Object recommendation method, device and system, computing equipment and storage medium

Abstract

The embodiment of the invention provides an object recommendation method, device and system, computing equipment and a storage medium, and the method comprises the steps: for any to-be-recommended target object, determining candidate users matched with the target object and the number of recommended users corresponding to the target object; on the basis of the matching relationship between the target object and the candidate user, constructing global matching information of a to-be-recommended object and the user, and performing optimization analysis on the global matching information by taking the condition that the target user number allocated for the target object reaches the recommended user number as a constraint condition; and determining a target recommendation object for each user. The method can improve the object recommendation effect.

Description

Technical Field

[0001] The embodiments of the present specification relate to the field of computer technology, and in particular to an object recommendation method, an object recommendation apparatus, an object recommendation system, a computing device, a computer-readable storage medium, and a computer program product. Background Art

[0002] With the development of computer technology and Internet technology, the Internet has become an important channel for product promotion.

[0003] Usually, when recommending products on an online platform, the preferences of each user are analyzed from the user's perspective, and then among all the advertisements that can be recommended, product advertisements that match the user's preferences are determined, and the product is recommended to the user based on the product advertisements.

[0004] In this way, product recommendations are made based on user needs, which makes it difficult to meet the promotional goals required for the products, and the effect of product recommendations still needs to be improved. Summary of the invention

[0005] In view of this, an embodiment of this specification provides an object recommendation method, which can improve the object recommendation effect. One or more embodiments of this specification also relate to an object recommendation device, an object recommendation system, a computing device, a computer-readable storage medium, and a computer program product.

[0006] According to a first aspect of an embodiment of this specification, there is provided an object recommendation method, including: For any target object to be recommended, determine the candidate users matching the target object and the number of recommended users corresponding to the target object; Based on the matching relationship between the target object and the candidate user, global matching information of the object to be recommended and the user is constructed, and with the number of target users allocated to the target object reaching the number of recommended users as a constraint condition, the global matching information is optimized and analyzed to determine the target recommended object for each user.

[0007] According to a second aspect of an embodiment of this specification, there is provided an object recommendation device, including: A determination module, for determining, for any target object to be recommended, candidate users matching the target object and the number of recommended users corresponding to the target object; The analysis module is used to construct global matching information between the object to be recommended and the user based on the matching relationship between the target object and the candidate user, and optimize the global matching information with the number of target users allocated to the target object reaching the number of recommended users as a constraint condition to determine the target recommended object for each user.

[0008] According to a third aspect of the embodiments of this specification, there is provided an object recommendation system, including: a server device and a plurality of terminal devices, wherein the plurality of terminal devices correspond to a plurality of users respectively; The server device is used to execute the above method, determine a target recommendation object for each user corresponding to the terminal device, and send information of the target recommendation object to the terminal device; The terminal device is used to display the received information of the target recommended object.

[0009] According to a fourth aspect of the embodiments of this specification, there is provided a computing device, including: a memory and a processor; The memory is used to store computer programs / instructions, and the processor is used to execute the computer programs / instructions. When the computer programs / instructions are executed by the processor, the steps in the above-mentioned object recommendation method are implemented.

[0010] According to a fifth aspect of the embodiments of this specification, a computer-readable storage medium is provided, storing a computer program / instruction, which, when executed by a processor, implements the steps in the above-mentioned object recommendation method.

[0011] According to a sixth aspect of the embodiments of this specification, a computer program product is provided, including a computer program / instruction, which, when executed in a processor, implements the steps in the above-mentioned object recommendation method.

[0012] In the embodiments of this specification, from the perspective of the target object to be recommended, the candidate users that match the target object and the corresponding number of recommended users can be determined. Based on the matching relationship between the target object and the candidate users, the global matching information between the object and the user can be determined, and based on the global matching information, an object recommendation method that makes the user and the object more interested as a whole can be obtained. With the number of recommended users corresponding to the target object as a constraint condition, the global matching information is optimized and analyzed, and the target recommended object for each user is finally determined. In this way, it can be ensured that on the basis of meeting the recommendation needs of the target object, an object recommendation method with better overall matching effect is obtained, thereby improving the recommendation effect of each object. BRIEF DESCRIPTION OF THE DRAWINGS

[0013] Figure 1 is a structural diagram of an object recommendation system provided in an embodiment of this specification; Figure 2 is a flow chart of an object recommendation method provided by an embodiment of this specification; Figure 3 It is a structural schematic diagram of a double-tower model provided in an embodiment of this specification; Figure 4 It is a schematic diagram of an anti-funneling crowd strategy provided in an embodiment of this specification; Figure 5 It is a schematic diagram of a matching method between a user and an object to be recommended provided in an embodiment of this specification; Figure 6 This is a flowchart of an overall process of matching a user with an object to be recommended provided by an embodiment of this specification; Figure 7 is a timing diagram of an object recommendation method provided in an embodiment of this specification; Figure 8 is a structural diagram of an object recommendation device provided in an embodiment of this specification; Fig. 9 It is a structural block diagram of a computing device provided in an embodiment of this specification. DETAILED DESCRIPTION

[0014] Many specific details are described in the following description to facilitate a full understanding of this specification. However, this specification can be implemented in many other ways than those described herein, and those skilled in the art can make similar generalizations without violating the connotation of this specification, so this specification is not limited to the specific implementation disclosed below.

[0015] The terms used in one or more embodiments of this specification are only for the purpose of describing specific embodiments, and are not intended to limit one or more embodiments of this specification. The singular forms of "a", "said" and "the" used in one or more embodiments of this specification and the appended claims are also intended to include plural forms, unless the context clearly indicates other meanings. It should also be understood that the term "and / or" used in one or more embodiments of this specification refers to and includes any or all possible combinations of one or more associated listed items. It should be understood that although the terms first, second, etc. may be used to describe various information in one or more embodiments of this specification, these information should not be limited to these terms. These terms are only used to distinguish the same type of information from each other. For example, without departing from the scope of one or more embodiments of this specification, the first may also be referred to as the second, and similarly, the second may also be referred to as the first. Depending on the context, the word "if" as used herein can be interpreted as "at the time of..." or "when..." or "based on determination".

[0016] In addition, 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 one or more embodiments of this specification 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 the relevant laws, regulations and standards of the relevant countries and regions, and provide corresponding operation entrances for users to choose to authorize or refuse.

[0017] The following is an explanation of the meaning of some of the terms mentioned in this specification. For example, the term "in response to" used in this specification indicates a state in which a corresponding event occurs or a condition is satisfied. The timing of executing a subsequent action executed in response to the event or condition is not necessarily strongly correlated with the time when the event occurs or the condition is satisfied. For example, in some cases, a subsequent action may be executed immediately when an event occurs or a condition is satisfied; in other cases, a subsequent action may be executed some time after the event occurs or the condition is satisfied.

[0018] The term "Standard Product Unit (SPU)" used in this specification refers to the smallest unit of commodity information aggregation, which is a set of reusable and easily searchable standardized information that describes the overall characteristics and attributes of a product, such as brand, model, and function. SPU is usually used for product classification and display to facilitate consumers to browse and compare. On e-commerce platforms, SPU is often used as a display unit for commodities. Consumers can find the commodities they need based on SPU and compare commodities of different brands or models.

[0019] The term "published content" used in this specification refers to content information published on a network platform, which may include text, images, audio, video and other information, and can be displayed in the form of notes, articles, video files or live broadcasts. The specific content information contained in the published content applicable to different scenarios can be adjusted accordingly, and the content information in the published content can be determined by the publisher. For example, the published content on the network platform can be used to promote SPU.

[0020] At present, content is recommended to users on various network platforms, and the requirements for content recommendation effects are getting higher and higher. On network platforms, advertisers can promote certain products (such as objects to be recommended), such as advertising in units of SPUs, and pushing the advertising content corresponding to the SPUs to users. The products in this specification can be virtual products or physical products. Usually, the network platform will analyze the preferences of each user, and then determine the SPU that meets the user's preferences among the various SPUs to be recommended, and then recommend the SPU to the user. In this recommendation method, the focus is on determining the SPU that should be recommended to a single user from the perspective of the user, which tends to be local single-point optimization, and does not match the SPU and the user from a global perspective. It may be difficult to achieve the promotion effect for the SPU needs. Therefore, the object recommendation effect still needs to be improved.

[0021] In addition, in the usual object recommendation process, ads are first placed on a large scale to various users for SPUs, and then the user range of ads is gradually narrowed based on user feedback, and SPUs are gradually and accurately recommended to more interested users to achieve conversion. In this recommendation method, the matching efficiency between SPUs and users is low, and it requires a high cost to achieve a certain promotion effect on SPUs.

[0022] The embodiments of this specification provide an object recommendation method, which can improve the object recommendation effect and ensure that the required promotion effect is achieved more quickly and effectively for the recommended object. This specification also relates to an object recommendation device, an object recommendation system, a computing device, a computer-readable storage medium, and a computer program product, which are described in detail one by one in the following embodiments.

[0023] The object recommendation method provided in the embodiments of this specification can be applied to an object recommendation system. Figure 1 Schematic diagram of the structure of an object recommendation system provided in an embodiment of this specification. Figure 1 As shown, the object recommendation system may include a server device 101 and multiple terminal devices 102. The server device 101 and the terminal devices 102 may perform two-way communication via a local area network connection, a wide area network connection, an Internet connection or other types of data networks.

[0024] The server device 101 may correspond to a content application platform, which corresponds to the target application. The server device 101 may be a background server of the target application, which is used to provide support for the operation of the target application. The target application may be installed in the terminal device 102, and the communication connection with the server device 101 is realized by running the target application, so as to use the function of the content application platform. The content application platform may be an application platform that integrates multiple contents, and the content application platform provides users with rich and diverse multimedia content, including but not limited to graphic information, live broadcast, video on demand, audio, graphic information, social interaction, etc. These multimedia contents may be published by users, and may also be referred to as published content. The content application platform may be a note publishing and browsing platform, and users may publish notes in the platform, that is, the published content in the content application platform may be notes, and the notes may include one or more of text information, image information and video information. Users may view notes published by other users, or may interact with other users based on the notes. For example, different users may interact with each other by liking, commenting or sharing notes. The notes viewed by users may be recommended by the content application platform or searched by users.

[0025] The multiple terminal devices 102 in the object recommendation system correspond to multiple users respectively. For example, each terminal device 102 corresponds to one user, or there may be a situation where multiple terminal devices 102 correspond to one user. The users in the embodiments of this specification are represented by user accounts, and one user account represents one user. The terminal device 102 can log in to the content application platform based on the user account, and accordingly, the terminal device 102 is considered to correspond to the user represented by the user account. If the same user account is logged in to multiple terminal devices 102, it is considered that the multiple terminal devices 102 correspond to the same user.

[0026] The server device 101 can execute the object recommendation method provided in the embodiment of this specification, determine the target recommended object for each user corresponding to the terminal device 102, and use the content application platform to send the determined target recommended object information, such as the advertising information of the target recommended object, to the terminal device 102. The advertising information can be a note. Accordingly, the terminal device 102 can display the received target recommended object information, such as showing the user the notes related to the target recommended object.

[0027] In one embodiment, the server device 101 may include a first server device and a second server device, the first server device corresponds to a content application platform, the second server device corresponds to a recommendation method planning platform (i.e., an advertising delivery platform), and the content application platform and the advertising delivery platform are connected. Accordingly, the terminal device 102 may include a terminal device corresponding to a user of the content application platform (i.e., an advertiser), and a terminal device corresponding to a user of the advertising delivery platform (i.e., an ordinary content consumption user). The advertiser may log in to the advertising delivery platform based on the user account, and then may select or configure the information of the product to be promoted (i.e., the object to be recommended) on the page corresponding to the advertising delivery platform to instruct the advertising delivery platform to deliver advertisements for the product. Each advertiser may select one or more products for advertising delivery, and different advertisers may select different products for advertising delivery, or may select the same product for advertising delivery. The advertisement may include product-related product cards, live broadcast information, and published content (such as text or video), etc. The advertising delivery platform may obtain user-related information from the content application platform, plan an object recommendation method based on the information, determine the user to whom each object to be recommended should be recommended, and then the content application platform may make corresponding recommendations to each user based on the object recommendation method.

[0028] The server device 101 is a server responsible for receiving, storing, transcoding, distributing and other tasks of the live stream, which can be an independent server, or a server network or server cluster composed of servers, including but not limited to a computer, a network host, a single network server, a set of multiple network servers or a cloud server composed of multiple servers. Among them, the cloud server is composed of a large number of computers or network servers based on cloud computing.

[0029] The terminal devices 102 can be any type of stationary or mobile computing device, including a mobile computer or mobile computing device (e.g., a tablet computer, a personal digital assistant, a laptop computer, a content-based computer, a netbook, etc.), a mobile phone (e.g., a smart phone), a wearable computing device (e.g., a smart watch, smart glasses, etc.) or other types of mobile devices, smart home devices or vehicle-mounted devices, etc., or a stationary computing device such as a desktop computer or a personal computer.

[0030] Figure 2 is a flowchart of an object recommendation method provided by an embodiment of this specification, and the object recommendation method can be applied to Figure 1 The object recommendation system shown in FIG. 1 can be specifically applied to the server device 101 in the object recommendation system. Figure 2As shown, the object recommendation method specifically includes the following steps 202 and 204. The device that executes the object recommendation method is referred to as an object recommendation device as an example for explanation.

[0031] Step 202: for any target object to be recommended, determine the candidate users matching the target object and the number of recommended users corresponding to the target object.

[0032] In the embodiments of this specification, the object to be recommended may refer to the product that the advertiser needs to promote, and each object to be recommended may correspond to an SPU. The object may be a physical product, such as a mobile phone, camera, computer, or cosmetics, etc.; or a virtual product, such as application software, e-book, or online course, etc. The target object refers to any object to be recommended, and the same recommendation process as the target object may be performed for each object to be recommended.

[0033] For the target object, the object recommendation device can determine the candidate users matching the target object among the users supporting the object recommendation. The users supporting the object recommendation are determined based on the platform for object recommendation. For example, if the object recommendation is performed through a content publishing platform, the users supporting the object recommendation include users of the content publishing platform. By way of example, the object recommendation device can determine the matching relationship between the target object and each user to determine the candidate users of the target object based on the matching relationship. The matching relationship between an object and a user can include the matching degree between the object and the user. The object recommendation device can determine the matching degree between each object and each user based on the object information of multiple objects to be recommended and the user information of multiple users supporting the object recommendation, and then obtain the matching relationship between each object and each user. In one way, the candidate user can be a user whose matching degree with the target object is greater than a set threshold. In another way, the number of candidate users can be set for the target object, and the candidate users can include the number of users with the highest matching degree.

[0034] In one implementation, before determining the candidate user matching the target object for any target object to be recommended in step 202, the object recommendation method provided in the embodiment of this specification further includes: determining the matching relationship between each object and each user based on the object information of multiple objects to be recommended and the user information of multiple users, and recording the matching relationship. Accordingly, step 202 includes: for any target object to be recommended, obtaining the recorded matching relationship between the target object and each user, and determining the candidate user matching the target object among the users based on the obtained matching relationship.

[0035] The object to be recommended is referred to as the object to be recommended below. The advertiser can send the object information of the object to be recommended (such as the target object) to the object recommendation device, and accordingly, the object recommendation device can obtain the object information of the object to be recommended. For example, the advertiser can use its terminal device to access the advertising delivery platform, perform operations in the operation page corresponding to the advertising delivery platform, and upload the configuration information of the object to be recommended to the advertising delivery platform. The configuration information may include the object information of the object to be recommended, such as the basic information such as the name, characteristics and attributes of the object to be recommended. The configuration information may also include recommendation demand information configured for the object to be recommended. The recommendation demand information may include recommendation information for the object to be recommended, as well as the exposure, click volume and conversion volume required for the recommendation information. The recommendation information is also the advertising information of the object to be recommended, and may include at least one of a product card, live broadcast information, published content (such as notes in the form of text, images or videos) and a webpage link for promoting the recommended object.

[0036] Among them, "exposure" refers to the situation where information is loaded and displayed to users. "Conversion" refers to the situation where users complete specific behaviors expected by advertisers after seeing advertising information. The specific behavior is determined by the purpose of the advertising information. For example, for advertising information about goods sold on e-commerce platforms, the specific behavior includes the user completing the order purchase; for advertising information about application software, the specific behavior includes downloading and installing the application software; for advertising information about a certain video, the specific behavior includes the user watching the video and the viewing time is greater than the target time.

[0037] The object recommendation device also obtains user information of users who support object recommendation. The object recommendation device can obtain user information of the corresponding user based on the user account registered on the content application platform. The user information of the user may include attribute information and behavior information of the user. For example, the attribute information of the user may include information such as the user's country, gender, place of origin, current geographical area, age, and occupation; the behavior information of the user may include the user's historical operation information on the browsed content, such as information on operations such as likes, dislikes, collections, comments, and sharing of certain content.

[0038] In some embodiments, the object recommendation device can also analyze the activity level of each user account, and obtain corresponding user information for user accounts with an activity level greater than a certain threshold; for user accounts with too low activity levels, it is not necessary to obtain corresponding user information. This can avoid obtaining a lot of useless redundant information and reduce information processing pressure.

[0039] The object recommendation device can obtain the object information of multiple objects to be recommended and the user information of multiple users, and then analyze the object information and user information to determine the matching degree between each object and each user. There is a matching degree between each object and each user, and the matching degree is used to indicate the user's interest in the object to be recommended. The object recommendation device can periodically determine the matching degree between each user and each object, and the period can be an hour-level duration, such as one hour, two hours or other durations.

[0040] In one embodiment, the object recommendation device can determine the object feature information of each object based on the object information of the multiple objects, and determine the user feature information of each user based on the user information of the multiple users; based on the object feature information of each object to be recommended and the user feature information of each user, calculate the matching degree of each object and each user.

[0041] The object recommendation device can extract the features of each user information to obtain user feature information, and extract the features of each object information to obtain object feature information; then compare the two types of feature information to determine the feature similarity, so as to determine the matching degree between the corresponding object and the user based on the feature similarity, and the matching degree is positively correlated with the feature similarity. In one method, the feature similarity can be directly used as the matching degree. For example, the object recommendation device can encode the obtained feature information into a vector, and obtain the feature similarity by calculating the vector similarity.

[0042] The object recommendation device can use the twin-tower model to model the interest correlation between users and objects, and calculate the matching degree between each object and each user based on the object information of each object to be recommended and the user information of each user. Before using the twin-tower model to calculate the matching degree, the twin-tower model can be trained using multiple sample user information and multiple sample object information, so that the twin-tower model can construct accurate user feature information for user information and accurate object feature information for object information, ensuring that the matching degree can be determined more accurately based on these feature information.

[0043] Figure 3 This is a schematic diagram of the structure of a double tower model provided in one embodiment of this specification. Figure 3As shown, the dual-tower model can include two independent sub-models (called "towers"), namely the user tower and the object tower. The input of the user tower is the user information of each user. The user tower can perform feature embedding processing on the input user information and convert the data into a user feature representation of a vector of fixed size for easy processing and calculation. Feature embedding can map high-dimensional data to a low-dimensional space, thereby retaining the semantic relationship and similarity between the data so that it can be used more effectively. Afterwards, the user feature representation of the vector can be input into a deep neural network (DNN) for nonlinear transformation processing to obtain a user embedding vector (User embedding), which is the above-mentioned user feature information. In this way, the user feature information of each user can be obtained through the user tower.

[0044] The input of the object tower can be the object information of each object to be recommended, such as the advertisement (Ad) information of the object. The object tower can perform feature embedding processing on the input advertising information and convert the data into a fixed-size vector advertising feature representation. After that, the vector advertising feature representation can be input into the DNN for nonlinear transformation processing to obtain an advertising embedding vector (Ad embedding). On this basis, features of the object granularity (such as SPU granularity) (such as brand, product type, specifications, etc.) can also be added to obtain an object embedding vector (SPU embedding), which is the above-mentioned object feature information, to ensure that the object feature information can more accurately represent the object to be recommended. In this way, the object feature information of each object to be recommended can be obtained through the object tower.

[0045] After obtaining the user embedding vector and the object embedding vector, the correlation between the corresponding user and the object can be determined by calculating the inner product between the user embedding vector and the object embedding vector. In the embodiment of this specification, the user tower outputs the user embedding vectors of multiple users, which can be expressed in the form of a matrix. The object tower also outputs the object embedding vectors of multiple objects, which can also be expressed in the form of a matrix. Afterwards, the correlation between each user embedding vector and each object embedding vector can be quickly determined by multiplying the two matrices to obtain the matching degree between each user and each object. Assuming that the number of multiple users is m and the number of multiple objects SPU to be recommended is n, the matching degree between the i-th user and the j-th object to be recommended is ; Among them, 0<i≤m, 0<j≤n.

[0046] In the embodiments of this specification, the number of users and the number of objects to be recommended can be relatively large. By using the double-tower model, the correlation between users and objects to be recommended can be directly obtained by multiplying the matrices, which can improve the efficiency of determining the correlation, that is, improve the efficiency of determining the interest indicator.

[0047] After the object recommendation device determines the matching degree between each object and each user based on the aforementioned method and obtains the matching relationship based on the matching degree, the matching relationship between each object and each user can be recorded. The time for recommending the target object can be specified by the advertiser, and the actual recommendation of the target object can be made some time after uploading the object information of the target object. When it is necessary to recommend the target object, such as after the advertiser of the target object triggers a recommendation request for the target object, the object recommendation device can obtain the recorded matching relationship between the target object and each user. Based on the matching relationship and the conditions that the candidate users of the target object need to meet (the above-mentioned conditions related to the matching degree threshold or the number of users), the candidate users who match the target object can be determined among the users who support accepting the object recommendation.

[0048] In an optional manner, the object recommendation device may also directly determine the candidate user of the target object based on the matching degree after obtaining the object information of the target object, and record the candidate user. When the target object actually needs to be recommended, the recorded candidate user of the target object is directly obtained. In another optional manner, if the advertiser needs to recommend the target object immediately after uploading the object information of the target object, the object recommendation device may directly determine the candidate user after determining the matching degree between the target object and each user, and perform object recommendation based on the candidate user, without additionally performing the matching degree or candidate user recording process.

[0049] The object to be recommended may correspond to associated information under multiple information dimensions (or granularity) (such as multi-dimensional associated information). In one embodiment, the object recommendation device can filter the object information required in the process of determining the degree of matching from the multi-dimensional associated information of the object. The type that the object information needs to meet can be preset. Accordingly, before determining the matching relationship between each object and each user based on the object information of the multiple objects to be recommended and the user information of the multiple users, the object recommendation method provided in the embodiment of this specification also includes: for any object among the multiple objects to be recommended, obtaining the multi-dimensional associated information of the any object; from the multi-dimensional associated information, filtering the standardized object information describing the any object. Afterwards, the standardized object information of the object can be used to determine the matching relationship between the object and the user.

[0050] Various related information of the object to be recommended can be divided according to the size (or thickness) of the information dimension to which it belongs, to obtain multi-dimensional related information. Information in a larger information dimension is used to describe the object to be recommended in a more holistic and macroscopic aspect, and information in a smaller information dimension is used to describe the object to be recommended in a more detailed aspect. A restriction range of the information dimension to be obtained can be set for the information to be obtained, so that the information under the information dimension within the restriction range can be used as the standardized object information to be obtained. In the embodiment of this specification, information of a larger information dimension needs to be obtained for the object to be recommended as standardized object information, such as a minimum allowable information dimension threshold can be set for the object to be recommended.

[0051] For different objects to be recommended, the division method of information dimensions can be the same or different. For different objects to be recommended, information under different information dimensions can be obtained as standardized object information, and the minimum information dimension thresholds corresponding to different objects to be recommended can be different; different objects to be recommended can also set the same minimum information dimension threshold, which is not limited here.

[0052] In one example, the multi-dimensional associated information of the object to be recommended includes the attribute information of the object to be recommended itself and the advertising information of the object to be recommended. The information of the object to be recommended itself can be information of SPU granularity, and the information dimension to which it belongs can be relatively large, which can usually meet the required information dimension of the required information. The attribute information of the object to be recommended may include the name, category, brand, function, purpose or price of the object to be recommended. The advertising information of the object to be recommended may include the category of the object to be promoted (such as clothing or electronic products, etc.), the brand of the object, the type of advertising content (promotional notes or marketing plans, etc.), the presentation method (such as text or video, etc.), and the style (such as funny, serious or relaxed, etc.).

[0053] The advertisement information of the object to be recommended may include information of multiple information dimensions, such as object category and object brand belonging to the larger information dimension, and advertisement content type, presentation method and style belonging to the smaller information dimension. The advertisement information may be specifically screened, and the information of the smaller information dimension that does not meet the information dimension requirements may be deleted, and the remaining information that meets the requirements may be used as the standardized object information.

[0054] In the embodiments of the present specification, standardized object information describing the object to be recommended is obtained based on the advertisement information combined with the information of the object to be recommended itself, so as to ensure that the obtained object information can more accurately describe the object to be recommended, and accordingly, it can ensure that a more accurate match with the user can be achieved based on the object information.

[0055] In the embodiments of this specification, the target object may have different recommendation stages, and the number of recommended users required in different recommendation stages may be different. The recommendation stage can be represented by state information. Accordingly, in step 202, for any target object to be recommended, the candidate users matching the target object and the number of recommended users corresponding to the target object are determined, including: for any target object to be recommended, the candidate users matching the target object are determined; based on the state information of the target object, the number of recommended users corresponding to the target object is determined. The method for determining the candidate users can refer to the above-mentioned related introduction, which will not be repeated here. The content related to the state information is introduced below.

[0056] In the embodiments of the present specification, the state information of the target object may represent the current advertising delivery state of the target object. At different times after the advertisement begins to be delivered, the target object may correspond to different advertising delivery states according to the specific delivery effect. Under different advertising delivery states, the recommendation scheme required to continue recommending the target object may be different accordingly, and the number of users to whom the target object needs to be recommended (i.e., the number of recommended users) is also adjusted as the advertising state changes, so as to achieve a better recommendation effect by adaptively adjusting the recommendation scheme.

[0057] In one example, the advertising delivery status includes an initial state, a mature state, and a declining state. In the initial state, insufficient data accumulation may result in a low conversion rate for the advertisement; in the mature state, the advertisement should achieve a larger amount of faster conversions; in the declining state, due to market saturation, reduced user interest, or the target users basically achieving conversion, the advertising effect will gradually decline, and the conversion rate will decrease accordingly. In the initial state, the delivery of advertisements can be increased slowly, so that the number of recommended users corresponding to the target object increases slowly; in the mature state, the delivery of advertisements can be increased more quickly, so that the number of recommended users corresponding to the target object increases rapidly; in the declining state, the delivery of advertisements can be reduced, so that the number of recommended users corresponding to the target object gradually decreases.

[0058] In the embodiments of this specification, the reverse funnel crowd strategy, which is opposite to the traditional marketing funnel model, is used to determine the corresponding number of recommended users for the target object. The traditional marketing funnel starts with a large number of potential users, gradually narrows the scope through a series of screening steps, and finally obtains a small number of high-quality users. The reverse funnel crowd strategy starts with high-quality and high-value users, and then attracts more ordinary users through the influence of these users. Figure 4This is a schematic diagram of an anti-funnel crowd strategy provided in an embodiment of this specification. The anti-funnel crowd strategy starts with the core users, identifies and attracts the core population that is most valuable to the target object, and penetrates the core population. Then, word-of-mouth communication is used to rely on the usage experience and positive feedback of the core population to form a good reputation, gradually break the circle and generalize, thereby attracting more interested people, and then gradually attracting peripheral general populations and potential people. Through this anti-funnel crowd strategy, the target object can attract more users, enhance user loyalty, reduce customer acquisition costs, and achieve better object recommendation effects.

[0059] In one embodiment, determining the number of recommended users corresponding to the target object based on the status information of the target object includes: determining the user type corresponding to the target object according to a preset period; and determining the number of recommended users corresponding to the target object based on the status information of the target object within the user range covered by the user type.

[0060] Among them, user types may include core groups, interest groups, general groups, and potential groups, and different user types cover different user ranges. For any object to be recommended (such as a target object), users of different user types have different degrees of interest in the target object, and the interest of the core group, interest group, general group, and potential group in the target object may gradually decrease. For example, the user range covered by the core group includes users whose interest in the target object is greater than a first threshold, and the user range covered by the interest group includes users whose interest in the target object is greater than a second threshold, and the first threshold is greater than the second threshold.

[0061] For the target object, the object recommendation device can determine which user type of users should currently recommend the target object to according to a preset period (that is, periodically), that is, determine the user type currently corresponding to the target object, and accordingly determine the user range covered by the user type. By way of example, the preset period can be one day, two days, or other durations. As the cycle progresses, the user type can gradually change from a core population to an interested population, and then gradually to a general population and a potential population, and the user range covered gradually becomes wider and wider. Optionally, the user range covered by the user type corresponding to the target object may include at least some of the aforementioned candidate users, that is, the user range covered by the user type can be determined among the candidate users.

[0062] In one method, the number of users corresponding to each user type is pre-set, and the object recommendation device can determine the degree of interest of each user in the target object, and then determine, based on the number of users corresponding to the user type, that the user range covered by the user type includes the number of users with the highest degree of interest. Assuming that the number of users corresponding to the core population is one thousand, when the current user type is the core population, the object recommendation device can take the first one thousand people who are most interested in the recommended object as the users covered by the core population. The degree of interest can be the same as the above-mentioned matching degree, and the method for determining the degree of interest can refer to the above-mentioned introduction to the matching degree. Optionally, the degree of interest can also be determined in a manner different from the matching degree, which is not limited here.

[0063] Under different user types, the number of users to whom the target object needs to be recommended (i.e., the number of recommended users) is also different. Continuing with the above assumption that the number of users corresponding to the core population is one thousand people, the object recommendation device can recommend the target object for these one thousand people. In the embodiment of this specification, under the user type determined in each cycle, the object recommendation device can also determine the corresponding number of recommended users for the user range covered by the user type, and then recommend the target object to the users of the recommended number of users, instead of directly taking all users covered by the user type as the actual recommended users. The recommended user data may be less than or equal to the number of users corresponding to the current user type. In the process of determining the advancement of the cycle, the user range covered by the user type gradually increases, and the corresponding number of users to be recommended also gradually increases. In different periods under the same user type, the required number of recommended users can also gradually increase. In this way, it can be ensured that the recommended users continue to influence the unrecommended users throughout the recommendation process, gradually expand the scope of influence, and achieve better recommendation and publicity effects.

[0064] In the embodiment of this specification, for products that need to be advertised, the object recommendation device periodically determines the current advertising stage of the product (indicated by the above-mentioned state information) and the user group (corresponding to the above-mentioned user type) that needs to be targeted for advertising in the current stage. Based on the advertising stage, the required advertising volume in this stage is determined, and the number of users to be advertised in the corresponding user group (corresponding to the number of recommended users) is obtained, and then the users to be ultimately advertised are determined based on the number of users.

[0065] The object recommendation device can specifically calculate the number of recommended users based on the current state information of the target object. In one embodiment, the state information of the target object can be represented by some numerical values, and the larger the numerical value, the later the advertising state in the life cycle of the advertising delivery can be represented. By using a certain function or calculation method, the numerical value representing the state information is calculated to obtain the specific number of recommended users.

[0066] In an optional determination method of the status information, the status information of the target object can be directly specified by the advertiser, such as the advertiser can periodically specify the status information of the target object in the next period. In another optional determination method, the object recommendation device determines the status information of the target object based on the duration of the advertisement of the target object. For example, the longer the duration of the advertisement, the later the recommendation stage indicated by the corresponding status information.

[0067] In another way of determining the status information, the object recommendation device can determine the status information of the target object based on the information of the user behavior received by the target object based on the delivered advertising information. The information of the user behavior may belong to the recommendation feedback information, which can characterize the current recommendation effect. User behavior may include exposure behavior, click behavior and conversion behavior, and correspondingly the recommendation feedback information may include exposure information, click information and conversion information. For example, the information of user behavior received from multiple terminal devices for the advertising information of the target object can be integrated to obtain the overall exposure information, click information and conversion information. The exposure information may include the total exposure and / or exposure rate, the click information may include the total clicks and / or click rate, and the conversion information may include the total conversion amount and / or conversion rate.

[0068] Accordingly, before determining the number of recommended users corresponding to the target object based on the state information of the target object in step 204, the object recommendation method provided in the embodiment of this specification further includes: determining the state information of the target object based on the recommendation feedback information of the target object, wherein the recommendation feedback information includes at least one of exposure information, click information, and conversion information. Exemplarily, the state information θ of the target object satisfies the following formula, θ=ω1×imp_cnt+ω2×click_cnt+ω3×conv_cnt. Among them, imp_cnt represents the exposure amount, click_cnt represents the click amount, conv_cnt represents the conversion amount, and ω1, ω2, and ω3 represent the weights corresponding to the exposure amount, the click amount, and the conversion amount, respectively.

[0069] In one embodiment, the number of recommended users corresponding to the target object and its status information can satisfy a set function, and the object recommendation device can determine the number of recommended users corresponding to the target object based on the status information of the target object and the set function. In the embodiment of this specification, the number of users required for advertising of the recommended object needs to continue to grow, so the function is a monotonically increasing function. If the function is an S-type function, correspondingly, the number of recommended users user_num=k1×sigmoid(θ-k2)+k3. Among them, k1, k2 and k3 are function hyperparameters, which are used to control the steepness and range of change of the function curve. The values ​​of k1, k2 and k3 can be obtained through certain training.

[0070] In the embodiment of this specification, the state information of the target object may also be determined periodically, and the determination period of the state information of the target object and the determination period of the number of recommended users may be the same, such as determining a new period on a daily basis. After each update of the state information, the determined number of recommended users is updated accordingly.

[0071] In the embodiments of this specification, after advertising is placed for any product (i.e., recommendation is made for the target object), the corresponding advertising effect (corresponding to the above-mentioned recommendation effect) can be observed to adaptively adjust the subsequent advertising delivery method according to the currently obtained advertising effect. The advertising effect of the product can be reflected by the behavior of the user who received the advertisement towards the advertisement. The object recommendation device determines the advertising effect based on the viewing, clicking, and jump purchase of each user who received the advertisement. The current advertising delivery stage of the product can be determined based on the current advertising effect. The advertising delivery stage is indicated by the status information of the product. Accordingly, the corresponding status information can be obtained based on the advertising effect.

[0072] Step 204: Based on the matching relationship between the target object and the candidate user, construct global matching information between the object to be recommended and the user, and optimize and analyze the global matching information with the number of target users allocated to the target object reaching the number of recommended users as a constraint condition to determine the target recommended object for each user.

[0073] In the embodiments of this specification, for any product that needs to be advertised (such as a target object), there is a required minimum delivery amount (corresponding to the number of recommended users), and the advertising delivery for the product needs to meet this minimum delivery amount. Candidate users for advertising delivery can be determined for each product, and based on the candidate users and the minimum delivery amount, it is determined to which users the advertisement of the product will be ultimately delivered. The object recommendation device can comprehensively consider the various products that need to be advertised based on the candidate users and the minimum delivery amount of the product, and based on the overall matching between the product and the user (corresponding to the global matching information), determine which users the advertisement of each product will ultimately be delivered to, and accordingly determine the advertisements that need to be delivered to each user (that is, determine the target recommendation object for each user).

[0074] For the target object, the object recommendation device can construct global matching information between the object to be recommended and the user based on the matching relationship between the target object and the corresponding candidate users, combined with the matching relationship between other objects to be recommended and the candidate users of the object. This is equivalent to the object recommendation device performing an overall analysis of the matching relationship between each object to be recommended and the candidate users corresponding to each object, and constructing global matching information between the object to be recommended and the user. By way of example, the object recommendation device can periodically calculate the global matching information to determine the target recommended object for each user. This period can be the same as the period for determining the matching degree, such as an hourly period. After each user's matching degree for each object to be recommended is updated, the target recommended object for each user is updated based on the updated matching degree.

[0075] The global matching information can be obtained based on the distribution between each user and each object in combination with the corresponding matching degree. The distribution between each user and each object will affect each other. For example, after a user is assigned to several objects, he can no longer be assigned to other objects. The object recommendation device can adjust the distribution between objects and users in order to obtain a better global matching information that meets the needs. If the global matching information includes a global matching degree represented by a numerical value, the distribution situation is adjusted in order to obtain a global matching degree with the maximum value. Based on the obtained better global matching information, the objects to be recommended can be assigned to suitable users as a whole to ensure that a better recommendation effect is achieved for each object as a whole. In the embodiment of this specification, the matching relationship between an object and a user includes the matching degree between the object and the user, and can also include a matching indicator indicating whether the object matches the user (that is, indicating the above-mentioned distribution situation).

[0076] In the process of adjusting the allocation (adjusting the matching index accordingly), it is also necessary to ensure that each object to be recommended meets its recommendation requirements, which may at least include the corresponding number of recommended users. For example, for the target object, it should be ensured that the number of users allocated to the target object is not less than the corresponding number of recommended users to ensure the recommendation effect of the target object. Accordingly, the object recommendation device needs to set constraints for the optimization of the global matching information to ensure that the adjustment of the allocation situation is required to meet the requirement that the number of target users allocated to the target object reaches the corresponding number of recommended users. After optimizing and analyzing the global matching information based on the constraints, the allocation method of users and objects corresponding to the maximum global matching degree can be obtained, and the target recommended object for each user is determined accordingly. Subsequently, each object can be recommended based on the allocation method, and the corresponding target recommended object can be recommended to each user.

[0077] In one embodiment, step 204 may include: determining the global matching information between the object to be recommended and the user based on the matching index and matching degree in the matching relationship between the target object and the corresponding candidate user; adjusting the matching index with the number of target users allocated to the target object reaching the corresponding number of recommended users as a constraint condition until the global matching information is greater than the target threshold; and determining the target recommended object for each user based on the adjusted matching index.

[0078] The matching index can indicate the above-mentioned distribution situation, and is used to characterize whether the target object is recommended to the candidate user. The matching index can include a numerical value, and different numerical values ​​can be used to characterize different distribution situations. For example, 1 and 0 can be used to represent recommending an object to a user and not recommending the object to a user, respectively; that is, for an object to be recommended, if the object is recommended to a user, the matching index between the user and the object is marked as 1; if the object is not recommended to the user, the matching index between the user and the object is marked as 0. Optionally, in the process of determining the global matching information, for each object to be recommended, only the matching degree and matching index between it and the candidate user can be considered, and the matching degree and matching index of each object and users other than the corresponding candidate user are not included in the calculation. In this way, the matching index of each object and users other than the corresponding candidate user can also be directly set to 0, and the matching index of the user is not adjusted.

[0079] The object recommendation device can determine the product of the matching degree of an object and a user and the corresponding matching index as the updated matching degree between the object and the user. The sum of the updated matching degrees between each object and each user is determined as the global matching degree between the object to be recommended and the user. For example, the object recommendation device can first set the matching index of each object and each user to 0 (or 1), and then adjust the matching index of each object and each user on the basis of ensuring that the set constraint conditions are met, and adjust some matching indicators to 1 (or 0) to observe the adjusted global matching information. In this way, a global matching degree greater than the target threshold is determined, and based on the matching index between each object and each user at this time, the user assigned to each object is determined, and the target recommended object for each user is obtained accordingly. In one method, only the candidate users of each object are considered, so the sum of the updated matching degrees between each object and its corresponding candidate users can be determined as the global matching degree between the object and the user.

[0080] The target threshold may be a pre-set empirical value, or may specify the maximum global matching degree to be determined. When the maximum global matching degree is determined, the global matching degree is considered to be greater than the target threshold. In other words, the object recommendation device needs to determine the final matching index between each object to be recommended and each user based on the first formula. The first formula is ,and ; ;s ij ≥0,d j ≥0. Among them, s ij represents the matching degree between the jth object and the ith user, x ij represents the matching index between the jth object and the ith user, d j Indicates the number of recommended users corresponding to the jth object. " " means that the matching index between the jth object and the ith user is 0 or 1, and this restriction is satisfied for all users and objects." " means that the sum of the matching indexes between the jth object and each user is greater than or equal to d j , and this restriction is satisfied for all objects to be recommended. The object recommendation device needs to solve x based on the first formula ij , and obtain the optimal matching method between users and objects.

[0081] The above adjustment of the matching index of the target object and its candidate users so that the number of target users allocated to the target object reaches the number of recommended users is satisfied, which is an optional implementation method for optimizing the analysis of the global matching information. In other implementations, the object recommendation device can also optimize the analysis of the global matching information by adjusting the candidate users of the target object, such as adding candidate users or replacing some candidate users with low activity, while ensuring that the constraint condition is met. After each adjustment, the object recommendation device can determine whether the constraint condition is met. If the constraint condition is met, it is determined whether the global matching information obtained after the adjustment is better than the original global matching information. If it is better, the adjustment will continue based on the allocation at this time until the global matching information obtained meets the condition for stopping the adjustment. For example, the condition includes that the global matching degree is greater than the target threshold, or the maximum global matching degree is obtained, or the number of adjustments is greater than the number threshold.

[0082] In the embodiments of this specification, corresponding recommendation weights can also be set for each user respectively, so as to determine the global matching degree between the object to be recommended and the user based on the recommendation weight. The above-mentioned method for determining the global matching degree can be adopted when treating each user equally when recommending an object, that is, it is considered that the recommendation weight corresponding to each user is 1. In one embodiment, the object recommendation device can determine the recommendation weight of the user based on the user's behavior information, so that the recommendation weights of different users can be different, and more attention can be paid to users with higher recommendation weights when recommending objects. Accordingly, the user information obtained above may include behavior information. The object recommendation method provided in the embodiments of this specification also includes: determining the recommendation weight of the candidate user based on the behavior information corresponding to the candidate user. In step 204, based on the matching relationship between the target object and the candidate user, constructing the global matching information between the object to be recommended and the user, including: determining the global matching information between the object to be recommended and the user based on the matching relationship between the target object and the candidate user and the recommendation weight of the candidate user.

[0083] The user's behavior information may include information about the user's historical operations in the content application platform, and the user's activity may be determined based on this information. The object recommendation device may determine the corresponding recommendation weight based on the user's activity, and the recommendation weight may be positively correlated with the activity. For example, the user's activity may be determined based on information such as the frequency of access to the content application platform derived from the behavior information, the frequency of content viewing in the content application platform, the amount of content requests sent to the content application platform, and the length of stay in the content application platform.

[0084] The higher the user's activity level in the content application platform, the easier it is for the user to see the object recommendation made to the user, and there is a greater probability of obtaining user feedback on the recommended information, obtaining a greater conversion probability, and correspondingly obtaining a better recommendation effect. Therefore, a larger recommendation weight can be set for users with higher activity levels, so that these users can play a greater role in determining the global matching degree and, accordingly, in the object recommendation process. A smaller recommendation weight can be set for users with lower activity levels to balance the lower conversion probability of recommending to the user. In this way, it is possible to determine the object recommendation plan in accordance with a situation that is more in line with the actual recommendation scenario, which can improve the overall object recommendation effect.

[0085] After determining the recommendation weight of each user, the object recommendation device can determine the global matching information by combining the matching degree of the object to be recommended and its candidate user and the recommendation weight of each candidate user. For example, the object recommendation device can determine the weighted matching degree between the target object and the candidate user based on the recommendation weight of a candidate user of the target object and its matching degree with the target object, and then determine the global matching information by combining the weighted matching degrees of each object to be recommended and each user. The object recommendation device can optimize and adjust the matching indexes of each object and each user by combining the recommendation weight and matching degree, and determine the matching indexes corresponding to the maximum global matching information.

[0086] In consideration of the recommendation weight of each user, the object recommendation device can determine the final matching index between each user and each object based on the second formula. The second formula is ,and ; ;s ij ≥0,d j ≥0, p i ≥0. Among them, p i Represents the recommendation weight of the i-th user. " " means that the sum of the recommendation weight of each user and the product of its matching index with the jth object is greater than or equal to d j , and this restriction is satisfied for all objects.

[0087] In the embodiments of this specification, it is also possible to impose certain restrictions on the number of recommended objects for each user to avoid placing too many advertisements to the user, to avoid the user's attention being overwhelmed by too many recommended objects, causing interference to the user, and affecting the user's experience on the content application platform. For example, a recommendation object threshold can be set for the user, and the recommendation object threshold is the maximum number of recommended objects that the user supports receiving. Correspondingly, the object recommendation method provided in the embodiments of this specification also includes: obtaining the recommendation object threshold of the candidate user. In step 204, the global matching information is optimized and analyzed with the number of target users assigned to the target object reaching the number of recommended users as a constraint condition, including: the global matching information is optimized and analyzed with the number of target users assigned to the target object reaching the corresponding number of recommended users, and the number of objects assigned to the candidate users not exceeding the corresponding recommendation object threshold as constraints.

[0088] The recommended object threshold of each user can be the same to ensure that the object recommendations for each user are relatively balanced, to ensure the consistency of object recommendations, and to ensure the consistency of user experience on the content application platform. For example, the recommended object threshold of a user can be determined based on the total number of users in the content application platform, the number of objects to be recommended, and the number of recommended users corresponding to each object. For example, the recommended object threshold of a user can be determined based on the result obtained by dividing the sum of the number of recommended users corresponding to each object by the total number of users.

[0089] In some implementations, the user's recommended object threshold may also be different. For example, the user can configure the object recommendation that he can receive, and the corresponding content application platform can determine the corresponding recommended object threshold for the user based on the user's configuration information. In this way, the object recommendation can be carried out in a targeted manner based on the different needs of the user, thereby improving the flexibility of the object recommendation.

[0090] When a recommendation object threshold is set for each user, the object recommendation device needs to consider the recommendation object threshold when determining the global matching information. If the recommendation object threshold of each user is also used as a constraint condition to optimize the analysis of the global matching information, it is ensured that the number of recommended objects allocated to each user does not exceed the corresponding recommendation object threshold. In this case, the first formula and the second formula above also need to satisfy , that is, the sum of the products of the matching indexes of the i-th user and the objects to be recommended is less than or equal to Q, and this restriction is satisfied for all users. When adding the constraint conditions related to the threshold of the recommended object, the optimization analysis method of the global matching information can still refer to the above-mentioned introduction to the optimization analysis based on the constraint conditions related to the number of recommended users, which will not be repeated here.

[0091] In the embodiments of this specification, for users who accept advertising recommendations, the maximum number of advertisements that can be accepted (corresponding to the recommendation object threshold) will be set for them. When advertising for each product, this number is used as a constraint to perform analysis and calculation, and the final advertisements to be delivered are determined for each user. This ensures that the number of advertisements delivered to the user does not exceed this number as much as possible, so as to avoid excessive interference to the user and ensure the user's normal content consumption experience.

[0092] In an optional manner, for each object to be recommended, the object recommendation device may not predetermine the corresponding candidate users, but directly use the first formula or the second formula to calculate the global matching information based on the matching degree between each object and all users who support accepting the object recommendation, and determine the allocation method between each object and each user.

[0093] Figure 5 This is a schematic diagram of a matching method between a user and an object to be recommended provided in an embodiment of this specification. Figure 5For example, allocation is performed for four users and three objects to be recommended, the recommendation weights p of the four users are 1, 3, 2 and 1 respectively, the number of recommended users d corresponding to the three objects are 2, 4 and 2 respectively, and the recommended object threshold Q for each user is equal to 2. For the recommendation weights and the number of recommended users, the first user and the third user can be allocated to the first object, the four users are all allocated to the second object, and the second and fourth users are allocated to the third object, so as to ensure that the number of recommended users corresponding to each object and the recommended object threshold of each user are met.

[0094] In the embodiment of this specification, in the process of assigning each user to each object, a corresponding number of recommended users is set for each object to perform a quantity constraint, ensuring that a certain number of users can be assigned to each object. A corresponding threshold of recommended objects is set for each user to perform a quota constraint, ensuring that the objects assigned to each user do not exceed a certain number at most, so as to avoid the user's attention being overwhelmed by too much object recommendation information.

[0095] In the embodiments of this specification, multiple different advertisers or one advertiser can operate in an advertising delivery platform (such as an operating platform for advertisers corresponding to a server-side device) to select multiple products to instruct the advertising delivery platform to deliver advertisements. Accordingly, the server-side device can obtain information about each product to be advertised. The server-side device can also determine a list of candidate users corresponding to each product. For example, the server-side device determines the candidate users corresponding to each product based on the matching relationship between each product and each user, and forms a list of candidate users. The server-side device then comprehensively considers each product, and based on the matching degree between each product and each candidate user, determines a user allocation method that can achieve a better global matching information between the product and the user, obtains the user to which each product should be assigned, and then determines the advertisement for the product that needs to be delivered to each user. After that, the server-side device can deliver product advertisements based on this allocation method.

[0096] Figure 6 This is a flowchart of an overall process of matching a user with an object to be recommended provided by an embodiment of this specification. Figure 6As shown, the object recommendation device can first use the trained dual-tower model to encode the user information of each user and the object information of the object to be recommended (such as SPU) respectively, and obtain the user feature information (user embedding) and object feature information (SPU embedding) represented by the embedding vector (embedding), and then determine the interest index of each user for each SPU based on the user feature information and object feature information. After that, the object recommendation device can use the matching module to match the user and SPU based on the interest index. Among them, for each SPU, the object recommendation device can determine the current anti-funnel crowd stage (that is, determine the current corresponding user type), and then determine the number of recommended users that should be currently allocated to each SPU. In the matching module, the graph decomposition processing can be performed based on the matching degree of each SPU with each user, the circled SPU and the number of recommended users corresponding to each SPU to obtain the SPU corresponding to each user. Since the method of matching users and SPUs in the embodiment of this specification is similar to the matching method of bipartite b-Matching, the matching process is referred to as performing graph decomposition processing here.

[0097] In the embodiments of this specification, the number of users and objects to be recommended in the content application platform can be very large, such as the number of users can reach hundreds of millions, and the number of objects can reach tens of thousands. The matching of users and objects is a complex mixed integer linear programming problem. The matching process is relatively complex and requires a large amount of calculation. The object recommendation device can simplify the matching process and use a parallel algorithm to accelerate the matching solution. In some embodiments, a general mixed integer programming (MIP) solver can also be used to solve the matching.

[0098] For example, in the solution process, the recommendation weight of each user can be equal, such as 1. When supply and demand are balanced, the matching problem can be transformed into a bipartite b-Matching problem, that is, finding a subgraph with the maximum weight. The matching of users and objects needs to be carried out continuously during the operation of the content application platform. It is a network flow problem, and it is necessary to use a polynomial complexity method to find the global optimal solution for this problem. If it is solved directly, the solution will be unacceptably time-consuming due to the huge number of edges (trillions), and it will not be executed on a single machine because the memory is far from enough. For example, the b-suitor algorithm can be used to find an approximate solution. The complexity of this algorithm is , where m is the number of edges (equal to the number of users × the number of recommended objects), and β is the maximum matching number of all points (equal to max d j), △ is the maximum degree of the point (equal to the number of users). In the embodiments of this specification, an approximate algorithm that is solved in parallel on multiple machines can be used to control the machine resource consumption and overall time consumption within an acceptable range.

[0099] Accordingly, before determining the matching relationship between each object and each user based on the object information of the multiple objects to be recommended and the user information of the multiple users, the object recommendation method provided in the embodiment of this specification also includes: dividing the user set into multiple subsets, wherein the subsets include the user information of the multiple users. Determining the matching relationship between each object and each user based on the object information of the multiple objects to be recommended and the user information of the multiple users includes: for each subset of the multiple subsets, calling the corresponding solver, and determining the matching relationship between each object and each user based on the object information of the multiple objects to be recommended and the information of the multiple users in the subset.

[0100] The user set may include all users in the content application platform who support receiving object recommendations. In the embodiments of this specification, these users may be pre-divided into multiple subsets, each of which includes user information of multiple users. The number of users in each subset may be the same, or there may be certain differences. The number of users is much larger than the number of objects to be recommended, and the division may be random to ensure that the distribution of users in each subset is relatively homogeneous, thereby reducing the impact on the final matching effect.

[0101] The multiple users and each user targeted in the aforementioned step 202 and step 204 may refer to multiple users in a subset, and the aforementioned step 202 and step 204 may be performed for each subset to determine the target recommended object for each user in the subset. A corresponding solver may be assigned to each subset, and the solver is used to solve the matching degree between the users in the corresponding subset and the objects to be recommended, and the target recommended object to be matched for each user is determined based on the number of recommended users corresponding to the objects to be recommended in the subset. The solver may determine the total number of recommended users required for the objects to be recommended based on their current state information. Based on the total number of recommended users, the number of recommended users corresponding to each subset is determined based on the number of subsets obtained by division. If the total number of recommended users is divided equally based on the number of subsets, the number of recommended users corresponding to each subset is obtained.

[0102] For example, the Spark framework can be used to split the matching problem between all users and recommended objects in the content application platform into multiple solvers (also called worker nodes) on the cluster for solving. Each solver can calculate in parallel, and the target recommended objects corresponding to the users in each subset can be determined in parallel to improve computing efficiency and avoid insufficient memory on a single machine.

[0103] After using multiple solvers to determine target recommendation objects for users in multiple subsets, the object recommendation device can integrate the obtained solution results to obtain target recommendation objects for all users in the content application platform. The object recommendation device can output the information of the target recommendation object for each user obtained after integration to the content recommendation platform for it to send the recommendation information of the target recommendation object to each user.

[0104] Accordingly, in step 204, based on the matching relationship between the target object and the candidate user, the global matching information between the object to be recommended and the user is constructed, and the target number of users assigned to the target object reaches the number of recommended users as a constraint condition, the global matching information is optimized and analyzed, and the target recommended object for each user is determined, including: calling a solver, based on the matching relationship between the target object and the candidate user, the global matching information between the object to be recommended and the user is constructed, and the target number of users assigned to the target object reaches the number of recommended users as a constraint condition, the global matching information is optimized and analyzed, and the target recommended object for each user is determined; integrating the solution results of each solver to determine the target recommended object for each user. By calling a solver, the target recommended object for each user in a subset can be obtained; after integrating the solution results of each solver, the target recommended object for each user in the user set can be obtained.

[0105] In the embodiments of this specification, each solver can also use multi-threaded parallelism in the process of solving to further improve the solving efficiency. For example, in a single spark worker, pyspark reads the embedding sets on both the user side and the object side from the hive database, first splits the user set and maps it to each worker, and each worker multiplies the embedding matrices on both sides to obtain the correlation between each user and each object, which can be called the estimated decomposition graph of each edge. The solving algorithm can use a parallel version of the b-suitor algorithm, use the C++ OpenMP library to achieve multi-threaded parallel acceleration, and be called in the pyspark environment after being encapsulated through the foreign function interface (CFFI, Foreign Function Interface). In the embodiments of this specification, a single solver can use a solving scale of 8 cores and 64GB of memory to match users and objects, ensuring a good balance between parallel acceleration and solving effect.

[0106] In the embodiments of this specification, the object recommendation device can use the double tower model to calculate the matching degree between the object to be recommended and the user, and characterize the user's interest in the recommended object. Based on the anti-funnel crowd strategy, the number of recommended users that can be assigned to each object is calculated, and at the same time, a super-large-scale Spu2user allocation algorithm is used to determine which users each object should be assigned to. In this way, object allocation can be performed efficiently and accurately, so that the advertising delivery system can more accurately match the objects to be recommended and the users, and help dynamically and accurately allocate the advertiser's products to different target users, thereby improving the product recommendation effect. In addition, for super-large-scale allocation problems, a distributed solution algorithm can be designed and implemented based on the Spark cluster framework. With extremely low resource consumption, the allocation problem can be efficiently solved, greatly improving the timeliness of the allocation.

[0107] In summary, in the object recommendation method provided in the embodiments of this specification, the candidate users that match the target object and the corresponding number of recommended users can be determined for the target object from the perspective of the target object to be recommended. The global matching information between the object and the user is determined based on the matching relationship between the target object and the candidate users, and based on the global matching information, an object recommendation method that makes the user and the object more interested as a whole can be obtained. Taking the number of recommended users corresponding to the target object as a constraint condition, the global matching information is optimized and analyzed, and finally the target recommended object for each user is determined. In this way, it can be ensured that on the basis of meeting the recommendation needs of the target object, an object recommendation method with better overall matching effect is obtained, thereby improving the recommendation effect of each object.

[0108] The object recommendation method provided in the embodiments of this specification can be applied to Figure 1 The object recommendation system shown. Figure 7 This is a timing diagram of an object recommendation method provided by an embodiment of this specification, which introduces the interaction process of each device in the object recommendation system. The server device in the object recommendation system can correspond to an advertising delivery platform for advertisers and a content application platform for content consumption users. The multiple terminal devices in the object recommendation system can include terminal devices corresponding to users of the advertising delivery platform and terminal devices corresponding to users of the content consumption users. Figure 7 The following is an introduction using two terminal devices (the first terminal device and the second terminal device) in the object recommendation system as an example, and the first terminal device is an advertiser's terminal device, and the second terminal device is a content consumer's terminal device. For other terminal devices, you can refer to the introduction of the first terminal device and the second terminal device. Figure 7 As shown, the method may include the following steps 702 to 718. Figure 7 The steps in this chapter can be cross-referenced with the previous ones.

[0109] Step 702: The first terminal device sends object information of a target object to be recommended to the server device.

[0110] The first terminal device may be installed with an application corresponding to the advertising delivery platform, or may access a webpage corresponding to the advertising delivery platform. The server device may correspond to the aforementioned object recommendation device, which is a backend server corresponding to the advertising delivery platform, and is used to support the operation of the advertising delivery platform. Advertisers may use the first terminal device to access the advertising delivery platform, and configure the information of the product to be advertised (such as the object information of the target object) on the corresponding page to trigger the advertising delivery platform to deliver advertisements for the product. Step 702 may refer to the relevant introduction about the object information in the aforementioned step 202, which will not be repeated here.

[0111] Step 704: The second terminal device sends the user information of the user to the server device.

[0112] The second terminal device may have an application corresponding to the content application platform installed, or may access a webpage corresponding to the content application platform. The second terminal device may log in to a user account during access to the content application platform, and the terminal device may obtain corresponding user information based on the user account, which is legal information authorized by the user. For example, the user information may include the user's own attribute information set by the user, and may also include interaction information between the user and the second terminal device.

[0113] The server device may correspond to the aforementioned object recommendation device, and the server device may be a background server corresponding to the content application platform, used to support the operation of the content application platform. The embodiments of this specification only introduce the user information that needs to be obtained in the process of object recommendation. When the user browses regular content, the second terminal device may also send a content request to the server device to request the server device to push the corresponding content, and the recommendation information of the object to be recommended may belong to these contents. The server device may be connected to multiple second terminal devices, and the multiple second terminal devices may all send the user information of the corresponding user to the server device.

[0114] Step 706: The server device determines a matching relationship between the target object and each user based on the object information of the target object and the received user information of the multiple users.

[0115] The user information of the multiple users may be received by the server device from different second terminal devices. The server device may determine the interest level of each user in the product for which advertisements are to be placed, so as to obtain a corresponding matching relationship. Step 706 may refer to the relevant introduction on determining the matching relationship in the aforementioned step 202, which will not be repeated here.

[0116] Step 708: The server device determines candidate users matching the target object based on the matching relationship, and determines the number of recommended users corresponding to the target object based on the state information of the target object.

[0117] The server-side device can determine the candidate user list corresponding to the product (such as the target object) for the product that needs to be advertised based on the interest level of each user in the product, so as to subsequently determine the users to whom the advertisement of the product needs to be delivered in the candidate user list. The server-side device can also determine the current advertising stage of the product and determine the corresponding required advertising delivery volume (corresponding to the number of recommended users), so as to subsequently determine the users to whom the advertisement needs to be delivered based on the advertising delivery volume.

[0118] Step 708 can refer to the relevant introduction in the aforementioned step 202, which will not be repeated here. After obtaining the recommendation demand for an object, the server device can perform steps 706 and 708 for the object to determine the matching relationship between the object and each user, and determine the number of candidate users and recommended users for the object, so as to obtain the matching relationship between each object and each user, and the number of recommended users for each object.

[0119] Step 710: The server device determines the global matching information between the object to be recommended and the user based on the matching relationship between the target object and its candidate users, and optimizes and analyzes the global matching information with the number of target users allocated to the target object reaching the corresponding number of recommended users as a constraint condition to determine the target recommended object for each user.

[0120] The target object is any object to be recommended. In other words, the server device can determine the global matching information between the object to be recommended and the user based on the matching relationship between each object and its candidate user, and optimize the global matching degree with the number of target users allocated to each object reaching the corresponding number of recommended users as a constraint condition to determine the target recommended object for each user. Step 710 can refer to the relevant introduction in the aforementioned step 204, which will not be repeated here.

[0121] In the embodiments of this specification, the server-side device determines the user allocation method that can achieve the best global matching information between the product and the user based on the matching relationship between the product and its candidate users, and comprehensively considers each product for the product that needs to be advertised (such as the target object), and obtains the user to which each product should be assigned, and then determines the advertisement of the product that needs to be placed for each user. After that, the server-side device can place product advertisements based on the allocation method, so as to ensure that the overall advertising effect of each product is good.

[0122] Step 712: The server device sends the recommendation information of the target recommendation object of the user corresponding to the second terminal device to the second terminal device.

[0123] After determining the target recommendation object for each user, the server device can recommend the corresponding target recommendation object to the user, such as sending the corresponding recommendation information to the terminal device corresponding to the user. The recommendation information of the target recommendation object is also the advertising information of the product, such as at least one of the product card, live broadcast information, published content (such as notes in the form of text, images or videos), and web page links. The recommendation information of each target recommendation object can be pre-specified by the advertiser of the target recommendation object, or can be determined by the server device itself through analysis.

[0124] Step 714: The second terminal device displays the received recommendation information of the target recommendation object.

[0125] After receiving the recommendation information sent by the server device, the second terminal device can display it based on the set display method, such as displaying the received recommendation information of the target recommendation object at a specified position in the display page of the content application platform to recommend the target recommendation object. For example, the second terminal device can display the product advertisement at a specific position (such as the third display position) on the homepage of the application corresponding to the content application platform, or display it in the form of a splash screen advertisement, or display it in the form of a pop-up advertisement.

[0126] Step 716: The server device sends recommendation effect information to the first terminal device based on the recommendation feedback information of the target object.

[0127] After recommending the target object, the server device can also collect the user's behavior information (i.e., recommendation feedback information) for the target object from the second terminal device to analyze the recommendation effect for the target object, i.e., the advertising effect. The server device can integrate and analyze the recommendation feedback information collected from each second terminal device to determine the indicators that can reflect the recommendation effect of the target object, and then send these indicators as the recommendation effect information of the target object to the first terminal device, so that the advertiser can also know the current advertising effect in a timely manner. Step 716 can refer to the relevant introduction about the recommendation feedback information and recommendation effect in the aforementioned step 202, which will not be repeated here.

[0128] Step 718: The first terminal device displays the received recommendation effect information.

[0129] When the first terminal device receives the recommendation effect information for the target object, it can display the recommendation effect information in the advertising delivery platform using a set display method, so that the advertiser can know the advertising delivery effect. For example, it can be displayed in a display area set on the page of the advertising delivery platform; it can be presented in the form of a chart or in the form of a text description.

[0130] Advertisers can further configure the target object based on the recommendation effect information. For example, they can instruct to stop advertising to the target object, increase the advertising intensity, slow down the advertising intensity, or increase the advertising budget. Accordingly, the advertising delivery platform re-plans the subsequent advertising delivery method based on the configuration made by the advertiser. The configuration process is similar to the method of configuring the object information in the aforementioned step 202, and no additional introduction is given here.

[0131] In the embodiments of this specification, the server device can determine the interest level (i.e., matching level) of each user in each object to be recommended for each user corresponding to each terminal device, determine global matching information based on the matching level, and obtain a method of recommending objects with a higher overall interest level based on the global matching information. In addition, the number of recommended users required for each object is determined based on its status information, and the number of recommended users corresponding to each object to be recommended is used as a constraint condition to optimize and analyze the global matching information, and finally determine the target recommended object for each user, and then the terminal device can display the recommendation information of the target recommended object to the user. In this way, it can be ensured that on the basis of meeting the recommendation requirements of each object, a method of recommending objects with better overall matching effect is obtained, thereby improving the recommendation effect of each object.

[0132] Corresponding to the above method embodiment, this specification also provides an object recommendation device embodiment, Figure 8 is a schematic diagram of the structure of an object recommendation device provided in an embodiment of this specification. Figure 8 As shown, the object recommendation device includes: A determination module 802 is used to determine, for any target object to be recommended, candidate users matching the target object and the number of recommended users corresponding to the target object; The analysis module 804 is used to construct global matching information between the object to be recommended and the user based on the matching relationship between the target object and the candidate user, and optimize the global matching information with the number of target users allocated to the target object reaching the number of recommended users as a constraint condition to determine the target recommended object for each user.

[0133] Optionally, the object recommendation device provided in the embodiment of the present specification further includes: a matching relationship determination module, which is used to determine the matching relationship between each object and each user based on the object information of multiple objects to be recommended and the user information of multiple users, and record the matching relationship before determining the candidate user matching the target object for any target object to be recommended; The determination module 802 is used to: for any target object to be recommended, obtain the recorded matching relationship between the target object and each user, and determine a candidate user matching the target object among each user based on the obtained matching relationship.

[0134] Optionally, the determination module 802 is used to: for any target object to be recommended, determine candidate users matching the target object; and determine the number of recommended users corresponding to the target object based on the state information of the target object.

[0135] Optionally, the object recommendation device provided in the embodiment of this specification also includes: a status determination module, which is used to determine the status information of the target object based on the recommendation feedback information of the target object before determining the number of recommended users corresponding to the target object based on the status information of the target object, wherein the recommendation feedback information includes at least one of exposure information, click information and conversion information.

[0136] Optionally, the determination module 802 is used to: determine the user type corresponding to the target object according to a preset period; and determine the number of recommended users corresponding to the target object based on the state information of the target object within the user range covered by the user type.

[0137] Optionally, the matching relationship between an object and a user includes: a matching index indicating whether the object matches the user, and a matching degree indicating the object and the user; the analysis module 804 is used to: Based on the matching index and matching degree in the matching relationship between the target object and the candidate user, the global matching information of the object to be recommended and the user is constructed; with the number of target users allocated to the target object reaching the number of recommended users as a constraint, the matching index is adjusted until the global matching information is greater than the target threshold; based on the adjusted matching index, the target recommended object for each user is determined.

[0138] Optionally, the object recommendation device provided in the embodiment of this specification further includes: a weight determination module, configured to determine the recommendation weight of the candidate user based on the behavior information corresponding to the candidate user; The analysis module 804 is used to determine the global matching information between the object to be recommended and the user based on the matching relationship between the target object and the candidate user and the recommendation weight of the candidate user.

[0139] Optionally, the object recommendation device provided in the embodiment of this specification also includes: an acquisition module for obtaining a recommended object threshold for candidate users; an analysis module 804 for optimizing the analysis of global matching information with the number of target users assigned to the target objects reaching the number of recommended users and the number of objects assigned to the candidate users not exceeding the recommended object threshold as constraints.

[0140] Optionally, the object recommendation device provided in the embodiment of this specification further includes: An information acquisition module, used for acquiring multi-dimensional association information of any object among the multiple objects to be recommended before determining a matching relationship between each object and each user based on the object information of the multiple objects to be recommended and the user information of the multiple users; The screening module is used to screen the standardized object information describing any object from the multi-dimensional association information.

[0141] Optionally, the object recommendation device provided in the embodiment of this specification further includes: a partitioning module, configured to partition the user set into a plurality of subsets before determining a matching relationship between each object and each user based on the object information of the plurality of objects to be recommended and the user information of the plurality of users, wherein the subsets include the user information of the plurality of users; The determination module 802 is used to call a corresponding solver for each subset in the plurality of subsets, and determine a matching relationship between each object and each user based on the object information of the plurality of objects to be recommended and the information of the plurality of users in the subset.

[0142] Optionally, the analysis module 804 is used to: call the solver, construct global matching information between the object to be recommended and the user based on the matching relationship between the target object and the candidate user, and optimize the global matching information with the number of target users allocated to the target object reaching the number of recommended users as a constraint condition to determine the target recommended object for each user; integrate the solution results of each solver to determine the target recommended object for each user.

[0143] The above is a schematic scheme of the object recommendation device of this embodiment. It should be noted that the technical scheme of the object recommendation device and the technical scheme of the corresponding object recommendation method described above belong to the same concept, and the details not described in detail in the technical scheme of the object recommendation device can be referred to the description of the technical scheme of the object recommendation method described above.

[0144] Fig. 99 is a block diagram of a computing device provided in an embodiment of the present specification. The components of the computing device 900 include, but are not limited to, a memory 910 and a processor 920. The processor 920 is connected to the memory 910 via a bus 930, and a database 950 is used to store data. The processor 920 is used to execute a computer program / instruction, and when the computer program / instruction is executed by the processor, the steps in the above-mentioned object recommendation method are implemented.

[0145] The computing device 900 also includes an access device 940 that enables the computing device 900 to communicate via one or more networks 960. Examples of these networks include a public switched telephone network (PSTN), a local area network (LAN), a wide area network (WAN), a personal area network (PAN), or a combination of communication networks such as the Internet. The access device 940 may include one or more of any type of network interface (e.g., a network interface card (NIC)) that is wired or wireless, such as an IEEE 802.11 wireless local area network (WLAN) wireless interface, a world-wide interoperability for microwave access (Wi-MAX) interface, an Ethernet interface, a universal serial bus (USB) interface, a cellular network interface, a Bluetooth interface, and a near field communication (NFC).

[0146] In one embodiment of the present specification, the above components of the computing device 900 and Fig. 9 Other components not shown in the figure may also be connected to each other, for example, via a bus. It should be understood that Fig. 9The computing device structure block diagram shown is for illustrative purposes only and is not intended to limit the scope of this specification. Those skilled in the art may add or replace other components as needed. The computing device 900 may be a mobile or stationary server or server cluster, or may be some stationary or mobile computing devices with large data storage capabilities. For example, mobile computers or mobile computing devices may include tablet computers, personal digital assistants, laptop computers, notebook computers, netbooks, etc., and stationary computing devices may include desktop computers or personal computers (PCs), etc. For the computing device embodiment, since it is basically similar to the object recommendation method embodiment, the description is relatively simple, and the relevant parts may refer to the partial description of the object recommendation method embodiment.

[0147] One embodiment of the present specification also provides a computer-readable storage medium, which stores a computer program / instruction, which implements the steps of the above-mentioned object recommendation method when executed by a processor. The computer program / instruction includes a computer program code, which may be in source code form, object code form, executable file or some intermediate form, etc. The computer-readable storage medium may include: any entity or device capable of carrying the computer program code, recording medium, USB flash drive, mobile hard disk, disk, optical disk, computer memory, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), electric carrier signal, telecommunication signal and software distribution medium, etc. It should be noted that the content contained in the computer-readable storage medium can be appropriately increased or decreased according to the requirements of legislation and patent practice in the jurisdiction. For example, in some jurisdictions, according to legislation and patent practice, computer-readable storage media do not include electric carrier signals and telecommunication signals.

[0148] An embodiment of the present specification also provides a computer program product, including a computer program / instruction, which implements the steps in the above-mentioned object recommendation method when the computer program / instruction is executed in a processor.

[0149] As for the computer-readable storage medium embodiment and the computer program product embodiment, since they are basically similar to the object recommendation method embodiment, the description is relatively simple, and the relevant parts can be referred to the partial description of the object recommendation method embodiment.

[0150] The above is a description of a specific embodiment of the specification. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps recorded in the claims can be performed in an order different from that in the embodiments and still achieve the desired results. In addition, the processes depicted in the drawings do not necessarily require the specific order or continuous order shown to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0151] It should be noted that, for the aforementioned method embodiments, for the sake of simplicity of description, they are all expressed as a series of action combinations, but those skilled in the art should be aware that the embodiments of this specification are not limited to the described action sequence, because according to the embodiments of this specification, some steps can be performed in other sequences or simultaneously. Secondly, those skilled in the art should also be aware that the embodiments described in the specification are all preferred embodiments, and the actions and modules involved are not necessarily required by the embodiments of this specification. In the above embodiments, the descriptions of each embodiment have their own emphasis, and for the parts that are not described in detail in a certain embodiment, please refer to the relevant descriptions of other embodiments. The preferred embodiments of this specification disclosed above are only used to help explain this specification. The optional embodiments do not describe all the details in detail, nor do they limit the invention to the specific implementation methods described. Obviously, according to the content of the embodiments of this specification, many modifications and changes can be made. This specification selects and specifically describes these embodiments in order to better explain the principles and practical applications of the embodiments of this specification, so that those skilled in the art can understand and use this specification well. This specification is only limited by the claims and their full scope and equivalents.

Claims

1. An object recommendation method, characterized in that: include: For any target object to be recommended, determine the candidate users matching the target object and the number of recommended users corresponding to the target object; Based on the matching relationship between the target object and the candidate user, global matching information of the object to be recommended and the user is constructed, and with the number of target users allocated to the target object reaching the number of recommended users as a constraint condition, the global matching information is optimized and analyzed to determine the target recommended object for each user.

2. The method according to claim 1, characterized in that Before determining a candidate user matching any target object to be recommended, the method further includes: Based on the object information of the multiple objects to be recommended and the user information of the multiple users, determining the matching relationship between each object and each user, and recording the matching relationship; The step of determining, for any target object to be recommended, a candidate user matching the target object includes: For any target object to be recommended, the recorded matching relationship between the target object and each user is obtained, and candidate users matching the target object are determined from among the users based on the obtained matching relationship.

3. The method according to claim 1, characterized in that The step of determining, for any target object to be recommended, candidate users matching the target object and the number of recommended users corresponding to the target object includes: For any target object to be recommended, determining a candidate user matching the target object; Based on the status information of the target object, the number of recommended users corresponding to the target object is determined.

4. The method according to claim 3, characterized in that Before determining the number of recommended users corresponding to the target object based on the state information of the target object, the method further includes: Based on the recommendation feedback information of the target object, the status information of the target object is determined, wherein the recommendation feedback information includes at least one of exposure information, click information and conversion information.

5. The method according to claim 3, characterized in that: The determining, based on the state information of the target object, the number of recommended users corresponding to the target object includes: Determine the user type corresponding to the target object according to a preset period; Within the user range covered by the user type, based on the status information of the target object, the number of recommended users corresponding to the target object is determined.

6. The method according to any one of claims 1 to 5, characterized in that: The matching relationship between an object and a user includes: a matching index indicating whether the object matches the user, and a matching degree indicating the object and the user; constructing global matching information of the object to be recommended and the user based on the matching relationship between the target object and the candidate user, and optimizing and analyzing the global matching information with the number of target users allocated to the target object reaching the number of recommended users as a constraint condition, and determining the target recommended object for each user, including: Based on the matching index and the matching degree in the matching relationship between the target object and the candidate user, construct global matching information between the object to be recommended and the user; Taking the number of target users allocated to the target object reaching the number of recommended users as a constraint condition, adjusting the matching index until the global matching information is greater than a target threshold; Based on the adjusted matching index, a target recommendation object for each user is determined.

7. The method according to any one of claims 1 to 5, characterized in that: Also includes: Determining a recommendation weight of the candidate user based on the behavior information corresponding to the candidate user; The constructing global matching information of the object to be recommended and the user based on the matching relationship between the target object and the candidate user includes: Based on the matching relationship between the target object and the candidate user, and the recommendation weight of the candidate user, global matching information between the object to be recommended and the user is determined.

8. The method according to any one of claims 1 to 5, characterized in that: Also includes: Obtaining a recommendation object threshold for the candidate user; The optimizing analysis of the global matching information with the number of target users allocated to the target object reaching the number of recommended users as a constraint condition includes: The global matching information is optimized and analyzed with the constraint that the number of target users assigned to the target object reaches the number of recommended users and the number of objects assigned to candidate users does not exceed the recommended object threshold.

9. The method according to claim 2, characterized in that: Before determining the matching relationship between each object and each user based on the object information of the plurality of objects to be recommended and the user information of the plurality of users, the method further includes: For any object among the multiple objects to be recommended, obtaining multi-dimensional association information of the any object; Standardized object information describing any one of the objects is filtered from the multi-dimensional association information.

10. The method according to claim 2, characterized in that Before determining the matching relationship between each object and each user based on the object information of the plurality of objects to be recommended and the user information of the plurality of users, the method further includes: Dividing the user set into a plurality of subsets, wherein the subsets include user information of a plurality of users; The determining the matching relationship between each object and each user based on the object information of the multiple objects to be recommended and the user information of the multiple users includes: For each subset in the plurality of subsets, a corresponding solver is called to determine a matching relationship between each object and each user based on object information of the plurality of objects to be recommended and information of the plurality of users in the subset.

11. The method according to claim 10, characterized in that The step of constructing global matching information of the object to be recommended and the user based on the matching relationship between the target object and the candidate user, and optimizing and analyzing the global matching information with the number of target users allocated to the target object reaching the number of recommended users as a constraint condition to determine the target recommended object for each user includes: The solver is called to construct global matching information of the object to be recommended and the user based on the matching relationship between the target object and the candidate user, and the number of target users allocated to the target object reaches the number of recommended users as a constraint condition, the global matching information is optimized and analyzed to determine the target recommended object for each user; Integrate the solution results of each solver to determine the target recommendation object for each user.

12. An object recommendation device, characterized in that: include: A determination module, for determining, for any target object to be recommended, candidate users matching the target object and the number of recommended users corresponding to the target object; The analysis module is used to construct global matching information between the object to be recommended and the user based on the matching relationship between the target object and the candidate user, and optimize the global matching information with the number of target users allocated to the target object reaching the number of recommended users as a constraint condition to determine the target recommended object for each user.

13. An object recommendation system, characterized in that: include: A server device and multiple terminal devices, wherein the multiple terminal devices correspond to multiple users respectively; The server device is used to execute the method according to any one of claims 1 to 11, determine a target recommended object for each user corresponding to the terminal device, and send information of the target recommended object to the terminal device; The terminal device is used to display the received information of the target recommended object.

14. A computing device, characterized in that include: Memory and processor; The memory is used to store computer programs / instructions, and the processor is used to execute the computer programs / instructions. When the computer programs / instructions are executed by the processor, the steps in any one of the methods described in claims 1 to 11 are implemented.

15. A computer-readable storage medium, characterized in that: A computer program / instruction is stored, and when the computer program / instruction is executed by a processor, the steps in any one of the methods described in claims 1 to 11 are implemented.

16. A computer program product, characterized in that The method comprises a computer program / instruction, and when the computer program / instruction is executed in a processor, the steps in any one of the methods of claims 1 to 11 are implemented.

Images