live broadcast room to the terminal.

[0192] 14) The terminal receives the item information and the live broadcast room screen, displays the live broadcast room screen, and displays a pop-up window of the target item based on the item information.

[0193] 15) The terminal receives the trigger operation input by the user for the shopping cart identification control and sends a display request to the server.

[0194] 16) The server receives the display request and sends the shopping cart page to the terminal.

[0195] 17) The terminal displays the shopping cart page and displays the target item at the top of the shopping cart page.

[0196] In the embodiment of the present application, after a user enters a merchant's live broadcast room based on a note, the terminal can directly recommend items in the note to the user through a pop-up window and a top display on the shopping cart page, without the user having to search or compare, which can greatly improve the user's search efficiency. Moreover, after the notes and the live broadcast room are linked, items can be accurately recommended to the user, thereby improving the conversion rate of the notes and increasing the merchant's sales.

[0197] It should be understood that, although the various steps in the flowcharts involved in the above-mentioned embodiments are displayed in sequence according to the indication of the arrows, these steps are not necessarily executed in sequence according to the order indicated by the arrows. Unless there is a clear explanation in this article, the execution of these steps does not have a strict order restriction, and these steps can be executed in other orders. Moreover, at least a part of the steps in the flowcharts involved in the above-mentioned embodiments can include multiple steps or multiple stages, and these steps or stages are not necessarily executed at the same time, but can be executed at different times, and the execution order of these steps or stages is not necessarily to be carried out in sequence, but can be executed in turn or alternately with other steps or at least a part of the steps or stages in other steps.

[0198] Based on the same inventive concept, the embodiment of the present application also provides an item recommendation device for implementing the item recommendation method involved above. The implementation scheme for solving the problem provided by the device is similar to the implementation scheme recorded in the above method, so the specific limitations in one or more item recommendation device embodiments provided below can refer to the limitations on the item recommendation method above, and will not be repeated here.

[0199] In an exemplary embodiment, Fig.12 As shown, an item recommendation device is provided, comprising:

[0200] A subject detection module 901 is used to perform subject detection on multiple target images in the published content to obtain subject detection results, where the target images include at least one object;

[0201] A sorting module 902 is used to determine a plurality of candidate items from a preset item library according to the subject detection result, and sort the plurality of candidate items to obtain an item sorting result;

[0202] The recommendation module 903 is used to determine the target item according to the item sorting result and the preset item set, and recommend the target item.

[0203] In some embodiments, the subject detection result includes the object subject in the target image and the subject level of the object subject, the subject level is used to characterize the degree of representativeness of the object subject to the target image, and the sorting module 902 is specifically used to find out the candidate objects corresponding to each object subject from the preset object library; based on the subject level of the object subject and the similarity between the object subject and each candidate object, determine the detection score of the candidate objects corresponding to each object subject; sort the objects according to the detection score of the candidate objects corresponding to each object subject, and obtain the object sorting result.

[0204] In some embodiments, the sorting module 902 is specifically used to sort multiple candidate items according to the similarity between the main object and each candidate item to obtain a similarity sorting result; and determine the similarity score of each candidate item according to the similarity sorting result.

[0205] In some embodiments, the sorting module 902 is specifically used to sort the first candidate items for the first target image according to the detection scores of the first candidate items corresponding to each item body, and obtain the first image sorting result; for other target images, sort the second candidate items according to the detection scores of the second candidate items corresponding to each item body, and obtain the other image sorting results; sort the first candidate items and the second candidate items based on the first image sorting result and the other image sorting results to obtain the item sorting result.

[0206] In some embodiments, the sorting module 902 is specifically used to select the first candidate item and the second candidate item in turn for sorting the candidate items of each subject score according to the ranking positions of each candidate item in the first picture sorting result and the other picture sorting results, so as to obtain the initial sorting result of each subject score; according to the size of the subject score, the initial sorting results of multiple subject scores are spliced ​​to obtain the item sorting result.

[0207] In some embodiments, the sorting module 902 is specifically used to match the item body with each item in a preset item library for each item body, and obtain the probability of whether each item and the item body are the same item; according to the probability of whether each item and the item body are the same item and the preset probability threshold, find the candidate item corresponding to the item body.

[0208] In some embodiments, the subject detection module 901 is specifically used to input each target image into a pre-established subject detection model to obtain the subject position of each object subject in the target image; and determine the subject level of each object subject according to the subject position.

[0209] In some embodiments, the recommendation module 903 is specifically used to determine the search item from multiple candidate items according to the item ranking result; search for matching items in a preset item set according to the search item; if a matching item is found, determine the matching item as the target item.

[0210] In some embodiments, the recommendation module 903 is further configured to re-determine the search item from multiple candidate items according to the item ranking result if no matching item is found, and execute the step of searching for a matching item in a preset item set according to the search item.

[0211] Each module in the above-mentioned item recommendation device can be implemented in whole or in part by software, hardware or a combination thereof. Each module can be embedded in or independent of a processor in a computer device in the form of hardware, or can be stored in a memory in a computer device in the form of software, so that the processor can call and execute the corresponding operations of each module.

[0212] In an exemplary embodiment, a computer device is provided. The computer device may be a server, and its internal structure diagram may be as shown in FIG. Fig.13 As shown. The computer device includes a processor, a memory, an input / output interface (Input / Output, referred to as I / O) and a communication interface. The processor, the memory and the input / output interface are connected through a system bus, and the communication interface is connected to the system bus through the input / output interface. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The database of the computer device is used to store XX data. The input / output interface of the computer device is used to exchange information between the processor and an external device. The communication interface of the computer device is used to communicate with an external terminal through a network connection. When the computer program is executed by the processor, an item recommendation method is implemented.

[0213] Those skilled in the art will understand that Fig.13 The structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components.

[0214] In an exemplary embodiment, a non-transitory computer-readable storage medium including instructions is also provided, such as a memory including instructions, and the instructions can be executed by a processor of an electronic device to perform the above method. For example, the non-transitory computer-readable storage medium can be a ROM, a random access memory (RAM), a CD-ROM, a magnetic tape, a floppy disk, an optical data storage device, etc.

[0215] In an exemplary embodiment, a computer program product is also provided, and when the computer program is executed by a processor, the above method can be implemented. The computer program product includes one or more computer instructions. When these computer instructions are loaded and executed on a computer, part or all of the above method can be implemented in whole or in part according to the process or function described in the embodiment of the present application.

[0216] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of relevant data must comply with relevant regulations.

[0217] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to the memory, database or other medium used in the embodiments provided in this application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetoresistive random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. As an illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM). The database involved in each embodiment provided in this application may include at least one of a relational database and a non-relational database. Non-relational databases may include distributed databases based on blockchains, etc., but are not limited to this. The processor involved in each embodiment provided in this application may be a general-purpose processor, a central processing unit, a graphics processor, a digital signal processor, a programmable logic device, a data processing logic device based on quantum computing, etc., but are not limited to this.

[0218] The technical features of the above embodiments may be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0219] The above-described embodiments only express several implementation methods of the present application, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the present application. It should be pointed out that, for a person of ordinary skill in the art, several variations and improvements can be made without departing from the concept of the present application, and these all belong to the protection scope of the present application. Therefore, the protection scope of the present application shall be subject to the attached claims.

Claims

1. An item recommendation method, characterized in that: The method comprises: Performing subject detection on multiple target images in the published content respectively to obtain subject detection results, wherein the target images include at least one object; Determining a plurality of candidate items from a preset item library according to the subject detection result, and sorting the plurality of candidate items to obtain an item sorting result; A target item is determined according to the item sorting result and a preset item set, and the target item is recommended.

2. The method according to claim 1, characterized in that The subject detection result includes an object subject in the target image and a subject level of the object subject, wherein the subject level is used to characterize the degree to which the object subject represents the target image. The determining of a plurality of candidate objects from a preset object library according to the subject detection result and sorting the plurality of candidate objects to obtain an object sorting result includes: Finding candidate items corresponding to each of the item bodies from the preset item library; Determine the detection score of the candidate items corresponding to each of the item subjects based on the subject level of the item subject and the similarity between the item subject and each of the candidate items; The items are sorted according to the detection scores of the candidate items corresponding to the item bodies to obtain the item sorting result.

3. The method according to claim 2, characterized in that The determining, based on the subject level of the object subject and the similarity between the object subject and each of the candidate objects, the detection score of the candidate objects corresponding to each of the object subjects comprises: Determining the subject score of each of the candidate items according to the subject level of the item subject; Determining a similarity score of each candidate item according to the similarity between the item subject and each candidate item; For each candidate item corresponding to the item subject, a detection score of the candidate item is obtained according to the subject score and the similarity score of the candidate item.

4. The method according to claim 2, characterized in that: The step of sorting the items according to the detection scores of the candidate items corresponding to the item bodies to obtain the item sorting results includes: For the first target image, the first candidate objects are sorted according to the detection scores of the first candidate objects corresponding to the object bodies to obtain a first image sorting result; For other target images, the second candidate objects are sorted according to the detection scores of the second candidate objects corresponding to the object bodies to obtain sorting results of other images; The first candidate items and the second candidate items are sorted based on the first image sorting result and the other image sorting results to obtain the item sorting result.

5. The method according to claim 4, characterized in that The step of sorting the first candidate items and the second candidate items based on the first image sorting result and the other image sorting results to obtain the item sorting result includes: For the candidate items of each subject score, according to the ranking of each candidate item in the first picture ranking result and the ranking of each candidate item in the other picture ranking results, the first candidate item and the second candidate item are selected in turn for ranking, so as to obtain the initial ranking result of each subject score; According to the size of the subject score, multiple initial sorting results of the subject scores are spliced ​​to obtain the item sorting result.

6. The method according to claim 2, characterized in that The step of searching for candidate items corresponding to each of the item bodies from the preset item library includes: For each of the item bodies, the item body is matched with each item in the preset item library to obtain a probability of whether each item and the item body are the same item; According to the probability of whether each of the items and the item body are the same item and a preset probability threshold, a candidate item corresponding to the item body is found.

7. The method according to claim 2, characterized in that The subject detection is performed on multiple target images in the published content to obtain subject detection results, including: For each of the target images, input the target image into a pre-established subject detection model to obtain the subject position of each object subject in the target image; The subject level of each of the object subjects is determined according to the subject position.

8. The method according to any one of claims 1 to 7, characterized in that: The determining the target item according to the item sorting result and the preset item set includes: Determine a search item from the plurality of candidate items according to the item ranking result; Searching for a matching item in the preset item set according to the search item; If the matching object is found, the matching object is determined as the target object.

9. The method according to claim 8, characterized in that The method further comprises: If the matching item is not found, the search item is re-determined from the plurality of candidate items according to the item ranking result, and the step of searching for a matching item in the preset item set according to the search item is performed.

10. An item recommendation device, characterized in that: The device comprises: A subject detection module, used to perform subject detection on a plurality of target images in the published content respectively to obtain subject detection results, wherein the target images include at least one object; A sorting module, used to determine a plurality of candidate items from a preset item library according to the subject detection result, and sort the plurality of candidate items to obtain an item sorting result; The recommendation module is used to determine the target item according to the item sorting result and the preset item set, and recommend the target item.

11. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 9 are implemented.

12. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 9 are implemented.

13. A computer program product, comprising a computer program, characterized in that When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 9 are implemented.

Images