Game props recommendation method, device, computer equipment, storage medium and product

By dividing hierarchical levels based on game skill information and recommending prop usage information of high-level objects, the problems of low efficiency and high cost of game prop recommendation in the existing technology are solved, and efficient and low-cost prop recommendations are achieved, especially the problem of cold props is solved. Startup problem.

CN114470791BActive Publication Date: 2024-11-26TENCENT TECHNOLOGY (SHENZHEN) CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202210135377.2
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2022-02-14
Publication Date
2024-11-26
Estimated Expiration
2042-02-14

AI Technical Summary

Technical Problem

The existing game prop recommendation method is inefficient and costly, requiring complex feature engineering and a deep understanding of game logic, making it difficult to solve the problem of cold start of props.

Method used

Determine the target hierarchy based on the game skill information of the target object, divide multiple levels, and recommend the prop usage information of high-level objects based on the hierarchy and feature similarity, recommend to low-level objects, isolate the game logic and prop recommendation process, and reduce labor costs .

Benefits of technology

It improves the efficiency and practicality of prop recommendations, solves the problem of cold start of props, has strong reusability and versatility, and can effectively recommend the latest props.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN114470791B_ABST
    Figure CN114470791B_ABST
Patent Text Reader

Abstract

The present application provides a method, device, computer equipment, storage medium and product for recommending game props, and relates to the technical fields of computer networks, games, etc. By determining the target level of the target object based on the game skill information of the target object, the object is divided into multiple levels with different game skills; and based on the target level and the object characteristics of the target object, a first object set is determined, so as to determine a first object set with a higher level than the target object and similar to the target object in combination with the level, object characteristics, etc. of the object; and subsequently further determine a recommended set of game props used by the object including the first object set, so as to recommend the game props in the recommended set to the target object; so that low-level objects can learn the prop usage habits of objects that are similar to him and have better game skills than him. And there is no need to deeply analyze the game logic and complex feature engineering, which improves the practicality of game prop recommendation.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present application relates to the technical fields of computer networks, games, etc., and the present application relates to a method, device, computer equipment, storage medium and product for recommending game props. Background Art

[0002] With the rapid development of the game industry, various types of games are emerging one after another, and the competition in the game industry is becoming increasingly fierce. Many of these games provide game props for virtual objects representing the user's image. Game props are virtual items used in the game. In order to increase the user's purchase rate of game props, many game platforms will actively recommend game props.

[0003] In the related art, game props recommendation can include: one is to manually configure a fixed prop list and recommend the same prop list to each object. The other is based on the user's interest characteristics and purchase behavior data, usually using a model to build a complex feature engineering, after a large amount of data mining and calculation, matching the game props that the user is interested in, so as to recommend the game props of interest.

[0004] When using the above schemes for item recommendation, either the item lists of all users are the same, and users have low purchasing desire, which means that the actual recommendation efficiency is low; or complex feature engineering and large amount of calculation are required, and algorithm engineers need to have an in-depth understanding of the advantages and disadvantages of each item in the game, which requires a lot of manpower to analyze, resulting in high recommendation costs. Therefore, the practicality of the game item recommendation method in the related art is poor. Summary of the invention

[0005] The present application provides a game props recommendation method, device, computer equipment, storage medium and product, which can solve the problem of poor practicality of game props recommendation methods in related technologies. The technical solution is as follows:

[0006] In one aspect, a method for recommending game props is provided, the method comprising:

[0007] Determining a target level of the target object based on the game skill information of the target object;

[0008] Determining a first object set based on the target level and the object features of the target object, wherein the level of the objects in the first object set is higher than the target level, and the feature similarity between the objects in the first object set and the target object meets the target condition;

[0009] Based on the prop usage information of the first object set, a recommendation set of the target object is determined, and game props in the recommendation set are recommended to the target object.

[0010] On the other hand, a game prop recommendation device is provided, the device comprising:

[0011] A level determination module, used to determine a target level of the target object based on the game skill information of the target object;

[0012] An object set determination module, configured to determine a first object set based on the target level and the object features of the target object, wherein the level of the objects in the first object set is higher than the target level, and the feature similarity between the objects in the first object set and the target object meets the target condition;

[0013] The recommendation module is used to determine the recommended set of the target object based on the prop usage information of the first object set, and recommend the game props in the recommended set to the target object.

[0014] On the other hand, a computer device is provided, including a memory, a processor, and a computer program stored in the memory, wherein the processor executes the computer program to implement the above-mentioned game prop recommendation method.

[0015] On the other hand, a computer-readable storage medium is provided, on which a computer program is stored, and when the computer program is executed by a processor, the above-mentioned game prop recommendation method is implemented.

[0016] On the other hand, a computer program product is provided, including a computer program, wherein when the computer program is executed by a processor, the game prop recommendation method described above is implemented.

[0017] The beneficial effects of the technical solution provided by the embodiment of the present application are:

[0018] The game props recommendation method provided by the embodiment of the present application determines the target level of the target object based on the game skill information of the target object, thereby dividing the object into multiple levels with different game skills; and determines the first object set based on the target level and the object characteristics of the target object, thereby determining the first object set with a higher level than the target object and similar to the target object in combination with the level, object characteristics, etc. of the object; and subsequently further determines a recommended set of game props used by the first object set to recommend the game props in the recommended set to the target object; so that the low-level object learns the prop usage habits from the high-level object, and since the high-level object is more proficient in the game skills than the low-level object, the possibility of converting the recommended props into effective props is increased, thereby improving the actual recommendation efficiency. In addition, there is no need to deeply analyze the game logic and the advantages and disadvantages of each prop, and the commonly selected props of the feature-similar objects of the higher level can be adaptively recommended based on the level and feature similarity, and there is no need for complex feature engineering and a deep understanding of the game logic, which greatly saves labor costs. By isolating the game logic, game props and game prop recommendation process of different games, and without the need for personalized machine learning models, the method of the present application can be reused in any game for prop recommendation, with strong reusability, versatility and portability. Even for the latest game props, the recommendation method of using props can be learned from the low level to the high level, thereby increasing the possibility of new props being recommended and purchased, solving the prop cold start problem well, and thus improving the practicality of game prop recommendations. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings required for use in describing the embodiments of the present application are briefly introduced below.

[0020] Figure 1 A schematic diagram of an implementation environment of a game prop recommendation method provided in an embodiment of the present application;

[0021] Figure 2 A flowchart of a method for recommending game props provided in an embodiment of the present application;

[0022] Figure 3 A schematic diagram of object layering provided in an embodiment of the present application;

[0023] Figure 4 A schematic diagram of a game prop recommendation process provided in an embodiment of the present application;

[0024] Figure 5 A schematic diagram of the structure of a game prop recommendation device provided in an embodiment of the present application;

[0025] Figure 6A schematic diagram of the structure of a computer device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0026] The embodiments of the present application are described below in conjunction with the drawings in the present application. It should be understood that the implementation methods described below in conjunction with the drawings are exemplary descriptions for explaining the technical solutions of the embodiments of the present application and do not constitute a limitation on the technical solutions of the embodiments of the present application.

[0027] It will be understood by those skilled in the art that, unless specifically stated, the singular forms "one", "said", and "the" used herein may also include plural forms. It should be further understood that the terms "including" and "comprising" used in the embodiments of the present application refer to that the corresponding features can be implemented as the presented features, information, data, steps, operations, elements and / or components, but do not exclude the implementation as other features, information, data, steps, operations, elements, components and / or combinations thereof supported by the technical field. It should be understood that when we say that an element is "connected" or "coupled" to another element, the one element may be directly connected or coupled to the other element, or it may refer to that the one element and the other element establish a connection relationship through an intermediate element. In addition, the "connection" or "coupling" used herein may include wireless connection or wireless coupling. The term "and / or" used herein indicates at least one of the items defined by the term, for example, "A and / or B" may be implemented as "A", or as "B", or as "A and B".

[0028] It is understandable that in the specific implementation of the present application, any user-related data such as game level, resource value of trading items, transaction records, object online time, object registration time, event field parameter times, etc., when the following embodiments of the present application are applied to specific products or technologies, user permission or consent is required, and the collection, use and processing of relevant data need to comply with relevant laws, regulations and standards of relevant countries and regions.

[0029] In order to make the objectives, technical solutions and advantages of the present application clearer, the implementation methods of the present application will be further described in detail below with reference to the accompanying drawings.

[0030] The game props recommendation method provided in the embodiment of the present application involves the following artificial intelligence technology, computer vision technology, big data, cloud games, etc. For example, cloud computing technology in artificial intelligence technology can be used to grade each object based on the game skill information of each object in the game; of course, distributed databases, distributed file systems and other technologies in big data technology can also be used to store multiple data such as the level, characteristics, game skill information, and prop usage information of each object in the game. For example, computer vision technology can also be used to display the recommended game props on the game page. For example, each object can use cloud game technology to play online games on a web page, or it can use a game platform or an independent game application that provides multiple game entrances to play games, which is not limited to the embodiment of the present application.

[0031] Artificial Intelligence (AI) is the theory, method, technology and application system that uses digital computers or machines controlled by digital computers to simulate, extend and expand human intelligence, perceive the environment, acquire knowledge and use knowledge to obtain the best results. In other words, artificial intelligence is a comprehensive technology in computer science that attempts to understand the essence of intelligence and produce a new intelligent machine that can respond in a similar way to human intelligence. Artificial intelligence is to study the design principles and implementation methods of various intelligent machines so that machines have the functions of perception, reasoning and decision-making.

[0032] Artificial intelligence technology is a comprehensive discipline that covers a wide range of fields, including both hardware-level and software-level technologies. Basic artificial intelligence technologies generally include sensors, dedicated artificial intelligence chips, cloud computing, distributed storage, big data processing technology, operating / interactive systems, mechatronics and other technologies. Artificial intelligence software technologies mainly include computer vision technology, speech processing technology, natural language processing technology, as well as machine learning / deep learning, autonomous driving, smart transportation and other major directions.

[0033] Computer Vision (CV) is a science that studies how to make machines "see". To put it more specifically, it means using cameras and computers to replace human eyes to identify and measure targets, and further perform graphic processing so that the computer processes the images into images that are more suitable for human observation or transmission to instruments for detection. As a scientific discipline, computer vision studies related theories and technologies, and attempts to establish an artificial intelligence system that can obtain information from images or multidimensional data. Computer vision technology usually includes image processing, image recognition, image semantic understanding, image retrieval, OCR, video processing, video semantic understanding, video content / behavior recognition, three-dimensional object reconstruction, 3D technology, virtual reality, augmented reality and other technologies.

[0034] Big data refers to a collection of data that cannot be captured, managed, and processed by conventional software tools within a certain time frame. It is a massive, high-growth, and diverse information asset that requires new processing models to have stronger decision-making power, insight discovery, and process optimization capabilities. With the advent of the cloud era, big data has also attracted more and more attention. Big data requires special technologies to effectively process large amounts of data within a tolerable time frame. Technologies applicable to big data include large-scale parallel processing databases, data mining, distributed file systems, distributed databases, cloud computing platforms, the Internet, and scalable storage systems.

[0035] Cloud gaming, also known as gaming on demand, is an online gaming technology based on cloud computing technology. Cloud gaming technology enables thin clients with relatively limited graphics processing and data computing capabilities to run high-quality games. In a cloud gaming scenario, the game is not played on the player's game terminal, but on a cloud server, which renders the game scene into a video and audio stream and transmits it to the player's game terminal over the network. The player's game terminal does not need to have powerful graphics computing and data processing capabilities, but only needs to have basic streaming media playback capabilities and the ability to obtain player input commands and send them to the cloud server.

[0036] Figure 1 A schematic diagram of the implementation environment of a game prop recommendation method provided by this application. Figure 1 As shown, the implementation environment includes: a computer device 101 and a terminal 102. The computer device 101 can be a background server of a game application or a cloud server supporting a game scene, etc. The terminal 102 can be a terminal of a target object. The computer device 101 can exchange data with the terminal 102 to send the game props recommended for the target object to the terminal 102.

[0037] In one possible scenario example, the terminal 102 may be installed with a game application, and the computer device 101 sends recommended game props to the terminal 102 based on the game application. The terminal 102 may display the recommended game props such as icons and names on the page of the game application.

[0038] In another scenario example, the computer device 101 and the terminal pole 102 can exchange data based on cloud gaming technology. For example, the terminal 102 can display the online game page of the cloud game, and the computer device 101 can send game screen rendering data including props to be recommended to the terminal 102. Of course, this application only uses the above two examples as examples, and the embodiments of the present application do not limit the scenario. For example, the computer device 101 can also send props to be recommended to the terminal 102 based on the game live broadcast application, and the terminal 102 pops up a recommendation page including props to be recommended in the page of the game live broadcast application, or displays the URL link corresponding to the recommendation page.

[0039] The target object is used to represent the user's virtual image in the game. For example, the target object can be a game character in the game. A user can set at least one virtual image in a game. The game props can be virtual items that interact with objects in the game and have an impact on the properties of objects in the game. Users can purchase game props through trading resources. The trading resources can be tools for measuring the virtual value of game props and media for purchasing game props. For example, trading resources can be virtual media that can be used to trade props in the game, such as game coins, virtual gold coins, virtual diamonds, etc.

[0040] The game applications include but are not limited to: shooting games, multiplayer online tactical competitive games, multiplayer gunfight survival games, role-playing games, real-time strategy games, racing games, music games, chess and card games, etc. Game props include but are not limited to: attack props, defense props, wearable props, virtual appliances, virtual medicines, virtual ammunition, virtual keys, virtual cards, fragments, etc.

[0041] The server can be an independent physical server, or a server cluster or distributed system composed of multiple physical servers, or a cloud server or server cluster that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communications, middleware services, domain name services, security services, CDN (Content Delivery Network), and big data and artificial intelligence platforms. The above-mentioned network may include but is not limited to: wired network, wireless network, wherein the wired network includes: local area network, metropolitan area network and wide area network, and the wireless network includes: Bluetooth, Wi-Fi and other networks that realize wireless communication. The terminal can be a smart phone (such as Android mobile phone, iOS mobile phone, etc.), intelligent voice interaction device, smart home appliance, vehicle terminal, tablet computer, laptop computer, digital broadcast receiver, MID (Mobile Internet Devices), PDA (personal digital assistant), desktop computer, vehicle terminal (such as vehicle navigation terminal, vehicle computer, etc.), smart speaker, smart watch, etc. The terminal and the server can be directly or indirectly connected by wired or wireless communication, but it is not limited to this. It can also be determined based on the actual application scenario requirements, and is not limited here.

[0042] Figure 2 The flowchart of a method for recommending game props provided in an embodiment of the present application is shown in FIG. The execution subject of the method may be a computer device. Figure 2 As shown, the method includes the following steps.

[0043] Step 201: The computer device determines a target level of the target object based on the game skill information of the target object.

[0044] The game skill information is used to indicate the proficiency of the game skills of the target object. The computer device can obtain the game skill information of the target object, and determine the target level corresponding to the game skill information according to the proficiency of the game skills of the target object indicated by the game skill information. The higher the proficiency of the game skills of the target object, the higher the target level. Exemplarily, the computer device can hierarchically divide each object according to the hierarchical division rules based on the game skill information of each object. The hierarchical division rules include: the more proficient the game skills of the object in the game, the higher the level corresponding to the object. The level of the object can be represented by a numerical value, and the level can change with the change of the size of the numerical value. For example, the first level is higher than the second level, the second level is higher than the third level, and so on; for example, the first level is the highest level, and the game skills of the objects in the first level are the most proficient; the second level is the second highest level, and the game skills of the objects in the second level are second, and so on.

[0045] In one possible implementation, the game skill information is information indicating the proficiency in mastering and applying game logic; the higher the proficiency of the target object in mastering and applying game logic, the higher the target level of the target object; the game skill information includes at least one of the game level, resource value of trading items, object online time, object registration time, and number of activity participations; accordingly, in this step, the process of the computer device determining the target level of the target object based on the game skill information of the target object can be implemented by at least one of the following corresponding methods.

[0046] In a first manner, the game skill information includes a game level. In response to the game level of the target object being in a target level interval, the computer device determines that the object level of the target object is a target level corresponding to the target level interval.

[0047] The game level may be the level of the target object in the game, and a higher level indicates that the target object has more skilled game skills and better grasp of the game logic in the game. The computer device may extract the game level of the target object from the object data of the target object.

[0048] The computer device may pre-store an association relationship between a level interval and a level of an object. The computer device may determine a target level interval in which the game level of the target object is located, and based on the target level interval, obtain the level corresponding to the target level interval from the association relationship between the level interval and the level of the object, and use the obtained level as the target level of the target object.

[0049] In one possible example, the computer device may divide the levels of various objects in the game based on the game level. Exemplarily, the computer device may divide objects belonging to the same level interval into the same level based on the level interval. It should be noted that the higher the game level, the more proficient the game skills of the target object, and the higher the corresponding target level, and the more core the object. For example, the first level is the highest level, the second level is the second highest level, and each game level in the level interval corresponding to the first level is greater than the game level in the level interval corresponding to the second level, and so on.

[0050] In one possible example, the computer device may use the inflection point position in the level distribution curve to determine the level interval corresponding to each level. Exemplarily, the process of the computer device determining the association relationship between the level interval and the level of the object includes: the computer device may determine the level distribution curve corresponding to the game based on the game level of each object in the game, and the level distribution curve includes the change relationship between the game level and the number of objects corresponding to it; the computer device determines multiple level intervals based on the inflection point position of the level distribution curve, and associates the multiple level intervals with multiple levels respectively based on the level included in the multiple level intervals; wherein, the higher the level in the level interval, the higher the level corresponding to the level interval. For example, the inflection point may be an extreme point, a maximum point, or a minimum point of the curve. For example, the level distribution curve may be a curve in a two-dimensional coordinate system with the game level as the horizontal axis and the number of objects as the vertical axis. Then the computer device may use the horizontal coordinate corresponding to the inflection point position in the level distribution curve as the endpoint of the level interval to be divided, and obtain multiple level intervals in sequence.

[0051] In a second manner, the game skill information includes a resource value of a trading item. In response to the resource value of the trading item of the target object being in a target resource interval, the computer device determines that the object level of the target object is the target level corresponding to the target resource interval.

[0052] The resource value of the trading prop can be the amount of trading resources consumed by the target object when trading props in the game. The larger the resource value of the trading prop, the more and more advanced the trading props are. The higher the prop trading degree of the target object, the greater the corresponding trading potential, and the more proficient the corresponding game skills. The computer device can extract the transaction record of the target object and count the resource value of the trading prop of the target object. It should be noted that the transaction record, the resource value of the trading prop and other data are all obtained after the user's consent.

[0053] The computer device may pre-store an association relationship between a resource interval and an object's hierarchy. The computer device may determine a target resource interval in which the resource value of the target object's trading item is located, and based on the target resource interval, obtain the hierarchy corresponding to the target resource interval from the association relationship between the resource interval and the object's hierarchy, and use the obtained hierarchy as the target hierarchy of the target object.

[0054] In one possible example, the computer device can divide the levels of various objects in the game based on the resource values ​​of the trading props. It should be noted that the larger the resource value of the trading props, the more or more advanced game props the target object uses when playing the game, the more proficient the corresponding game skills are, the higher the corresponding target level is, and the more core the object is. For example, the first level is the highest level, the second level is the second highest level, and the resource values ​​in the resource interval corresponding to the first level are greater than the resource values ​​in the resource interval corresponding to the second level, and so on.

[0055] In one possible example, the computer device may use the inflection point position in the resource distribution curve of the trading props to determine the resource intervals corresponding to each level. Exemplarily, the process of the computer device determining the association relationship between the resource interval and the level of the object includes: the computer device may determine the resource distribution curve corresponding to the game based on the resource values ​​of the trading props of each object in the game, and the resource distribution curve includes the change relationship between the resource values ​​of the trading props and the number of objects corresponding to them; the computer device determines multiple resource intervals based on the inflection point position of the resource distribution curve, and associates the multiple resource intervals with multiple levels respectively based on the size of the resource values ​​included in the multiple resource intervals; wherein, among the multiple resource intervals, the larger the resource value in the resource interval, the higher the level corresponding to the resource interval. wherein, the process of the computer device using the resource distribution curve to obtain the association relationship between the resource interval and the level is similar to the above-mentioned first method of using the level distribution curve to obtain the association relationship between the level interval and the layer, and will not be repeated here.

[0056] In a third mode, the game skill information includes the online duration of the object. In response to the online duration of the target object being within a target duration interval, the computer device determines that the object level of the target object is the target level corresponding to the target duration interval.

[0057] The online time of the object can be the cumulative time that the target object has been online at least once in a specified period. The specified period can be the past month, the past week, etc. A large value of the online time of the object indicates that the longer the target object has been playing the game, the more active the target object is in the game, and the more skilled the corresponding game skills are; for example, object A has been online for 2 hours in the past month and object B has been online for 50 hours in the past month. Obviously, object B has been playing the game more and has more skilled game skills. The computer device can extract the online records of the target object in the specified period, and count the cumulative time of at least one online time in the specified period. It should be noted that the online records, object online time and other data are all obtained with the user's consent.

[0058] The computer device can pre-store an association relationship between duration intervals and levels of objects. The computer device can determine a target duration interval in which the online duration of the target object is located, and based on the target duration interval, obtain the level corresponding to the target duration interval from the association relationship between the duration interval and the level of the object, and use the obtained level as the target level of the target object.

[0059] In one possible example, the computer device can divide the levels of each object in the game based on the online time of the object. It should be noted that the longer the online time of the object, the more game time the target object has, the more skilled the game skills mastered in more game time, the higher the target level, and the more core the object. For example, the value of each online time in the time interval corresponding to the first level is greater than the value of the online time in the time interval corresponding to the second level, and so on. In one possible example, the computer device can use the inflection point position in the online time distribution curve to determine the time interval corresponding to each level. This process is the same as the process of determining the corresponding interval based on the inflection point position in the curve in the first and second methods mentioned above, and will not be repeated here.

[0060] In a fourth manner, the game skill information includes an object registration time. In response to the object registration time of the target object being in a target time interval, the computer device determines that the object level of the target object is a target level corresponding to the target time interval.

[0061] The object registration time can be the time when the user registers the game account in the game, the time when the game is first launched, or the time when the user configures the target object in the game. The earlier the object registration time, the longer the target object or the user represented by the target object has been in contact with the game, the richer the game experience accumulated by the target object, and the more proficient the corresponding game skills. It should be noted that the object registration time is obtained with the user's consent.

[0062] The computer device may pre-store an association relationship between a time interval and a hierarchy of an object. The computer device may determine a target time interval in which the object registration time of the target object is located, and based on the target time interval, obtain the hierarchy corresponding to the target time interval from the association relationship between the time interval and the hierarchy of the object, and use the obtained hierarchy as the target hierarchy of the target object.

[0063] The earlier the object registration time is, the longer the target object's game time is, the more proficient the corresponding game skills are, the higher the corresponding target level is, and the more core the object is. For example, the first level is the highest level, the second level is the second highest level, and each time in the time interval corresponding to the first level is earlier than the time in the time interval corresponding to the second level, and so on. In one possible example, the computer device can use the inflection point position in the registration time distribution curve to determine the time interval corresponding to each level. This process is the same as the process of determining the corresponding interval based on the inflection point position in the curve in the first and second methods mentioned above, and will not be repeated here.

[0064] In a fifth manner, the game skill information includes the number of times the target object participates in an activity. In response to the target object's number of times the target object participates in the activity being within a target number interval, the computer device determines that the object level of the target object is the target level corresponding to the target number interval.

[0065] The number of activity participations may be game-related activities that the user has participated in within a specified period. For example, the number of activity parameters within the past month or the past week. Exemplarily, the game may be a specified game. The game-related activities are not limited to online and offline activities; for example, they may be online activities where you get rewards for completing tasks, free point card activities, discounted props promotion activities, etc.; they may also be live activities provided by offline live game booths. Exemplarily, the more times an activity is participated in, the more active the target object or the user represented by the target object is in the game, the more interested they are in the game, and the more proficient their corresponding game skills are. It should be noted that the number of activity participations is obtained with the user's consent.

[0066] The computer device may pre-store an association relationship between frequency intervals and object levels. The computer device may determine a target frequency interval in which the target object's activity participation times are located, and based on the target frequency interval, obtain the level corresponding to the target frequency interval from the association relationship between the frequency interval and the object level, and use the obtained level as the target level of the target object.

[0067] The more times an activity is participated in, the higher the interest of the target object in the game, the more proficient the corresponding game skills are, the higher the corresponding target level is, and the more core the object is. For example, the first level is the highest level, the second level is the second highest level, and the value of the number of times each activity is participated in in the number interval corresponding to the first level is greater than the value of the number of times each activity is participated in in the number interval corresponding to the second level, and so on. In one possible example, the computer device can use the inflection point position in the activity participation frequency distribution curve to determine the number interval corresponding to each level. This process is the same as the process of determining the corresponding interval based on the inflection point position in the curve in the first and second methods mentioned above, and will not be repeated here.

[0068] It should be noted that the above five methods are methods for determining the target level based on one type of information. The computer device may also combine any two or more of the above five methods to determine the target level. In one example, an example is given by combining the above first and second methods. The computer device may be configured with level intervals and resource intervals corresponding to each level. The computer device may determine the target level interval where the game level is located and the target resource interval where the resource value of the trading item is located based on the game level of the target object and the resource value of the trading item, and determine the target level corresponding to the target level interval and the target resource interval according to the target level interval and the target resource interval.

[0069] In another example, an example is given by combining the first and second methods mentioned above. The computer device can also calculate the scores corresponding to each information, and determine the target level based on the scores corresponding to at least two types of information. The computer device is configured with an association relationship between multiple levels and multiple score intervals. The computer device can also calculate the first score of the target object based on the weight of the game level and the level score corresponding to the game level of the target object; and calculate the second score of the target object based on the weight of the resources of the trading props and the score corresponding to the resource value of the trading props of the target object; the sum of the first score and the second score is used as the total score of the target object; and the level corresponding to the target score interval where the total score is located is used as the target level of the target object. It should be noted that the implementation process of the combination of the first and second methods is only illustrated by the above two examples, but the actual method of combining the above first and second methods is not limited. It should be noted that the above is only an example of the combination of the first and second methods. Of course, for any other combination of the above five methods, the corresponding implementation process is the same as the implementation process of the combination of the first method and the second method as exemplified above, and will not be described one by one here.

[0070] Figure 3 A schematic diagram of an object hierarchy provided in an embodiment of the present application. Figure 3As shown, the computer device can stratify the objects in the game and divide out high-level and highly active users. Exemplary, for example, it can be stratified according to the game level, the resource value of the trading props, the object online time, the object registration time, and the number of times the activity is participated in, or it can be stratified according to the frequency of the trading props, and multiple objects can be divided into multiple different levels of core player levels. The core degree of users at different levels is different, and the smaller the number of layers, the more core users. For example, the highest level is the first level for the most core players, followed by the second level for core players, the third level for sub-core players, the fourth level for sub-core players, and the fifth level for active players. In addition, the payment, game level, and online time lower than the fifth level of inactive (for example, the number of logins is less than 3 times, the payment is 0, etc.) can be divided into other players. Among them, the low-level B group can learn the prop usage habits of the high-level A group.

[0071] By determining the target level of the target object based on the game skill information, the hierarchical division of each object in the game based on the game skill is achieved, and props recommendations for different levels are achieved based on subsequent steps, and the low-level group learns the use of game props from the high-level group. There is no need for complex feature engineering or a deep understanding of the game logic, which makes the prop recommendation process of this application highly reusable. Furthermore, the game skill information includes at least one of the game level, the resource value of the trading props, the object's online time, the object registration time, and the number of activity participations; using one or more of the information included in the game skill information to accurately locate the level of each object, the level of the object can accurately measure the object's skill proficiency, further improving the accuracy of prop recommendations.

[0072] Step 202: The computer device determines a first object set based on the target level and the object characteristics of the target object.

[0073] The level of the objects in the first object set is higher than the target level, and the feature similarity between the objects in the first object set and the target object meets the target condition. The computer device determines the second object set based on the target level, and the level of the objects in the second object set is higher than the target level; the computer device screens out the first object set whose feature similarity with the target object meets the target condition from the second object set based on the features of the target object. Wherein, the target condition may include that the feature similarity is not lower than the target similarity threshold. Wherein, the computer device may determine the feature similarity between the target object and each object in the second object set based on the features of the target object and the objects in the second object set; the computer device screens out the first object set from the second object set based on the feature similarity between the target object and each object in the second object set. Exemplarily, the computer device may obtain the feature vector of the target object based on the object data of the target object. For example, the computer device may use a trained feature extraction network to input the object data of the target object into the feature extraction network to obtain the feature vector output by the feature extraction network. The feature vector can represent the characteristics of the target object in multiple dimensions. The object data includes but is not limited to: the target object's game skill information, wins and losses in each game, team information during the game, etc. It should be noted that the object data is obtained after the user's authorization and consent.

[0074] In one possible implementation, before determining the first set of objects based on the object features of the target level and the target object, the computer device may also first determine the features of each object in the game application, so as to determine a set of objects with similar features based on the features of the object in step 202. Exemplarily, the process of determining the features of each object may include: the computer device determines the features of each object based on team information of at least two objects in the game application. Exemplarily, the team information of each object may include a teammate object of the object, and the teammate object may be an object that is a teammate of the object in the game team to which the object belongs. The computer device may learn the team features of each object based on the team information of at least two objects to obtain the features of each object.

[0075] In one possible example, the computer device may perform machine learning in a graph manner to obtain the characteristics of each object. Exemplarily, the process of determining the characteristics of each object based on the team information of at least two objects in the game application may be implemented by following steps S1 to S3.

[0076] Step S1: The computer device constructs an undirected graph based on the teaming information of the at least two objects.

[0077] The undirected graph includes at least two nodes corresponding to at least two objects, and there is an edge between two nodes corresponding to two objects that team up with each other in the at least two objects. Exemplarily, based on the team information of at least two objects in the game application, an undirected graph is constructed with at least two objects as nodes, wherein whether the objects team up is used as a basis for whether there is a change between the corresponding nodes. For example, when two objects have teamed up with each other in the past month, there is an edge between the two nodes corresponding to the two objects. Thus, a large undirected graph can be obtained, and the undirected graph may not include weights.

[0078] For example, the team formation information of all objects in a game application within one month can be used to construct an undirected graph G = (V, E); wherein V represents a node in the undirected graph, a node represents a corresponding object, and E represents an edge in the undirected graph.

[0079] Step S2: The computer device obtains at least two sequences from the undirected graph by random walk.

[0080] Each sequence includes objects arranged in the order of walking during the walk. After constructing the graph G, take any node as the starting node, and randomly select a node from multiple nodes that have edges with the previous node each time you walk until the maximum step length t is reached, that is, the number of walks reaches t times, and a sequence of length t is obtained. In addition, this truncated random walk method provides two advantages: on the one hand, local exploration walks are easy to parallelize, and several different walkers can walk multiple different routes at the same time. On the other hand, information is obtained from truncated random walks, and when the graph structure changes slightly, there is no need to repeat the learning.

[0081] A random walk can be a random walk of fixed length. The purpose of a random walk is to perform sampling work to collect training data. The principle of a random walk mainly includes: for each node in an undirected graph as the root, randomly find the next node to walk, and the walk length is a fixed t. Each node can have multiple walkers walking in parallel, which are hyperparameters that can be configured based on needs. After a random walk, multiple sequence samples will be obtained. These sequence samples can be used as training samples.

[0082] Step S3: The computer device inputs the at least two sequences into a target model, and performs unsupervised training on the target model to obtain the features of each object.

[0083] The computer device inputs the at least two sequences into the target model, trains the target model by unsupervised training, and obtains the characteristics of each object. Each sequence includes multiple objects arranged in a walk order. In one possible example, the step of obtaining at least two sequences from the undirected graph by random walk may include: for each node in the undirected graph, the computer device performs a random walk based on the target walk number and the walk step length, taking each node as the starting point, to obtain the at least two sequences. Accordingly, the computer device inputs the at least two sequences into the target model, and performs unsupervised training on the target model, and the step of obtaining the features of each object may include: the computer device obtains the initialization representation matrix of the at least two objects; for each sequence obtained by each walk, the computer device traverses at least two subsequences included in each sequence based on the target window size, with each node in each sequence as the center point, each subsequence includes the center point and the adjacent nodes within the target window size range of the center point; for each sequence, the following steps are performed: for each subsequence of each sequence, the computer device calculates the value of the target loss function based on each subsequence and the center point of each subsequence, and optimizes the initialization representation matrix based on the value of the target loss function, until the value of the loss function reaches the target optimization condition, the optimization is stopped, and the representation matrix of at least two objects is obtained. The target optimization condition may include that the value of the loss function does not exceed the target optimization threshold, is within the target optimization threshold interval, etc. The characterization matrix may be a node characterization matrix Φ, where Φ is a matrix of |V|*d, and each node has a d-dimensional vector, that is, each object corresponds to a d-dimensional Embedding vector. Among them, the target loss function may be a function that characterizes conditional probability; the target loss function is used to characterize the conditional probability of each subsequence appearing under the condition that the center point of each subsequence appears. Then the process in which the computer device calculates the value of the target loss function based on each subsequence and the center point of each subsequence may include: the computer device calculates the conditional probability of the subsequence appearing under the condition that the center point appears based on each subsequence and the center point of each subsequence, and the computer device obtains the value of the target loss function based on the conditional probability.

[0084] The feature of each object can be the Embedding vector of the object. Among them, the process in which the computer device determines the feature similarity between the target object and each object in the second object set can include: the computer device calculates the cosine similarity between the target object and each object in the second object set based on the Embedding vector of the target object and the Embedding vector of each object in the second object set, and uses the cosine similarity as the feature similarity. Thereby, the cosine similarity is calculated for the Embedding vector of each object in each layer and the Embedding vector of each object in the upper layer higher than its level, and similar objects of each object in the object set higher than its level are obtained. Alternatively, the Euclidean distance between two Embedding vectors can also be calculated, and the Euclidean distance can be used as the feature similarity. Accordingly, the target condition can include but is not limited to: the Euclidean distance is not higher than the target distance threshold, the cosine similarity is not lower than the target similarity threshold, etc.

[0085] Random walk mainly takes each node in the undirected graph as the root and randomly finds the next node to walk. The walk length is fixed t. Each node can have several walkers walking in parallel. These are hyperparameters that can be configured based on needs. After random walk, multiple sequence samples will be obtained. These sequence samples can be used as training samples. The specific process of obtaining sequences based on random walks and using the model to generate embedding vectors is as follows:

[0086] Algorithm 1 DeepWalk(G,w,d,γ,t)

[0087] Input: graph G(V, E)

[0088] windowsizew

[0089] Embeddingsized

[0090] walkspervertexγ

[0091] walklengtht

[0092] Output: matrixofvertexrepresentationsΦ∈R ∣V∣×d

[0093] 1:Initialization:SampleФfrom U ∣V∣×d

[0094] 2: Build a binary Tree T from V

[0095] 3: for i = 0 to γ ​​do

[0096] 4: O = Shuffle (V)

[0097] 5: for each v i ∈O do

[0098] 6: Wv i = Random Walk(G,v i , t)

[0099] 7: SkipGram(Ф,Wv i , w)

[0100] 8: endfor

[0101] 9: endfor

[0102] Among them, "Algorithm1DeepWalk(G, w, d, γ, t)" means that the DeepWalk algorithm can be used for implementation, where G is a directed graph, V is a node set in the directed graph, and E is an edge in the graph. "Input: graph G(V, E); windowsize w; embeddingsize d; walkspervertexγ; walklength t;" means that the input of the algorithm is graph G(V, E), the window size is w, the dimension of the embedding vector is d, the number of random walks for each node is γ, and the step length of the random walk is t. "Output: matrixofvertexrepresentationsΦ∈R ∣V∣×d " indicates that the output node representation matrix Φ is a matrix of |V|*d. Each node has a d-dimensional vector, corresponding to the Embedding vector of each object. In the first step, the vector space of each node is initialized, and a U ∣V∣×d The initial representation matrix. Step 2: Establish a Huffman tree. The Huffman tree can be constructed based on the number of random walk nodes. Step 3: Indicates the loop walk from 0 to γ ​​times. Step 4: Disorder the node set V to get O. Step 5: Traverse each node in O. do indicates entering the loop. Step 6: Get the nodes from v i The node starts a random walk sequence with a step length of t. Step 7: Use the SkipGram algorithm to train to optimize the parameters, which can include the optimization of the node representation matrix Φ. Step 8: Exit the inner loop. Step 9: Exit the outer loop.

[0103] The SkipGram algorithm involved in the above algorithm flow is a training method of word2vec in the field of natural language processing. In the DeepWalk algorithm, random walks generate a large number of training samples, and then SkipGram is used for unsupervised training to obtain the embedding vector of each node. The specific process of the SkipGram algorithm is as follows:

[0104] Algorithm 2 SkipGram(Ф,Wv i , w)

[0105] 1: for each v j ∈Wv i do

[0106] 2: for each u k ∈Wv i [jw:j+w]do

[0107] 3: J(Ф)=-logPr(u k |Ф(v j ))

[0108] 4:

[0109] 5: endfor

[0110] 6: endfor

[0111] Among them, "Algorithm 2 SkipGram(Ф,Wv i ,w)” represents the SkipGram algorithm in Algorithm 2, Φ represents the output node representation matrix, Wv i represents the sequence generated by random walk, and w represents the window size. In the first step, the outer loop is used to generate the sequence Wv i Loop through each node in Wv i Each node v in the sequence i , enter the loop. Step 2, inner loop, operate on the sequence of window size w of each node: u k ∈Wvi[jw:j+w], traverse v i For each node with a distance w before and after the node, we get a subsequence u k , and enter the loop. Step 3, for each subsequence u k , calculate the value of the target loss function J(Ф) corresponding to each cycle, where Pr(u k |Ф(v j )) indicates that in Ф(v j ) appears under the condition that u k The probability that uk In v j The probability of context occurrence; the optimization goal here is to move towards u k In v j The greater the probability of occurrence under the condition of occurrence, that is, Pr(u k |Ф(v j )) is larger, that is, J(Ф) is closer to 0. Step 4, it means that the gradient descent method is used for iterative training, where α represents the learning rate. Step 5, exit the inner loop. Step 6, exit the outer loop.

[0112] It should be noted that the DeepWalk algorithm is used in the above example to calculate the Embedding vector of the object. In addition, you can also try other graph representation algorithms, such as LINE (Large-scale Information Network Embedding) algorithm, Node2Vec (Scalable Feature Learning for Networks) algorithm, Struc2Vec (Learning Node Representations from Structural Identity) algorithm, etc., conduct comparative experimental analysis and select the best algorithm.

[0113] By filtering out the first set of objects at a higher level than the target object based on the target level, we can make prop recommendations based on the prop usage habits of the higher-level objects. The higher the level of the players, the more proficient they are in the game logic, and the more reasonable the prop combinations they choose. For players at each level, they can learn the prop selection habits, prop combination habits, etc. of similar higher-level players, and then learn the game skills of higher-level players, further improving the practicality of prop recommendations.

[0114] Step 203: The computer device determines a recommended set of the target object based on the prop usage information of the first object set.

[0115] The computer device may determine a recommended set that meets the prop screening condition based on the prop usage information. Exemplarily, the prop usage information may include a prop usage record of the object. The prop screening condition at least includes: props used by each object in the first object set. The props used by the object may include, but are not limited to: currently used props, historically used props, etc.

[0116] In a possible implementation, the computer device may further filter the props of the objects in the first object set based on the props already owned by the target object, the prop transaction level of the target object, etc., to obtain a recommended set. Exemplarily, the prop screening conditions may also include: not exceeding the prop transaction level of the target object, not including the existing props of the target object, etc. Accordingly, step 203 may be implemented by following the steps 2031-2033.

[0117] Step 2031: The computer device determines a first recommendation set based on the prop usage information of each object in the first object set.

[0118] The computer device may obtain the props of each object in the first object set based on the prop usage information, and determine the first recommended set, which includes the props of each object in the first object set. Exemplarily, based on the number of times the props are used, the props commonly used by the objects in the first object set may be further screened out as the first recommended set. The prop screening condition may also include: the number of times used by each object in the first object set exceeds the target number threshold. Then the computer device may determine the props of each object in the first object set based on the prop usage information; and count the number of times the props of each object in the first object set are used, and screen out the first recommended set whose number of uses exceeds the target number threshold from the props of each object in the first object set. Exemplarily, the computer device may determine the props used by each object in the first object set within the target period based on the prop usage information; for example, props that have been used or are being used in the past month or the past week. Among them, the computer device can count the number of times each prop is used by at least one object in the first object set during the target period. For example, for a prop a used by object A in the first object set, the cumulative value of the number of times prop a is used by object A and other objects in the first object set in the past month can be counted; and the commonly used props of the first object set can be further obtained.

[0119] In one possible example, the item screening condition may also include: game props used by the objects in the first object set in each game in which the game result is a win. Then the computer device may also determine a first recommended set in combination with the game results of the objects in each game. Exemplarily, the computer device may also determine the first recommended set based on the prop usage information of each object in the first object set and the game results of each object in each game, and the first recommended set includes props used by the objects in the first object set in games in which the game result is a win. Of course, it is also possible to further screen out the first recommended set whose number of uses in each game in which the game result is a win exceeds the target number threshold; or, screen out the first recommended set whose number of uses in each game won within the target period exceeds the target number threshold, and so on, which will not be described one by one here.

[0120] Step 2032: The computer device determines the item trading level of the target object based on the resource value of the trading item of the target object.

[0121] The item transaction level refers to the target object's ability to trade game items; for example, the value of resources spent by the target object to purchase game items. In one possible example, the computer device can evaluate the target object's consumption level based on the value of resources spent by the target object when trading items at least twice. For example, the average, maximum, or median value of the game virtual resources spent by the target object on multiple purchases of game items in the past month is used as the target object's item transaction level.

[0122] In one possible example, the computer device can also predict the target object's transaction potential based on the target object's historical transaction records in the past, thereby further evaluating the consumption level. Then the computer device can count the resource values ​​of the target object's transaction items within a specified time, and determine the target object's transaction potential value based on the resource values ​​of the transaction items within the specified time; the computer device combines the transaction potential value with the resource values ​​of the transaction items within the specified time to determine the target object's item transaction level. The transaction potential value is used to represent the target object's potential ability to trade items, for example, the potential to purchase items. For example, it can be counted that the game coins purchased by player A each month in the past three consecutive months are 1800, 2000, and 2200, with a maximum value of 2200. The transaction potential value is also the potential increase in purchasing power of 200, and the maximum value is that the target object's consumption level in the fourth month can be 2400.

[0123] Step 2033: The computer device selects a second recommendation set from the first recommendation set based on the existing props of the target object and the transaction level of the props.

[0124] The second recommendation set does not include the existing props and the resource values ​​of the props in the second recommendation set do not exceed the prop transaction level. Based on the existing props, the computer device screens out a candidate set that does not include the existing props from the first recommendation set, and based on the resource values ​​of each prop in the candidate set, determines the props in the candidate set that do not exceed the prop transaction level as the second recommendation set. In one possible example, the computer device may also associate and store the association relationship between the second recommendation set and the target object, so as to subsequently recommend the game props in the second recommendation set to the target object. Among them, the existing props refer to props that the target object currently owns and can use directly; for example, props that have been purchased, given by friends, or already exist in the target object's prop library, etc.

[0125] Based on the prop usage information of the objects in the first object set, a recommended set including props used by the first object set is determined, thereby finding props used by players at a higher level than the target object for subsequent recommendation. Furthermore, certain prop screening conditions can be combined to further screen out props that better meet the needs of the target object, such as screening out props that do not exceed the consumption level of the target object, filtering out props that the target object already has, etc., so that the final second recommended set can better meet the prop needs of the target object, thereby improving the personalization and accuracy of prop recommendations.

[0126] Step 204: The computer device recommends the game props in the recommendation set to the target object.

[0127] The computer device may sort the props in the recommendation set and recommend them in a certain order. Alternatively, the computer device may recommend a certain number of game props in the recommendation set to the target object according to a preconfigured number of recommendations. In one possible implementation, the computer device may make recommendations based on the heat value of the props. Then step 204 may include: the computer device counts the heat values ​​of at least two props in the recommendation set in the first object set; the computer device arranges at least two props in the recommendation set in descending order based on the heat values ​​of the at least two props; the computer device recommends the game props in the recommendation set to the target object in descending order. For example, the computer device recommends the recommended number of game props in the recommendation set to the target object in descending order. Exemplarily, the heat value of the prop is used to indicate the popularity of the prop in the first object set. For example, the heat value may include the number of usage objects corresponding to the prop in the first object set. The recommendation set may be a second recommendation set, and the computer device may count the number of users of each item in the second recommendation set, and sort the items in the second recommendation set in descending order based on the number of users. The number of users may be the number of objects that use the item in the first object set; for example, if item A has been used by 10 players in the past month, the number of users is 10.

[0128] In another possible implementation, the computer device may sort the props in the recommendation set in descending order according to the attribute values ​​of the game props in the recommendation set, and recommend them in descending order. Exemplarily, the attribute value is used to indicate the ability value of the game props when used in the game; for example, the range of a virtual gun, the attack strength of a virtual sword, the defense strength of a virtual shield, etc. in the game.

[0129] In a possible implementation, when recommending props, the computer device may recommend in combination with the game level that the player is playing; or, it may also recommend in combination with the configuration information of the game props. In Example 1, the computer device may determine the game level that the target object is currently playing, filter out the target props corresponding to the game level from the recommendation set, and recommend the target props to the target object. For example, when the computer device detects that the target object is playing the game level, it may send the target props to the terminal where the target object is located, so that the terminal displays the target props in the game screen corresponding to the game level. For example, the icon of the target props, the number of game coins spent, the attribute value, the attack power, the usage rules, etc. are displayed. For example, the target props corresponding to the game level may include: game props specific to the game level, the number of objects using props in the game level exceeds the target value, etc. For example, if more than 100 objects in the first object set use props b when passing game level E, then props b may be the target props corresponding to the game level. In Example 2, the computer device can determine the configuration information of the game props in the recommended set for the objects in the first object set, and recommend the game props in the recommended set and their corresponding configuration information to the target object. For example, the computer device can send the game props and configuration information in the recommended set to the terminal where the target object is located, and the terminal can display the game props and their configuration information. The configuration information of the game props may include, but is not limited to: the configuration of the attribute parameters of the game props themselves, and the configuration of the accessories required for the game props. For example, the attack value of an attacking virtual tool, the driving speed of a virtual vehicle, etc. For example, when the object in the first object set uses a virtual shotgun, the sight configured is a quadruple scope, then when recommending the virtual shotgun, its corresponding configuration information including the sight being a quadruple scope can also be recommended.

[0130] It should be noted that the computer device may store the association relationship between the target object and the recommendation set (or the second recommendation set) accordingly, and each time a recommendation is made, one or more game props in the recommendation set may be recommended based on the above implementation or the method shown in the example. The above only illustrates the recommendation process through several implementations and examples, and the embodiment of the present application does not specifically limit the number of props recommended each time and the recommendation process.

[0131] In one possible implementation, for the object at the highest level, the computer device may no longer determine the first object set and the recommended set based on the above steps 202-203, but may make targeted recommendations. In one example, if the target level is the highest level in the game application, the computer device may recommend the props with the highest prop attribute value in the game application to the target object. That is, if the target level is the highest level, steps 202-203 may be replaced by: the computer device obtains the props with the highest prop attribute value in the game application, obtains the recommended set, and based on step 204, recommends the game props in the recommended set to the target object. Of course, filtering may also be performed based on the existing props of the target object. For example, the computer device obtains at least two game props with the highest prop attribute values, and filters out the existing props of the target object in the at least two game props to obtain the recommended set. For example, the most powerful attacking virtual tools, the latest virtual shotguns with the highest configuration, and so on are recommended. In another example, if the target level is not the highest level in the game application, the computer device executes the above steps 202-203, first determines the first object set, and obtains the recommendation set based on the prop usage information of the objects in the first object set, and then performs prop recommendations through step 204.

[0132] In one possible implementation, when determining the recommended set based on the above steps 2031-2033, if based on step 2033, after filtering the first recommended set according to the prop screening condition that does not exceed the prop trading level of the target object and does not include the existing props of the target object, the second recommended set obtained is empty, that is, after filtering the props in the first recommended set based on the prop screening condition, there are no game props that meet the conditions. Then step 204 can be replaced by: the computer device recommends a prop exchange package to the target object, and the prop exchange package can be a universal prop package that can be exchanged for other props. Of course, when the recommended set is empty (or the second recommended set is empty), the computer device can also obtain the prop exchange package corresponding to the target level according to the target level of the target object, and recommend the prop exchange package to the target object. For example, a master gift package corresponds to a sub-core player, an advanced gift package corresponds to an ordinary player, and the like.

[0133] There are two related technologies for game props recommendation: one is manual configuration by game planners, and the other is personalized recommendation through machine learning algorithms.

[0134] Among them, the recommendations manually configured by game planners have the following technical defects: First, the recommended content of game props is single, that is, all users see the same content, and even some props that users have already purchased are still recommended repeatedly, resulting in a long-term waste of promotion resources. Second, because the prop list is fixed, in most cases, users cannot quickly find their favorite game props from the first page of the prop list. They usually have to turn the page many times, which is a very poor user experience. Third, because the required props are difficult to find, the user's desire to buy is greatly reduced, which greatly reduces the game revenue. Fourth, the game list is not updated in a timely manner. Usually, the release of the game needs to follow the version, and the latest and best game props cannot be seen by users quickly.

[0135] Personalized recommendation through machine learning algorithms was first applied in the field of e-commerce. It is based on the user's interest characteristics and purchase behavior data, uses models to build complex feature engineering, and recommends information and products that the user is interested in through data mining calculations. It is systematized to form an advanced intelligent recommendation system, called a personalized recommendation system, which provides users with completely personalized decision support and information services during the shopping process, thereby bringing a better product purchase experience, and invisibly increasing the number of product sales and revenue. Similarly, game props, as a virtual commodity, also fall into the category of personalized recommendation. Currently, the commonly used recommendation algorithms in the industry include those based on tree models, deep networks (DeepFM (Deep Factorization Machine, Deep Factorization Machine) network, DIN (Deep Interest Network, Deep Interest Network)), and collaborative filtering algorithms. Although the existing commonly used personalized recommendation algorithms are relatively mature, they may have the following disadvantages when applied in the game field: First, they require more complex feature engineering. Second, algorithm engineers need to have a deep understanding of the game logic and the advantages and disadvantages of each prop, which requires a lot of manpower costs to analyze, and their recommendation schemes are deeply coupled with specific businesses, and the designed schemes are difficult to reuse and migrate on different businesses. Third, a large number of new props are often generated in the game for sale, and the currently commonly used recommendation solutions cannot solve the prop cold start problem well.

[0136] In view of the shortcomings of traditional game planners' manual configuration of recommendations, such as single recommended content, fixed prop lists, and inability to update recommendation results in a timely manner, the game prop recommendation method provided in the present application can adaptively stratify objects in the game on a regular basis and then make recommendations. The recommendation results can be automatically updated routinely. In view of the problems existing in the currently commonly used recommendation algorithms, such as the need for complex feature engineering, the need for algorithm engineers to have a deep understanding of game logic, and the difficulty in solving the problem of cold start of props, the game prop recommendation method provided in the present application can adaptively recommend to the target object the props that are recently frequently selected by objects that are similar to him and better than him, without the need for in-depth analysis of the game logic and the advantages and disadvantages of each prop. There is no need for complex feature engineering or a deep understanding of game logic, which greatly saves costs. This method has strong reusability, versatility, and portability, especially for the problem of cold start of props. It determines the recommendation set directly based on the usage information of the first object set, and the highest-level objects directly recommend the latest props and the strongest props. Therefore, by learning prop usage habits from high-level objects through low-level objects, even the latest props can be effectively recommended with a greater probability, which can effectively solve the problem of cold start of props.

[0137] In order to better illustrate the game props recommendation process of this application, Figure 4 The game prop recommendation process shown in the figure further introduces the game prop recommendation method of the present application by examples. In view of the shortcomings of the existing technical solutions, the game prop recommendation method of the present application will make full use of the object's in-game log data and apply machine learning / deep learning and other technologies to allow the recommendation system to learn the object's embedding vector. Without the need for in-depth analysis of the game logic and the advantages and disadvantages of each prop, the system can adaptively recommend to the object props that are frequently chosen by objects that are similar to and better than the object. The game prop recommendation method of the present application does not require complex feature engineering, nor does it require a deep understanding of game logic. It has strong reusability, versatility, and portability, and can solve the problem of cold start of props. Figure 4 As shown, the game prop recommendation process of the present application may include the following steps (1) to (7).

[0138] Step (1), object stratification: According to the game level, payment characteristics, online time and other characteristics of the object, each object is divided into several layers. The higher the level (the smaller the number of layers), the more core the object is and the higher the game skills are. The purpose of stratification is to recommend to the object the props that are frequently used by people who are similar to him but better than him.

[0139] Step (2), calculate the object embedding vector: according to the object's team characteristics in the game, use the object as the node composition, use the DeepWalk algorithm to perform random walks, and learn the embedding vector of each node.

[0140] Step (3), object similarity calculation: For each object Embedding vector in each layer and the Embedding vectors of each object in the upper layer with a game level higher than it, calculate the cosine similarity. Except for the objects in the topmost first layer, each object in other layers will obtain the top N (where N can be determined according to the actual situation) most similar objects.

[0141] Step (4), determining the recommended item candidate set: In step (3), except for the objects in the first layer, each object in other layers can match the top N (the first N, N is a positive integer) most similar objects. Count the most recently and frequently used game items among these top N similar objects. Thus, each object except those in the first layer will obtain a top N recommended item candidate set.

[0142] Step (5), rule filtering: Use the maximum amount of a single purchase of an item by each object in the most recent half year as the payment ability of the object. In step (4), for the recommended item candidate set obtained for each object except those in the first layer, filter out the items that do not meet the payment ability of the object according to the item price and the payment ability of the object, and at the same time filter out the items that the object already has. Thus, each object except those in the first layer can obtain the recommended item candidate set S.

[0143] Exemplarily, for object u in the i-th layer (i > 1, obtained from user stratification in section 3.2.1), its Embedding vector is denoted as e i , and the game level is p i . Find the set of objects with a game level higher than p i (in actual business, other information can also be used for filtering) among the objects in the first layer to the (i - 1)-th layer. Assume this set has n objects, and each object in this set has a corresponding Embedding vector. The set of these Embedding vectors is denoted as Es = {e 1 , e 2 , e 3 ,..., e j ,...e n}. Calculate the similarity between the Embedding vector e i of user u and each Embedding vector in Es according to the cosine similarity. Sort in descending order of similarity to obtain the top k (k < n) most similar upper-layer objects. Count the most recently (according to the actual business situation, it can be the most recent week, or the most recent month, etc.) and frequently used items among these k objects, and k items can be obtained as the recommended candidate set for object u.

[0144] According to the above logic, each user in other layers except those in the first layer can obtain a recommended candidate set S of k items.

[0145] Step (6), usage ranking: In step (5), except for the first layer of objects, each object in other layers can obtain the recommended item candidate set S. Count again how many different objects in the upper layer have used each item in S. According to the number of users of each item, sort the items in S in reverse order to obtain the final recommended item candidate set S'.

[0146] After step (5), except for the first layer of objects, each of the remaining objects can obtain k prop recommendation candidate sets S. However, there are props in the recommended candidate set whose prices are higher than the object's payment ability or the props the object already has. If such props are recommended to the object, the recommendation will be invalid because the object will not buy them, which wastes operating resources and may cause the object to have a bad game experience. Therefore, we can first filter by payment ability and existing props to obtain an updated prop candidate set S'.

[0147] For the object to be recommended, the maximum amount of a single payment made by the object in the past six months can be counted as the object's payment ability, and the props with prices higher than the payment ability in the recommendation candidate set can be filtered out. In addition, the props already owned by the object can be checked and the props already owned by the object in the recommendation candidate set can be filtered out to obtain the updated prop candidate set S'.

[0148] Except for the first layer of objects, each of the remaining objects can obtain the item candidate set S'. However, during online recommendation, each object can be recommended m (determined according to business needs, for example, m can be 1, 2, 5, etc.) items. In this case, not all the items in S' need to be recommended at once. Therefore, the items in S' need to be sorted. For example, for the object u in the i (i>1) layer, its recommendation candidate set is denoted as S", and the number of people using each item in S" from the 1st to the i-1th layers is counted. It can be considered that the more people use it, the more reasonable the recommendation result is, and the higher the credibility of the recommendation is. The items in S" are sorted in reverse order according to the number of users. After the above processing, except for the first layer of objects, each of the remaining objects can obtain the sorted recommendation candidate set S".

[0149] Step (7), recommending props for the first layer of objects: Because the first layer of objects are the most core, valuable, and highest-level objects in the game, for these objects, it can be defaulted that no personalized recommendations are required, and the strongest and latest props in the game can be directly recommended.

[0150] In short, the purpose of stratification is to differentiate objects based on their mastery of game skills and their level of activity. It can be assumed that the higher the level of an object, the more proficient it is in the game logic, and the more reasonable the props it chooses. For objects in each layer, we hope to learn the prop selection habits of similar upper-layer objects. For the core player group at the top, this group has the highest level of mastery of game logic and can accurately choose the props that benefit them the most. The recommendation system only needs to recommend the latest and strongest props to this group.

[0151] The game props recommendation method provided by the embodiment of the present application determines the target level of the target object based on the game skill information of the target object, thereby dividing the object into multiple levels with different game skills; and determines the first object set based on the target level and the object characteristics of the target object, thereby determining the first object set with a higher level than the target object and similar to the target object in combination with the level, object characteristics, etc. of the object; and subsequently further determines a recommended set of game props used by the first object set to recommend the game props in the recommended set to the target object; so that the low-level object learns the prop usage habits from the high-level object, and since the high-level object is more proficient in the game skills than the low-level object, the possibility of converting the recommended props into effective props is increased, thereby improving the actual recommendation efficiency. In addition, there is no need to deeply analyze the game logic and the advantages and disadvantages of each prop, and the commonly selected props of the feature-similar objects of the higher level can be adaptively recommended based on the level and feature similarity, and there is no need for complex feature engineering and a deep understanding of the game logic, which greatly saves labor costs. By isolating the game logic, game props and game prop recommendation process of different games, and without the need for personalized machine learning models, the method of the present application can be reused in any game for prop recommendation, with strong reusability, versatility and portability. Even for the latest game props, the recommendation method of using props can be learned from the low level to the high level, thereby increasing the possibility of new props being recommended and purchased, solving the prop cold start problem well, and thus improving the practicality of game prop recommendations.

[0152] Figure 5 A schematic diagram of the structure of a game prop recommendation device provided in an embodiment of the present application, such as Figure 5 As shown, the game props recommendation device may include:

[0153] A level determination module 501 is used to determine a target level of a target object based on the game skill information of the target object;

[0154] An object set determination module 502 is used to determine a first object set based on the target level and the object features of the target object, wherein the level of the objects in the first object set is higher than the target level, and the feature similarity between the objects in the first object set and the target object meets the target condition;

[0155] The recommendation module 503 is used to determine a recommended set for the target object based on the prop usage information of the first object set, and recommend game props in the recommended set to the target object.

[0156] In a possible implementation, the device further includes:

[0157] A feature determination module, for determining a feature of each object based on team information of at least two objects in the game application;

[0158] The object set determination module 502 is used to determine a second object set based on the target level, wherein the level of the objects in the second object set is higher than the target level; determine the feature similarity between the target object and each object in the second object set based on the features of the target object and the objects in the second object set; and filter out the first object set from the second object set based on the feature similarity between the target object and each object in the second object set.

[0159] In a possible implementation, the feature determination module includes:

[0160] A construction unit, configured to construct an undirected graph based on the teaming information of the at least two objects, wherein the undirected graph includes at least two nodes corresponding to the at least two objects, and an edge exists between two nodes corresponding to two objects that are teamed with each other in the at least two objects;

[0161] A sequence acquisition unit, used to acquire at least two sequences from the undirected graph in a random walk manner, each sequence including nodes arranged in a walk order during the walk;

[0162] The feature acquisition unit is used to input the at least two sequences into a target model and perform unsupervised training on the target model to obtain the features of each object.

[0163] In a possible implementation, the sequence acquisition unit is used to perform a random walk on each node in the undirected graph based on a target walk number and a walk step length, taking each node as a starting point, to obtain the at least two sequences;

[0164] Correspondingly, the feature acquisition unit is used to obtain the initialization characterization matrix of the at least two objects; for each sequence obtained by each walk, based on the target window size, with each node in each sequence as the center point, traverse at least two subsequences included in each sequence, each subsequence includes a center point and adjacent nodes within the target window size range of the center point; for each sequence, perform the following steps: for each subsequence of each sequence, based on each subsequence and the center point of each subsequence, calculate the value of the target loss function, and optimize the initialization characterization matrix based on the value of the target loss function, until the value of the target loss function reaches the target optimization condition, stop the optimization, and obtain the characterization matrix of at least two objects.

[0165] In a possible implementation, the game skill information is information indicating the proficiency of mastering and applying the game logic; the higher the proficiency of the target object in mastering and applying the game logic, the higher the target level of the target object;

[0166] The game skill information includes at least one of the game level, resource value of the trading item, online time of the subject, registration time of the subject, and number of times of activity participation;

[0167] Accordingly, the level determination module 501 is used for at least one of the following:

[0168] In response to the game level of the target object being within a target level interval, determining the object level of the target object to be a target level corresponding to the target level interval, the game skill information including the game level;

[0169] In response to the resource value of the trading item of the target object being within the target resource interval, determining the object level of the target object to be the target level corresponding to the target resource interval;

[0170] In response to the object online duration of the target object being within the target duration interval, determining the object level of the target object to be the target level corresponding to the target duration interval;

[0171] In response to the object registration time of the target object being within the target time interval, determining the object level of the target object to be the target level corresponding to the target time interval;

[0172] In response to the target object's activity participation times being within a target times interval, the object level of the target object is determined to be a target level corresponding to the target times interval.

[0173] In one possible implementation, the recommendation module 503 is used to determine a first recommendation set based on the prop usage information of each object in the first object set; determine the prop trading level of the target object based on the resource value of the trading prop of the target object; and filter out a second recommendation set from the first recommendation set based on the existing props of the target object and the prop trading level, wherein the second recommendation set does not include the existing props and the resource value of the props in the second recommendation set does not exceed the prop trading level.

[0174] In one possible implementation, the recommendation module 503 is used to count the popularity values ​​of at least two props in the recommendation set in the first object set; based on the popularity values ​​of the at least two props, arrange the at least two props in the recommendation set in descending order; and recommend the game props in the recommendation set to the target object in descending order.

[0175] In a possible implementation, the recommendation module 503 includes at least one of the following:

[0176] Determine the game level currently played by the target object, filter out the target props corresponding to the game level from the recommendation set, and recommend the target props to the target object;

[0177] Determine the configuration information of the game props in the recommended set for the objects in the first object set, and recommend the game props in the recommended set and their corresponding configuration information to the target object.

[0178] In a possible implementation, the recommendation module 503 is further configured to recommend, if the target level is the highest level in the game application, an item with the highest item attribute value in the game application to the target object.

[0179] The game props recommendation device provided by the embodiment of the present application determines the target level of the target object based on the game skill information of the target object, thereby dividing the object into multiple levels with different game skills; and determines the first object set based on the target level and the object characteristics of the target object, thereby determining the first object set with a higher level than the target object and similar to the target object in combination with the level, object characteristics, etc. of the object; and subsequently further determines a recommended set of game props used by the first object set to recommend the game props in the recommended set to the target object; so that the low-level object learns the prop usage habits from the high-level object, and since the high-level object is more proficient in the game skills than the low-level object, the possibility of converting the recommended props into effective props is increased, thereby improving the actual recommendation efficiency. In addition, there is no need to deeply analyze the game logic and the advantages and disadvantages of each prop, and the commonly selected props of the feature-similar objects of the higher level can be adaptively recommended based on the level and feature similarity, and there is no need for complex feature engineering and a deep understanding of the game logic, which greatly saves labor costs. By isolating the game logic, game props and game prop recommendation process of different games, and without the need for personalized machine learning models, the method of the present application can be reused in any game for prop recommendation, with strong reusability, versatility and portability. Even for the latest game props, the recommendation method of using props can be learned from the low level to the high level, thereby increasing the possibility of new props being recommended and purchased, solving the prop cold start problem well, and thus improving the practicality of game prop recommendations.

[0180] The game prop recommendation device of this embodiment can execute the game prop recommendation method shown in the above embodiments of this application. The implementation principle is similar and will not be repeated here.

[0181] Figure 6 1 is a schematic diagram of the structure of a computer device provided in an embodiment of the present application. Figure 6 As shown, the computer device includes: a memory, a processor and a computer program stored in the memory, and the processor executes the above computer program to implement the steps of the game prop recommendation method. Compared with the related art, the following can be achieved:

[0182] By determining the target level of the target object based on the game skill information of the target object, the object is divided into multiple levels with different game skills; and based on the target level and the object features of the target object, a first object set is determined, so as to determine a first object set with a higher level than the target object and similar to the target object in combination with the level of the object, the object features, etc.; and then further determining a recommended set of game props used by the first object set to recommend the game props in the recommended set to the target object; so that the low-level objects learn the prop usage habits from the high-level objects, and since the high-level objects are more proficient in the game skills than the low-level objects, the possibility of converting the recommended props into effective props is increased, thereby improving the actual recommendation efficiency. In addition, there is no need to deeply analyze the game logic and the advantages and disadvantages of each prop. Based on the level and feature similarity, the commonly selected props of the feature-similar objects of the higher level can be adaptively recommended, and there is no need for complex feature engineering and a deep understanding of the game logic, which greatly saves labor costs. By isolating the game logic, game props and game prop recommendation process of different games, and without the need for personalized machine learning models, the method of the present application can be reused in any game for prop recommendation, with strong reusability, versatility and portability. Even for the latest game props, the method can learn the recommended way of using props from the low level to the high level, which increases the possibility of new props being recommended and purchased, solves the prop cold start problem well, and thus improves the practicality of game prop recommendation. In an optional embodiment, a computer device is provided, such as Figure 6 As shown, Figure 6 The computer device 600 shown includes: a processor 601 and a memory 603. The processor 601 and the memory 603 are connected, such as through a bus 602. Optionally, the computer device 600 may also include a transceiver 604, which may be used for data interaction between the computer device and other computer devices, such as data transmission and / or data reception. It should be noted that in actual applications, the transceiver 604 is not limited to one, and the structure of the computer device 600 does not constitute a limitation on the embodiments of the present application.

[0183] Processor 601 may be a CPU (Central Processing Unit), a general purpose processor, a DSP (Digital Signal Processor), an ASIC (Application Specific Integrated Circuit), an FPGA (Field Programmable Gate Array) or other programmable logic devices, transistor logic devices, hardware components or any combination thereof. It may implement or execute various exemplary logic blocks, modules and circuits described in conjunction with the disclosure of this application. Processor 601 may also be a combination that implements computing functions, such as a combination of one or more microprocessors, a combination of a DSP and a microprocessor, etc.

[0184] The bus 602 may include a path to transmit information between the above components. The bus 602 may be a PCI (Peripheral Component Interconnect) bus or an EISA (Extended Industry Standard Architecture) bus. The bus 602 may be divided into an address bus, a data bus, a control bus, etc. For ease of representation, Figure 6 Only one thick line is used in the diagram, but this does not mean that there is only one bus or only one type of bus.

[0185] The memory 603 can be a ROM (Read Only Memory) or other types of static storage devices that can store static information and instructions, a RAM (Random Access Memory) or other types of dynamic storage devices that can store information and instructions, or an EEPROM (Electrically Erasable Programmable Read Only Memory), a CD-ROM (Compact Disc Read Only Memory) or other optical disk storage, optical disk storage (including compressed optical disk, laser disk, optical disk, digital versatile disk, Blu-ray disk, etc.), magnetic disk storage medium\other magnetic storage devices, or any other medium that can be used to carry or store computer programs and can be read by a computer, without limitation herein.

[0186] The memory 603 is used to store the computer program for executing the embodiment of the present application, and the execution is controlled by the processor 601. The processor 601 is used to execute the computer program stored in the memory 603 to implement the steps shown in the above method embodiment.

[0187] Among them, computer equipment includes but is not limited to: servers, service clusters, cloud computing centers, etc.

[0188] An embodiment of the present application provides a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, the steps and corresponding contents of the aforementioned method embodiment can be implemented.

[0189] The embodiment of the present application also provides a computer program product, including a computer program, which can implement the steps and corresponding contents of the aforementioned method embodiment when executed by a processor.

[0190] The terms "first", "second", "third", "fourth", "1", "2", etc. (if any) in the specification and claims of this application and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the numbers used in this way can be interchanged where appropriate, so that the embodiments of the present application described herein can be implemented in an order other than that shown or described in the drawings.

[0191] It should be understood that, although each operation step is indicated by arrows in the flowchart of the embodiment of the present application, the implementation order of these steps is not limited to the order indicated by the arrows. Unless clearly stated herein, in some implementation scenarios of the embodiment of the present application, the implementation steps in each flowchart can be performed in other orders according to demand. In addition, some or all of the steps in each flowchart may include multiple sub-steps or multiple stages based on actual implementation scenarios. Some or all of these sub-steps or stages may be executed at the same time, and each sub-step or stage in these sub-steps or stages may also be executed at different times respectively. In different scenarios of execution time, the execution order of these sub-steps or stages may be flexibly configured according to demand, and the embodiment of the present application does not limit this.

[0192] The above is only an optional implementation method for some implementation scenarios of the present application. It should be pointed out that for ordinary technicians in this technical field, without departing from the technical concept of the solution of the present application, other similar implementation methods based on the technical ideas of the present application are also within the protection scope of the embodiments of the present application.

Claims

1. A method for recommending game props, characterized in that: The method comprises: Determining a target level of the target object based on the game skill information of the target object; Based on the team information of at least two objects in the game application, an undirected graph is constructed, wherein the undirected graph includes at least two nodes corresponding to the at least two objects, and there is an edge between two nodes corresponding to two objects that form a team with each other in the at least two objects; For each node in the undirected graph, based on the target number of walks and the walk step length, a random walk is performed with each node as a starting point to obtain at least two sequences, each sequence including nodes arranged in a walk order during the walk; The at least two sequences are input into a target model, and the target model is trained unsupervisedly to obtain the features of each object, including: obtaining an initialization representation matrix of the at least two objects; for each sequence obtained by each walk, based on the target window size, taking each node in each sequence as the center point, traversing at least two subsequences included in each sequence, each subsequence including the center point and adjacent nodes within the target window size range of the center point; for each sequence, performing the following steps: for each subsequence of each sequence, based on each subsequence and the center point of each subsequence, calculating the value of the target loss function, and optimizing the initialization representation matrix based on the value of the target loss function, until the optimization is stopped when the value of the target loss function reaches the target optimization condition, and obtaining the representation matrix of the at least two objects; Determining a first object set based on the target level and the object features of the target object, wherein the level of the objects in the first object set is higher than the target level, and the feature similarity between the objects in the first object set and the target object meets the target condition; Based on the prop usage information of the first object set, a recommendation set of the target object is determined, and game props in the recommendation set are recommended to the target object.

2. The method according to claim 1, characterized in that The determining of the first object set based on the target level and the object features of the target object includes: Based on the target level, determining a second object set, wherein the level of objects in the second object set is higher than the target level; Based on the features of the target object and the objects in the second object set, determining the feature similarity between the target object and each object in the second object set respectively; The first object set is filtered out from the second object set based on feature similarities between the target object and each object in the second object set.

3. The method according to claim 1, characterized in that The game skill information is information indicating the proficiency of the target object in mastering and applying the game logic; the higher the proficiency of the target object in mastering and applying the game logic, the higher the target level of the target object; The game skill information includes at least one of the game level, the resource value of the trading props, the online time of the object, the registration time of the object, and the number of times the activity is participated; Accordingly, determining the target level of the target object based on the game skill information of the target object includes at least one of the following: In response to the game level of the target object being within a target level interval, determining the object level of the target object to be a target level corresponding to the target level interval, wherein the game skill information includes the game level; In response to the resource value of the trading item of the target object being within the target resource interval, determining the object level of the target object to be the target level corresponding to the target resource interval; In response to the object online duration of the target object being within a target duration interval, determining the object level of the target object to be a target level corresponding to the target duration interval; In response to the object registration time of the target object being within a target time interval, determining the object level of the target object to be a target level corresponding to the target time interval; In response to the target object's activity participation times being within a target times interval, the object level of the target object is determined to be a target level corresponding to the target times interval.

4. The method according to claim 1, characterized in that: The determining the recommended set of target objects based on the prop usage information of the first set of objects includes: Determine a first recommendation set based on the prop usage information of each object in the first object set; Determining the item trading level of the target object based on the resource value of the trading item of the target object; Based on the existing props of the target object and the prop trading level, a second recommendation set is screened out from the first recommendation set, the second recommendation set does not include the existing props and the resource values ​​of the props in the second recommendation set do not exceed the prop trading level.

5. The method according to claim 1, characterized in that The recommending the game props in the recommendation set to the target object includes: Counting the popularity values ​​of at least two items in the recommendation set in the first object set; Arrange the at least two items in the recommendation set in descending order based on the popularity values ​​of the at least two items; The game props in the recommendation set are recommended to the target object in descending order.

6. The method according to claim 1, characterized in that The step of recommending the game props in the recommendation set to the target object includes at least one of the following: Determine the game level currently played by the target object, select the target props corresponding to the game level from the recommendation set, and recommend the target props to the target object; Determine the configuration information of the objects in the first object set for the game props in the recommendation set, and recommend the game props in the recommendation set and their corresponding configuration information to the target object.

7. The method according to claim 1, characterized in that The method further comprises: If the target level is the highest level in the game application, the prop with the highest prop attribute value in the game application is recommended to the target object.

8. A game prop recommendation device, characterized in that: The device comprises: A level determination module, used to determine a target level of the target object based on the game skill information of the target object; The feature determination module includes a construction unit, which is used to construct an undirected graph based on the team information of at least two objects in the game application, wherein the undirected graph includes at least two nodes corresponding to the at least two objects, and there is an edge between two nodes corresponding to two objects that form a team with each other in the at least two objects; The feature determination module further includes a sequence acquisition unit, which is used to perform a random walk on each node in the undirected graph based on a target number of walks and a walk step length, taking each node as a starting point, to obtain at least two sequences, each sequence including nodes arranged in a walk order during the walk process; The feature determination module also includes a feature acquisition unit, which is used to input the at least two sequences into the target model and perform unsupervised training on the target model to obtain the features of each object, including: obtaining the initialization representation matrix of the at least two objects; for each sequence obtained by each walk, based on the target window size, taking each node in each sequence as the center point, traversing at least two subsequences included in each sequence, each subsequence includes the center point and adjacent nodes within the target window size range of the center point; for each sequence, performing the following steps: for each subsequence of each sequence, based on each subsequence and the center point of each subsequence, calculating the value of the target loss function, and optimizing the initialization representation matrix based on the value of the target loss function, until the optimization is stopped when the value of the target loss function reaches the target optimization condition, and obtaining the representation matrix of at least two objects; An object set determination module, configured to determine a first object set based on the target level and the object features of the target object, wherein the level of the objects in the first object set is higher than the target level, and the feature similarity between the objects in the first object set and the target object meets the target condition; The recommendation module is used to determine the recommended set of the target object based on the prop usage information of the first object set, and recommend the game props in the recommended set to the target object.

9. A computer device comprising a memory, a processor and a computer program stored in the memory, characterized in that: The processor executes the computer program to implement the game item recommendation method according to any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the game item recommendation method according to any one of claims 1 to 7 is implemented.

11. A computer program product, comprising a computer program, characterized in that When the computer program is executed by a processor, the game item recommendation method according to any one of claims 1 to 7 is implemented.