Methods, devices, storage media, and electronic devices for predicting the use of virtual appearances

Abstract

This application discloses a method, apparatus, storage medium, and electronic device for predicting the usage of virtual appearances. The method includes: acquiring a first feature corresponding to a first virtual appearance; acquiring a second feature corresponding to a virtual character; acquiring a third feature corresponding to a virtual game; combining the first, second, and third features to predict the usage data of the first virtual appearance over a future time period, obtaining a first prediction result; and, based on the first, second, and third features, determining at least one second virtual appearance from the appearance resources of the virtual game that satisfies relevant conditions with the first virtual appearance, correcting the first prediction result using the at least one second virtual appearance, obtaining a second prediction result. This application solves the technical problem of low accuracy in predicting the usage of virtual appearances.

Description

Technical Field

[0001] This application relates to the field of computers, and more specifically, to a method, apparatus, storage medium, and electronic device for predicting the use of a virtual appearance. Background Technology

[0002] In scenarios involving predicting the use of virtual appearances, future usage is typically predicted based on historical usage patterns. However, this single-dimensional prediction mechanism lacks comprehensive consideration of multi-dimensional data, overlooking a large amount of potentially useful information. This incompleteness of information directly interferes with the accuracy of the prediction, leading to low accuracy in virtual appearance usage predictions. Therefore, the problem of low accuracy in virtual appearance usage predictions exists.

[0003] There is currently no effective solution to the above problems. Summary of the Invention

[0004] This application provides a method, apparatus, storage medium, and electronic device for predicting the use of virtual appearances, in order to at least solve the technical problem of low accuracy in predicting the use of virtual appearances.

[0005] According to one aspect of the embodiments of this application, a method for predicting the use of a virtual appearance is provided, comprising: obtaining a first feature corresponding to a first virtual appearance, wherein the first virtual appearance is used to change the appearance of a virtual character in a virtual game, and the first feature is used to represent attribute data and time information associated with the first virtual appearance; obtaining a second feature corresponding to the virtual character, wherein the second feature is used to represent the appearance resources of the virtual character and the usage data of the virtual character; obtaining a third feature corresponding to the virtual game, wherein the third feature is used to represent the activity data of the virtual game and the usage information of game resources by game users in the virtual game, wherein the game users include users using the virtual character, and the game resources include the appearance resources of the virtual character; combining the first feature, the second feature, and the third feature to predict the usage data of the first virtual appearance in a future time period to obtain a first prediction result; and, if at least one second virtual appearance is determined from the appearance resources of the virtual game based on the first feature, the second feature, and the third feature, and satisfies a relevant condition with the first virtual appearance, then the first prediction result is corrected using the at least one second virtual appearance to obtain a second prediction result.

[0006] According to another aspect of the embodiments of this application, a virtual appearance usage prediction device is also provided, comprising: a first acquisition unit, configured to acquire a first feature corresponding to a first virtual appearance, wherein the first virtual appearance is used to change the appearance of a virtual character in a virtual game, and the first feature is used to represent attribute data and time information associated with the first virtual appearance; a second acquisition unit, configured to acquire a second feature corresponding to the virtual character, wherein the second feature is used to represent the appearance resources of the virtual character and the usage data of the virtual character; and a third acquisition unit, configured to acquire a third feature corresponding to the virtual game, wherein the third feature is used to represent the activity data of the virtual game and the usage data of the virtual game. The information pertains to the usage of game resources by game users, including users who use the virtual characters, and the game resources include the appearance resources of the virtual characters; a prediction unit is used to combine the first feature, the second feature, and the third feature to predict the usage data of the first virtual appearance in a future time period, and obtain a first prediction result; a correction unit is used to correct the first prediction result by using the at least one second virtual appearance that satisfies relevant conditions with the first virtual appearance, based on the first feature, the second feature, and the third feature, from the appearance resources of the virtual game, and obtain a second prediction result.

[0007] As an optional solution, the first acquisition unit includes: a first acquisition module, used to acquire attribute features corresponding to the first virtual appearance, wherein the attribute features are used to represent the first attribute data of the first virtual appearance and the second attribute data of the virtual character, and the first feature includes the attribute features; and a second acquisition module, used to acquire time features corresponding to the first virtual appearance, wherein the time features are used to represent the opening time of the right to use the first virtual appearance, and the first feature includes the time features.

[0008] As an optional solution, the first acquisition module includes at least one of the following: a first acquisition submodule, used to acquire features representing the appearance series to which the first virtual appearance belongs; a second acquisition submodule, used to acquire features representing the appearance quality to which the first virtual appearance belongs; a fourth acquisition submodule, used by the third acquisition submodule, used to acquire features representing the appearance type to which the first virtual appearance belongs; and a fifth acquisition submodule, used to acquire features representing the amount of elements required to obtain the right to use the first virtual appearance.

[0009] As an optional solution, the second acquisition module mentioned above includes at least one of the following: a sixth acquisition submodule, used to acquire features representing the time period in which the opening time is located; and a seventh acquisition submodule, used to acquire features representing the preset time point covered by the opening time.

[0010] As an optional solution, the second acquisition unit includes: 77 a third acquisition module, used to acquire the first usage feature corresponding to the virtual character, wherein the first usage feature is used to represent the first usage data of the appearance resource of the virtual character before the development and use rights of the first virtual appearance, and the second feature includes the first usage feature; and a fourth acquisition module, used to acquire the second usage feature corresponding to the virtual character, wherein the second usage feature is used to represent the second usage data of the virtual character in a recent time period, and the second feature includes the second usage feature.

[0011] As an optional solution, the third acquisition module mentioned above includes at least one of the following: an eighth acquisition submodule, used to acquire features representing the quantity of historical appearance resources of the virtual character before the first virtual appearance usage right was granted; a ninth acquisition submodule, used to acquire features representing the usage amount of the historical appearance resources under different appearance qualities; and a tenth acquisition submodule, used to acquire features representing the usage amount of the historical appearance resources within a specific time period.

[0012] As an optional solution, the fourth acquisition module mentioned above includes at least one of the following: an eleventh acquisition submodule, used to acquire features representing the win rate of the virtual character in the recent time period; a twelfth acquisition submodule, used to acquire features representing the usage amount of the virtual character in the recent time period; and a thirteenth acquisition submodule, used to acquire features representing the number of users of the virtual character in the recent time period.

[0013] As an optional solution, the third acquisition unit includes: a fifth acquisition module, used to acquire the activity characteristics corresponding to the virtual game, wherein the activity characteristics reflect the activity level of the virtual game within a preset time period; a sixth acquisition module, used to acquire the first consumption characteristics corresponding to the virtual game, wherein the first consumption characteristics represent the situation where all game users of the virtual game exchange the right to use the game resources by consuming elements, and the third characteristic includes the first consumption characteristics; and a seventh acquisition module, used to acquire the second consumption characteristics corresponding to the virtual game, wherein the second consumption characteristics represent the situation where users exchange the right to use the game resources by consuming elements, and the third characteristic includes the second consumption characteristics.

[0014] As an optional solution, the fifth acquisition module mentioned above includes at least one of the following: a fourteenth acquisition submodule, used to acquire features representing the number of daily active users of the virtual game; and a fifteenth acquisition submodule, used to acquire features representing the number of matches in each mode of the virtual game.

[0015] As an optional solution, the sixth acquisition module mentioned above includes at least one of the following: a sixteenth acquisition submodule, used to acquire a feature representing the proportion of users who exchange the right to use the game resources in exchange for the aforementioned consumable elements in the virtual game among all game users; a seventeenth acquisition submodule, used to acquire a feature representing the number of users who exchange the right to use the game resources in each interval through the aforementioned consumable elements in the virtual game; and an eighteenth acquisition submodule, used to acquire a feature representing the average consumption amount of the aforementioned consumable elements by all users.

[0016] As an optional solution, the seventh acquisition module mentioned above includes at least one of the following: a nineteenth acquisition submodule, used to acquire features representing the number of users who exchange the above-mentioned consumable elements for the right to use the above-mentioned game resources in each interval; and a twentieth acquisition submodule, used to acquire features representing the average consumption amount of the above-mentioned consumable elements by the above-mentioned users.

[0017] As an optional solution, the prediction unit includes: a fusion module for fusing the first feature, the second feature, and the third feature to obtain multi-dimensional features; and a first prediction module for inputting the multi-dimensional features into a data prediction model to obtain the first prediction result output by the data prediction model, wherein the data prediction model is trained using multiple sample data and is used to predict the gradient regression model of the usage data of the appearance resource in the future time period.

[0018] As an optional solution, the above-mentioned apparatus further includes: a training module, used to perform multiple rounds of training on the initial data prediction model before inputting the multi-dimensional features into the data prediction model and obtaining the first prediction result output by the data prediction model, until a trained data prediction model is obtained, wherein the i-th round of training is any one of the multiple rounds of training, and the i-th round of training is as follows: an eighth acquisition module, used to acquire the sample data of the i-th round of training and the data prediction model of the i-th round of training; a second prediction module, used to use the data prediction model of the i-th round of training to predict the sample data of the i-th round of training. The system performs the following steps: First, it obtains a first predicted value. A fitting module is used to fit an initial decision tree with the residual between the true value corresponding to the sample data in the i-th training round and the first predicted value as the objective. A splitting module is used to maximize the information gain by splitting nodes of the initial decision tree until a stopping condition is met, resulting in the i-th training round decision tree. An update module is used to update the first predicted value using the i-th training round decision tree to obtain a second predicted value. A first determination module is used to determine the data prediction model trained in the i-th training round as the trained data prediction model if the second predicted value meets the convergence condition.

[0019] As an optional solution, the above apparatus further includes: a determining unit, configured to, before correcting the first prediction result using the at least one second virtual appearance to obtain a second prediction result, determine at least two sets of appearance resources from the appearance resources of the virtual game that are associated with the first virtual appearance, wherein different sets of appearance resources correspond to different types of association, and the sets of appearance resources include the second virtual appearance; and a recall unit, configured to, before correcting the first prediction result using the at least one second virtual appearance to obtain a second prediction result, apply a method to each of the at least two sets of appearance resources based on the amount of resources in the sets of appearance resources. The system processes the data using the appropriate recall method to obtain at least two recall resource sets; an allocation unit is used to allocate corresponding fusion weights to each of the at least two recall resource sets before correcting the first prediction result using the at least one second virtual appearance to obtain a second prediction result; a fusion unit is used to perform weighted fusion of the at least two recall resource sets according to the fusion weights before correcting the first prediction result using the at least one second virtual appearance to obtain a second prediction result, to obtain first corrected data; the correction unit includes: a first correction module, used to correct the first prediction result using the first correction data to obtain the second prediction result.

[0020] As an optional solution, the recall unit includes: a removal module, used to remove the highest first proportion and the lowest second proportion from the first resource set whose resource quantity is greater than or equal to a preset threshold, and then take the average value of the remaining samples as the representative value of the recall resource set to obtain the first recall resource set; and a second determination module, used to determine the mean value of all samples in the second resource set whose resource quantity is less than the preset threshold and greater than 0, and adjust the mean value according to the resource quantity to obtain the second recall resource set.

[0021] As an optional solution, the allocation unit includes at least one of the following: a first allocation module, configured to allocate a first weight to each appearance resource based on its importance in each of the recall resource sets, wherein the importance is positively correlated with the first weight, and the fusion weight includes the first weight; a second allocation module, configured to allocate a second weight to each appearance resource based on its quantity in each of the recall resource sets, wherein the quantity of resources is positively correlated with the second weight, and the fusion weight includes the second weight; and a third allocation module, configured to allocate a third weight to each appearance resource based on its dispersion in each of the recall resource sets, wherein the dispersion is positively correlated with the third weight, and the fusion weight includes the third weight.

[0022] As an optional solution, the above-mentioned apparatus further includes: a fourth acquisition unit, configured to, after determining at least two sets of appearance resources that are associated with the first virtual appearance from the appearance resources of the virtual game, and processing each of the at least two sets of appearance resources using the corresponding recall method according to the resource quantity in the appearance resource sets, and obtaining 0 of the recalled resource sets, acquire second correction data based on the usage data of the appearance resources of the virtual game that are of the same appearance quality as the first virtual appearance; the correction unit includes: a second correction module, configured to use the second correction data to correct the first prediction result to obtain the second prediction result.

[0023] According to another aspect of the embodiments of this application, a computer program product or computer program is provided, the computer program product or computer program including computer instructions stored in a computer-readable storage medium. A processor of an electronic device reads the computer instructions from the computer-readable storage medium, and executes the computer instructions, causing the electronic device to perform the virtual appearance usage prediction method as described above.

[0024] According to another aspect of the embodiments of this application, an electronic device is also provided, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the above-described virtual appearance usage prediction method through the computer program.

[0025] In this embodiment, a first feature corresponding to a first virtual appearance is obtained, wherein the first virtual appearance is used to change the appearance of a virtual character in a virtual game, and the first feature is used to represent the attribute data and time information associated with the first virtual appearance; a second feature corresponding to the virtual character is obtained, wherein the second feature is used to represent the appearance resources of the virtual character and the usage data of the virtual character; a third feature corresponding to the virtual game is obtained, wherein the third feature is used to represent the activity data of the virtual game and the usage information of game resources by game users in the virtual game, wherein game users include users who use the virtual character, and game resources include the appearance resources of the virtual character; combining the first feature, the second feature, and the third feature, the usage data of the first virtual appearance in a future time period is predicted to obtain a first prediction result; if, based on the first feature, the second feature, and the third feature, at least one second virtual appearance is determined from the appearance resources of the virtual game that satisfies the relevant conditions with the first virtual appearance, the first prediction result is corrected using the at least one second virtual appearance to obtain a second prediction result.

[0026] First, the first feature corresponding to the first virtual appearance is obtained. The first virtual appearance is used to change the appearance of virtual characters in the virtual game, and the first feature represents the attribute data and time information associated with the first virtual appearance. This data provides basic information from the perspective of the first virtual appearance itself, providing raw material for subsequent analysis of its usage.

[0027] Next, the second feature corresponding to the virtual character is obtained. This second feature represents the virtual character's appearance resources and usage data. The virtual character's appearance resources reflect the range of appearances that the character can be paired with. Due to differences in their appearance resources and usage habits, different virtual characters have different probabilities of using the first virtual appearance, providing an important basis for prediction at the virtual character level.

[0028] Next, the third feature corresponding to the virtual game is obtained. This third feature represents the virtual game's activity data and information on game users' usage of game resources. Game users include those using virtual characters, and game resources include the appearance resources of virtual characters. The virtual game's activity data reflects the game's overall popularity. The information on game users' usage of game resources reflects the user group's consumption habits and aesthetic trends, providing macro-level and group-level data for prediction from both the perspective of the game as a whole and the user group.

[0029] After acquiring the first, second, and third features, the usage data of the first virtual appearance in the future is predicted by combining these three dimensions, resulting in the first prediction result. This prediction method, which combines multiple dimensions, fully considers the appearance's own attributes, the characteristics of the virtual character, and the overall game and user group situation, avoiding the information gaps inherent in single-dimensional predictions and making the prediction results more comprehensive and accurate. For example, predicting solely based on the historical usage of the first virtual appearance may overlook the potential impact of recent increases in game activity on its usage, while combining the third feature allows for a more accurate grasp of this trend.

[0030] Furthermore, after identifying at least one second virtual appearance from the virtual game's appearance resources that satisfies relevant conditions with the first virtual appearance based on the first, second, and third features, the first prediction result is corrected using at least one second virtual appearance to obtain a second prediction result. The usage data of the second virtual appearance can provide a reference for the prediction of the first virtual appearance. If the second virtual appearance has a high usage under similar conditions, then the first virtual appearance may also have similar performance in the future, thus allowing for reasonable adjustments to the first prediction result. For example, if it is found that the usage of a second virtual appearance with a similar style to the first virtual appearance has increased significantly recently, then the predicted usage of the first virtual appearance can be appropriately increased, further improving the accuracy of the prediction.

[0031] This embodiment acquires multi-dimensional feature data, combines this data to make an initial prediction, and uses similar appearance data for correction, thereby achieving the goal of comprehensively considering various factors affecting the use of virtual appearances. This improves the accuracy of virtual appearance usage prediction and solves the technical problem of low accuracy in virtual appearance usage prediction. Attached Figure Description

[0032] The accompanying drawings, which are included to provide a further understanding of this application and form part of this application, illustrate exemplary embodiments and are used to explain this application, but do not constitute an undue limitation of this application. In the drawings:

[0033] Figure 1This is a schematic diagram of an application environment for an optional virtual appearance usage prediction method according to an embodiment of this application;

[0034] Figure 2 This is a schematic diagram of the flow of an optional virtual appearance usage prediction method according to an embodiment of this application;

[0035] Figure 3 This is a schematic diagram of an optional virtual appearance usage prediction method according to an embodiment of this application;

[0036] Figure 4 This is a schematic diagram of another optional virtual appearance usage prediction method according to an embodiment of this application;

[0037] Figure 5 This is a schematic diagram of another optional virtual appearance usage prediction method according to an embodiment of this application;

[0038] Figure 6 This is a schematic diagram of another optional virtual appearance usage prediction method according to an embodiment of this application;

[0039] Figure 7 This is a schematic diagram of another optional virtual appearance usage prediction method according to an embodiment of this application;

[0040] Figure 8 This is a schematic diagram of another optional virtual appearance usage prediction method according to an embodiment of this application;

[0041] Figure 9 This is a schematic diagram of another optional virtual appearance usage prediction method according to an embodiment of this application;

[0042] Figure 10 This is a schematic diagram of another optional virtual appearance usage prediction method according to an embodiment of this application;

[0043] Figure 11 This is a schematic diagram of another optional virtual appearance usage prediction method according to an embodiment of this application;

[0044] Figure 12 This is a schematic diagram of another optional virtual appearance usage prediction method according to an embodiment of this application;

[0045] Figure 13 This is a schematic diagram of an optional virtual appearance usage prediction device according to an embodiment of this application;

[0046] Figure 14 This is a schematic diagram of the structure of an optional electronic device according to an embodiment of this application. Detailed Implementation

[0047] To enable those skilled in the art to better understand the present application, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present application, and not all embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative effort should fall within the scope of protection of the present application.

[0048] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of this application described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0049] According to one aspect of the embodiments of this application, a method for predicting the use of virtual appearances is provided. Optionally, as an optional implementation, the above-mentioned method for predicting the use of virtual appearances can be applied to, but is not limited to, situations such as... Figure 1 The environment shown may include, but is not limited to, user equipment 102 and server 112. User equipment 102 may include, but is not limited to, a display 104, a processor 106 and a memory 108. Server 112 includes a database 114 and a processing engine 116.

[0050] The specific process can be summarized in the following steps:

[0051] In step S102, the user device 102 acquires the first feature corresponding to the first virtual appearance, the second feature corresponding to the virtual character, and the third feature corresponding to the virtual game;

[0052] Step S104: Send the first feature, the second feature, and the third feature to the server 112 via network 110;

[0053] In steps S106-S110, server 112 uses processing engine 116 to combine the first feature, the second feature, and the third feature to predict the usage data of the first virtual appearance in the future time period, and obtains a first prediction result. Then, based on the first feature, the second feature, and the third feature, server 112 determines at least one second virtual appearance from the appearance resources of the virtual game that meets the relevant conditions with the first virtual appearance. Finally, server 112 uses at least one second virtual appearance to correct the first prediction result, and obtains a second prediction result.

[0054] In step S112, the second prediction result is sent to the user equipment 102 via the network 110. The user equipment 102 displays the second prediction result on the display 104 via the processor 106 and stores the second prediction result in the memory 108.

[0055] remove Figure 1 Beyond the examples shown, the aforementioned user equipment can be a terminal device configured with a target client, and may include, but is not limited to, at least one of the following: mobile phones (such as Android phones, iOS phones, etc.), laptops, tablets, PDAs, MIDs (Mobile Internet Devices), desktop computers, smart TVs, computers, smart voice interaction devices, smart home appliances, vehicle terminals, aircraft, etc. This embodiment can be applied to various scenarios, including but not limited to digital humans, virtual humans, games, virtual reality, extended reality (XR), etc.

[0056] The target client can be a video client, instant messaging client, browser client, educational client, etc. The aforementioned network can include, but is not limited to, wired networks and wireless networks. The wired network includes local area networks (LANs), metropolitan area networks (MANs), and wide area networks (WANs). The wireless network includes Bluetooth, Wi-Fi, and other networks that enable wireless communication. The aforementioned server can be a single server, a server cluster consisting of multiple servers, or a cloud server providing basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, CDN, and big data and artificial intelligence platforms. The above is merely an example, and no limitations are imposed in this embodiment.

[0057] Alternatively, as an alternative implementation method, such as Figure 2 As shown, the virtual appearance usage prediction method can be executed by an electronic device, such as... Figure 1 The user equipment or server shown includes the following specific steps:

[0058] S202, Obtain the first feature corresponding to the first virtual appearance, wherein the first virtual appearance is used to change the appearance of the virtual character in the virtual game, and the first feature is used to represent the attribute data and time information associated with the first virtual appearance;

[0059] In an optional embodiment, the first virtual appearance: among the rich elements of a virtual game, it is a part specifically used to change the appearance of the virtual character, such as clothing elements, special effects elements, background elements, and accessory elements. For example... Figure 3 As shown, in a role-playing game, the first virtual appearance 304 can be a bear headgear, and the virtual character 302 can wear the bear headgear.

[0060] In an optional embodiment, the first feature is: attribute data and time information associated with the first virtual appearance. The attribute data covers multiple dimensions of the appearance, such as style, type, rarity, and special effects; the time information can record key time nodes such as its launch, update, and peak usage.

[0061] S204, Obtain the second feature corresponding to the virtual character, wherein the second feature is used to represent the appearance resources of the virtual character and the usage data of the virtual character;

[0062] In an optional embodiment, a virtual character is a core element of a virtual game, which can be an object controlled and played by the player in the game, such as a game hero or a game vehicle. Different virtual characters can have different roles, skills, and characteristics; for example, in a virtual game, there are warriors, mages, and priests who provide support and healing.

[0063] In an optional embodiment, the second feature is used to represent the appearance resources of the virtual character and the usage data of the virtual character. The appearance resources reflect the range of appearances that the character can wear, and the usage data reflects the character's gaming behavior habits and preferences for different appearances.

[0064] To illustrate further, for example... Figure 4 As shown, virtual character 402 has n optional clothing appearances. Clothing appearance 1 and clothing appearance 2 are those that virtual character 402 has already unlocked and can use, while clothing appearance n is those that virtual character 402 has not yet unlocked and can use.

[0065] S206, Obtain the third feature corresponding to the virtual game, wherein the third feature is used to represent the activity data of the virtual game and the usage information of game resources by game users in the virtual game. Game users include users who use virtual characters, and game resources include the appearance resources of virtual characters.

[0066] In an alternative embodiment, virtual games: platforms that provide elements such as virtual characters and virtual appearances, where players engage in various game activities, forming a social and entertainment ecosystem. For example, a massively multiplayer online role-playing game (MMORPG) with a player base and game content.

[0067] In an optional embodiment, the third feature is used to represent activity data of the virtual game and information on game users' use of game resources. Activity data reflects the game's popularity and player engagement, while user usage information reflects users' preferences for game resources and their consumption behavior.

[0068] To further illustrate, one could optionally monitor the number of online players in the game in real-time or periodically to understand the game's real-time popularity. For example, if a game has 100,000 online players during prime time in the evening, it indicates that this period is the game's peak.

[0069] S208, combining the first feature, the second feature, and the third feature, predict the usage data of the first virtual appearance in the future time period to obtain the first prediction result;

[0070] In an optional embodiment, the first feature, the second feature, and the third feature represent the relevant feature data of the first virtual appearance, the virtual character, and the virtual game, respectively. They are related to each other and influence each other, and together constitute the basis for prediction.

[0071] In an optional embodiment, the first prediction result is a predicted value about the usage of the first virtual appearance in the future time period, obtained by comprehensively analyzing multi-dimensional feature data, including usage amount, usage frequency, usage time period, etc.

[0072] To further illustrate, the data for the first, second, and third features can be organized and matched to form a complete dataset. For example, the attribute data of the first virtual appearance can be correlated with the appearance resources and usage data of the virtual character to analyze the impact of the virtual character's existing appearance on the use of the first virtual appearance; then, combined with the game's activity data and user usage information, the potential demand of the overall game environment for the first virtual appearance can be understood.

[0073] S210, if at least one second virtual appearance is determined from the appearance resources of the virtual game based on the first feature, the second feature, and the third feature, and at least one second virtual appearance that satisfies the relevant conditions with the first virtual appearance, the first prediction result is corrected using the at least one second virtual appearance to obtain a second prediction result.

[0074] In an optional embodiment, the second virtual appearance is another appearance in the virtual game besides the first virtual appearance, which meets certain related conditions with the first virtual appearance, such as similar style, similar function, and the same use scenario. Their performance in actual use can provide a reference for the prediction of the first virtual appearance.

[0075] In an optional embodiment, the second prediction result—a more accurate prediction of the future usage of the first virtual appearance after being corrected by the second virtual appearance—can provide game operators with a more reliable basis for decision-making.

[0076] To further illustrate, one could optionally establish similarity standards between the first virtual appearance and the second virtual appearance, based on the characteristics of the first virtual appearance, such as similarity in appearance style, usage scenario, and functional features. For example, if the first virtual appearance is a retro-style clothing appearance, the style similarity could be set to over 80%, meaning the second virtual appearance also needs to have obvious retro elements.

[0077] Besides similarity criteria, other influencing factors can be considered, such as the release time and rarity of the second virtual appearance. For example, second virtual appearances with similar release times and rarity are more valuable for predicting and correcting the first virtual appearance.

[0078] It should be noted that this embodiment aims to construct a comprehensive and accurate system for predicting and correcting the use of virtual appearances. In virtual game environments, virtual appearances play a crucial role in enhancing the player's gaming experience and promoting game consumption. However, predictions based on a single dimension often fail to accurately grasp the usage trends of virtual appearances. Correspondingly, this embodiment comprehensively acquires feature data from three levels: the first virtual appearance, the virtual character, and the virtual game. It first makes a preliminary prediction of the future usage data of the first virtual appearance, and then uses actual data from the second virtual appearance related to the first virtual appearance to correct the prediction results. This provides game operators with more valuable decision-making basis, helping games optimize appearance design, rationally allocate resources, and formulate precise marketing strategies.

[0079] To further illustrate, optional examples include... Figure 5 As shown, the first feature corresponding to the first virtual appearance 502 is obtained; the second feature corresponding to the virtual character 506 is obtained; and the third feature corresponding to the virtual game 508 is obtained.

[0080] By combining the first feature, the second feature, and the third feature, the usage data of the first virtual appearance 502 in the future time period (example of dashed arrow) is predicted to obtain the first prediction result 510;

[0081] Based on the first feature, the second feature, and the third feature, if at least one second virtual appearance 504 is determined from the appearance resources of the virtual game 508 that satisfies the relevant conditions with the first virtual appearance 502, the first prediction result 510 is corrected using at least one second virtual appearance 504 to obtain the second prediction result 512.

[0082] Optionally, the content of this embodiment can be applied to other embodiments described above, or combined with other embodiments described above.

[0083] The embodiments provided in this application first obtain a first feature corresponding to a first virtual appearance. The first virtual appearance is used to change the appearance of a virtual character in a virtual game. The first feature represents the attribute data and time information associated with the first virtual appearance. This data provides basic information from the perspective of the first virtual appearance itself, providing raw material for subsequent analysis of its usage.

[0084] Next, the second feature corresponding to the virtual character is obtained. This second feature represents the virtual character's appearance resources and usage data. The virtual character's appearance resources reflect the range of appearances that the character can be paired with. Due to differences in their appearance resources and usage habits, different virtual characters have different probabilities of using the first virtual appearance, providing an important basis for prediction at the virtual character level.

[0085] Next, the third feature corresponding to the virtual game is obtained. This third feature represents the virtual game's activity data and information on game users' usage of game resources. Game users include those using virtual characters, and game resources include the appearance resources of virtual characters. The virtual game's activity data reflects the game's overall popularity. The information on game users' usage of game resources reflects the user group's consumption habits and aesthetic trends, providing macro-level and group-level data for prediction from both the perspective of the game as a whole and the user group.

[0086] After acquiring the first, second, and third features, the usage data of the first virtual appearance in the future is predicted by combining these three dimensions, resulting in the first prediction result. This prediction method, which combines multiple dimensions, fully considers the appearance's own attributes, the characteristics of the virtual character, and the overall game and user group situation, avoiding the information gaps inherent in single-dimensional predictions and making the prediction results more comprehensive and accurate. For example, predicting solely based on the historical usage of the first virtual appearance may overlook the potential impact of recent increases in game activity on its usage, while combining the third feature allows for a more accurate grasp of this trend.

[0087] Furthermore, after identifying at least one second virtual appearance from the virtual game's appearance resources that satisfies relevant conditions with the first virtual appearance based on the first, second, and third features, the first prediction result is corrected using at least one second virtual appearance to obtain a second prediction result. The usage data of the second virtual appearance can provide a reference for the prediction of the first virtual appearance. If the second virtual appearance has a high usage under similar conditions, then the first virtual appearance may also have similar performance in the future, thus allowing for reasonable adjustments to the first prediction result. For example, if it is found that the usage of a second virtual appearance with a similar style to the first virtual appearance has increased significantly recently, then the predicted usage of the first virtual appearance can be appropriately increased, further improving the accuracy of the prediction.

[0088] This embodiment acquires multi-dimensional feature data, combines this data to make an initial prediction, and uses similar appearance data for correction, thereby achieving the goal of comprehensively considering various factors affecting the use of virtual appearances, and thus realizing the technical effect of improving the accuracy of virtual appearance usage prediction.

[0089] As an optional approach, obtaining the first feature corresponding to the first virtual appearance includes:

[0090] S1-1, Obtain the attribute features corresponding to the first virtual appearance, wherein the attribute features are used to represent the first attribute data of the first virtual appearance and the second attribute data of the virtual character, and the first feature includes the attribute features;

[0091] S1-2, Obtain the time feature corresponding to the first virtual appearance, wherein the time feature is used to represent the opening time of the right to use the first virtual appearance, and the first feature includes the time feature.

[0092] In an optional embodiment, the first virtual appearance is an appearance element with a specific visual expression that is provided to a virtual character in a virtual scene (such as a game, a virtual social platform, etc.), such as character skins in a game, character clothing in a virtual social scene, etc.

[0093] In an optional embodiment, the first feature is a set of various characteristics used to describe the first virtual appearance, which integrates information about the first virtual appearance in different dimensions in order to define and identify it more comprehensively and accurately.

[0094] In an optional embodiment, the attribute features are: first attribute data representing the first virtual appearance and second attribute data representing the virtual character. The first attribute data is data related to the characteristics of the first virtual appearance itself, such as the appearance's color, texture, and shape; the second attribute data is data on how the virtual appearance affects the attributes of the virtual character, such as changes in the character's attack power, defense power, and other attributes after wearing a specific appearance.

[0095] In an optional embodiment, the time feature is used to indicate the open time of the right to use the first virtual appearance, that is, the time period during which the user can use the first virtual appearance, such as limited to a specific event period, a certain time period, or permanently open for use.

[0096] It should be noted that this embodiment describes the specific method for obtaining the first feature corresponding to the first virtual appearance. The first feature is not a single feature, but a combination of multiple features of different types. Here, it is mainly obtained from two aspects: attribute features and time features, in order to comprehensively construct the first feature of the first virtual appearance.

[0097] Optionally, the content of this embodiment can be applied to other embodiments described above, or combined with other embodiments described above.

[0098] Through the embodiments provided in this application, by acquiring attribute features, the system can accurately record the visual presentation details of the first virtual appearance and its impact on the attributes of the virtual character, ensuring that the appearance is correctly rendered and the corresponding attribute changes are applied in the virtual scene. For example, in a game, after a player selects a specific appearance, the character's appearance can be accurately displayed, and attributes such as attack power are adjusted according to the settings. By acquiring time features, the system can control the usage permissions of the first virtual appearance according to preset rules, allowing users to use it during open hours and restricting its use outside of open hours, thus achieving effective management of the virtual appearance's usage time. Combining these two aspects, the first virtual appearance can be defined completely and accurately, providing users with a rich and controllable virtual experience and enhancing the fun and playability of the virtual scene.

[0099] As an optional approach, the attribute features corresponding to the first virtual appearance are obtained, including at least one of the following:

[0100] S2-1, Obtain the features used to represent the appearance series to which the first virtual appearance belongs;

[0101] S2-2, Obtain the features used to represent the appearance quality of the first virtual appearance;

[0102] S2-3, Obtain the feature used to represent the appearance type to which the first virtual appearance belongs;

[0103] S2-4, obtain the feature representing the amount of elements required to obtain the right to use the first virtual appearance.

[0104] In an optional embodiment, the first virtual appearance is: in a virtual environment (such as a game, a virtual social platform, etc.), the appearance of a virtual character or virtual scene element with specific visual performance and function, such as character skins, attack props appearances, and architectural decorations in a virtual scene.

[0105] In an optional embodiment, attribute features are a set of data used to describe various characteristics of the first virtual appearance. These features reflect the attribute status of the first virtual appearance from different perspectives, helping the system to identify, classify, and manage the first virtual appearance.

[0106] In an alternative embodiment, an appearance series is a collection of first virtual appearances that have similar design styles, themes, or associations, such as a series of character skins themed around "ancient mythology".

[0107] In an optional embodiment, appearance quality is used to measure the level of the first virtual appearance in terms of craftsmanship, rarity, and functional strength, and can generally be divided into different quality levels such as common, rare, epic, and legendary.

[0108] In an optional embodiment, appearance type: a classification based on the purpose, function, or presentation of the first virtual appearance, such as character appearance, attack item appearance, item appearance, etc.

[0109] In an optional embodiment, the amount of elements consumed refers to the quantity of specific virtual elements that a user needs to consume to obtain the right to use the first virtual appearance. These virtual elements can be in-game currency, items, points, etc. For example... Figure 6 As shown, the right to use the first virtual appearance 602 requires a certain number of points (consuming element 604) to unlock. After unlocking, the right to use the first virtual appearance 602 is obtained, and the user can use the first virtual appearance 602. For example, consuming 2 points will grant the right to use the first virtual appearance 602.

[0110] It should be noted that this embodiment describes various methods for acquiring attribute features across different dimensions when obtaining the attribute features corresponding to the first virtual appearance. It does not require acquiring all features simultaneously, but rather acquiring at least one of them. By acquiring attribute features from aspects such as appearance series, appearance quality, appearance type, and the amount of elements required to acquire usage rights, the attribute situation of the first virtual appearance can be more comprehensively and meticulously depicted, providing rich data support for subsequent management, display, and user interaction of the first virtual appearance in the virtual scene.

[0111] Optionally, the content of this embodiment can be applied to other embodiments described above, or combined with other embodiments described above.

[0112] Through the embodiments provided in this application, by acquiring appearance series features, the system can categorize and manage appearances with the same theme or style, and present them according to series during display, enhancing visual coherence and thematic consistency. Acquiring appearance quality features allows for the setting of different attribute bonuses and special effects based on quality levels, while also reflecting differences in acquisition methods, enhancing the game's depth and playability. Acquiring appearance type features helps the system perform specific operations for different types of appearances, such as character appearance equipping and attack item appearance replacement, and also facilitates users in quickly finding the appearances they need. Acquiring the amount of consumable elements allows for precise control of the difficulty for users to acquire appearances, maintaining the balance of the virtual economic system and preventing appearance acquisition from being too easy or too difficult, thus affecting user experience and the game ecosystem.

[0113] As an optional approach, the temporal characteristics corresponding to the first virtual appearance are obtained, including at least one of the following:

[0114] S3-1, Obtain the features used to represent the time period in which the opening time is located;

[0115] S3-2, Obtain the features used to represent the preset time points of open time coverage.

[0116] In an optional embodiment, the first virtual appearance is an appearance form with specific visual expression or function set for a virtual character or virtual scene element in a virtual scene (such as a game, a virtual social space, etc.), such as a limited character skin in a game, a festival-themed decoration in a virtual social scene, etc.

[0117] In an optional embodiment, the time feature is used to describe the relevant attributes of the first virtual appearance in the time dimension, mainly involving the opening time information of the right to use the appearance, so as to determine when the user can use the appearance.

[0118] In an optional embodiment, the time period refers to the time span that the first virtual appearance is open for use and follows a certain regularity, such as a period of days, weeks, months, or years, or a specific activity cycle (such as the season cycle in a game).

[0119] In an optional embodiment, the preset time point is a pre-defined specific moment when the first virtual appearance becomes available for use, such as a specific date, hour, and minute, like midnight on a specific holiday in a game. For example... Figure 7 As shown, the opening hours 702 cover the time period (including Wednesday the 17th, Thursday the 18th, Friday the 19th, and Saturday the 20th), including specific dates (marked with ""). (Marked Saturday the 20th).

[0120] It should be noted that this embodiment describes how, when acquiring the time characteristics corresponding to the first virtual appearance, relevant information can be obtained from two different dimensions, not simultaneously; at least one of them must be obtained. By acquiring the time period characteristics, the time span pattern of the first virtual appearance's open use can be clearly identified; by acquiring the preset time point characteristics, the specific moment of its open use can be accurately located. These two methods improve the acquisition of the time characteristics of the first virtual appearance from different levels, providing comprehensive data for the system to accurately control the usage time of the appearance.

[0121] Optionally, the content of this embodiment can be applied to other embodiments described above, or combined with other embodiments described above.

[0122] Through the embodiments provided in this application, after obtaining the time period characteristics, the system can automatically control the opening and closing of the first virtual appearance according to the set periodic pattern, without frequent manual intervention, thus improving management efficiency. At the same time, users can also plan their usage time in advance according to the periodic pattern, enhancing the user experience. By obtaining the preset time point characteristics, the system can open appearance usage permissions at precise times, avoiding time errors and ensuring that appearances can be used on time in specific activities or scenarios. This enhances the timeliness and fun of the virtual scene and also helps maintain the stability of the virtual economic system, preventing economic imbalances caused by inaccurate time control.

[0123] As an optional approach, obtaining the second feature corresponding to the virtual character includes:

[0124] S4-1, Obtain the first usage feature corresponding to the virtual character, wherein the first usage feature is used to represent the first usage data of the virtual character's appearance resources before the development and use rights of the first virtual appearance, and the second feature includes the first usage feature;

[0125] S4-2, Obtain the second usage feature corresponding to the virtual character, wherein the second usage feature is used to represent the second usage data of the virtual character in a recent time period, and the second feature includes the second usage feature.

[0126] In an optional embodiment, a virtual character is a character with a specific appearance, attributes, and behavioral abilities that is controlled by the user or set by the system in a virtual environment (such as a game, a virtual social platform, a virtual animation, etc.). Examples include a hero character controlled by a player in a game, or a virtual image of a user in a virtual social platform.

[0127] In an optional embodiment, the second feature is a data set used to comprehensively describe the various characteristics of the virtual character. It covers relevant data of the virtual character in different usage scenarios and time stages to fully reflect the usage and status of the virtual character.

[0128] In an optional embodiment, the first usage feature represents the first usage data of the virtual character's appearance resources prior to the development and use of the first virtual appearance. This data reflects the frequency of use and usage preferences of the virtual character's appearance resources before a specific appearance is available, such as the types and durations of other skins that a character frequently uses before a specific skin is released.

[0129] In an optional embodiment, the second usage feature represents the virtual character's second usage data within a recent time period. This data focuses on the virtual character's recent behavioral performance and reflects the character's usage habits and activity level at the current stage, such as the number of game sessions the character has participated in recently and the frequency of skill usage.

[0130] It should be noted that this embodiment focuses on acquiring the second feature corresponding to the virtual character, and adopts a data collection approach from two different dimensions. On the one hand, it acquires the first usage data of the virtual character's appearance resources before the development and use rights of the first virtual appearance, as the first usage feature; on the other hand, it acquires the second usage data of the virtual character in a recent time period, as the second usage feature. By integrating the usage features from these two different stages, a comprehensive and integrated second feature of the virtual character is constructed, providing rich and accurate data support for subsequent related operations based on these features, such as personalized recommendations and virtual character status analysis.

[0131] Optionally, the content of this embodiment can be applied to other embodiments described above, or combined with other embodiments described above.

[0132] By acquiring the first usage characteristic through the embodiments provided in this application, the system can deeply analyze the usage habits and preferences of virtual characters before a specific appearance appears, providing historical basis for subsequent accurate recommendations of related appearances or adjustments to virtual character attributes. For example, if it is found that a character frequently uses appearances with a specific style, similar styles can be prioritized when recommending new appearances. Acquiring the second usage characteristic allows for real-time monitoring of the virtual character's recent activity level and usage trends. Based on this data, the system can adjust relevant strategies for virtual characters in a timely manner, such as providing more rewards or special tasks for highly active characters, improving the user's interaction experience with virtual characters and the overall activity of the virtual environment. Combining these two characteristics, the system can gain a more comprehensive and accurate understanding of virtual characters, providing users with more personalized and intelligent services.

[0133] As an optional approach, the first usage characteristic corresponding to the virtual character is obtained, including at least one of the following:

[0134] S5-1, Obtain the feature representing the number of historical appearance resources of a virtual character before the first virtual appearance usage right is granted;

[0135] S5-2, Obtain features to represent the usage of historical appearance resources under different appearance qualities;

[0136] S5-3, Obtain features that represent the usage of historical appearance resources within a specific time period.

[0137] In an optional embodiment, the first usage feature is a set of relevant data features used to describe the virtual character's usage of appearance resources before the first virtual appearance usage right is granted, reflecting the virtual character's past usage habits and preferences for appearance resources.

[0138] In an optional embodiment, historical appearance resources refer to various appearance elements that the virtual character already possessed and used before the first virtual appearance usage rights were granted, such as character skins, attack item appearances, and mount appearances in the game.

[0139] In an optional embodiment, appearance quality is an indicator that measures the level of historical appearance resources in terms of craftsmanship, rarity, and functional strength. It can usually be divided into different levels such as common, rare, epic, and legendary.

[0140] In an optional embodiment, a specific time period refers to a pre-defined time interval with a clear start and end time, such as a certain time period during a day, a certain number of days in a week, or a specific week in a month.

[0141] It should be noted that this embodiment revolves around obtaining the first usage characteristics corresponding to the virtual character, providing three different acquisition methods. It is not required that all three be acquired simultaneously, but rather that at least one of them be obtained. By obtaining the historical quantity characteristics of appearance resources before the virtual character's first virtual appearance usage right was granted, we can understand the scale of appearance resources the character previously possessed; by obtaining the usage characteristics of historical appearance resources at different appearance qualities, we can understand the character's preference for appearances of different qualities; by obtaining the usage characteristics of historical appearance resources within a specific time period, we can know the changes in the frequency of appearance resource usage by the character at different times. Combining this information provides a comprehensive and detailed characterization of the virtual character's appearance usage before the first virtual appearance was granted, providing strong data support for subsequent personalized services and appearance recommendations for virtual characters.

[0142] Optionally, the content of this embodiment can be applied to other embodiments described above, or combined with other embodiments described above.

[0143] By acquiring historical appearance resource quantity characteristics through the embodiments provided in this application, the system can clearly grasp the total amount of appearance resources previously owned by a virtual character, facilitating resource management and statistical analysis. For example, users can be categorized into different levels based on resource quantity to provide differentiated services. Acquiring usage characteristics of different appearance qualities allows for in-depth analysis of virtual characters' preferences for different quality appearances. When recommending new appearances, priority is given to pushing quality types that match their preferences, improving recommendation accuracy and user satisfaction. Acquiring usage characteristics over specific time periods reveals the activity level of virtual characters' appearance usage at different times. Based on this data, the system can increase the promotion of relevant appearances during active periods, optimize appearance display strategies, and enhance the user's interactive experience with virtual characters and the overall activity level of the virtual environment.

[0144] As an optional approach, the second usage characteristic corresponding to the virtual character is obtained, including at least one of the following:

[0145] S6-1, Obtain the features used to represent the win rate of virtual characters in recent time periods;

[0146] S6-2, Obtain features representing the usage of virtual characters in a recent time period;

[0147] S6-3, Obtain features representing the number of users of a virtual character in a recent time period.

[0148] In an optional embodiment, the second usage feature is a set of relevant data features used to describe the usage of the virtual character in a recent time period, reflecting the recent usage status and popularity of the virtual character from multiple dimensions.

[0149] In an optional embodiment, the recent time period is a pre-defined time interval that is relatively close to the present, which can be set to different durations such as one day, one week, or one month according to specific needs.

[0150] In an optional embodiment, the win rate is used: the proportion of games a virtual character wins out of the total number of games played in a recent period, reflecting the character's performance level in recent battles.

[0151] In an optional embodiment, usage refers to the total number of times a virtual character has been used in a recent time period, reflecting the frequency of the character's use in the recent period.

[0152] In an optional embodiment, the number of users refers to the number of different users who have used the virtual character in a recent time period, reflecting the character's recent popularity and audience reach.

[0153] It should be noted that this embodiment focuses on obtaining the second usage characteristics corresponding to the virtual character, providing three different acquisition methods. It is not required that all three be achieved simultaneously, but rather that at least one of them be obtained. By obtaining the virtual character's win rate characteristics in recent game matches, we can understand the character's strength or weakness in recent game battles; by obtaining usage characteristics, we can grasp the character's recent usage frequency; and by obtaining user quantity characteristics, we can know how many different users have used the character recently. Combining this information, we can comprehensively reflect the recent usage status of the virtual character from different perspectives, providing rich and valuable data for subsequent adjustments to operational strategies and personalized recommendations for the virtual character.

[0154] Optionally, the content of this embodiment can be applied to other embodiments described above, or combined with other embodiments described above.

[0155] Through the embodiments provided in this application, by acquiring win rate characteristics in matches, the system can evaluate and adjust the abilities of virtual characters based on this data. If the win rate is too high, its skills or attributes can be appropriately weakened; if the win rate is too low, it can be strengthened to maintain game balance. Acquiring usage characteristics provides a clear understanding of the recent popularity of virtual characters. For characters with high usage, promotional efforts can be increased, such as launching related events or skins; for characters with low usage, the reasons can be analyzed and targeted measures can be taken to increase their attractiveness. Acquiring user number characteristics clarifies the size of the virtual character's audience. Based on the changing trends in user numbers, the character's operational direction can be adjusted, such as developing new content or functions suitable for the target audience to improve user gaming experience and engagement.

[0156] As an optional approach, obtaining the third feature corresponding to the virtual game includes:

[0157] S7-1, Obtain the activity characteristics corresponding to the virtual game, whereby the activity characteristics are used to reflect the activity level of the virtual game within a preset time period;

[0158] S7-2, Obtain the first consumption feature corresponding to the virtual game, wherein the first consumption feature is used to represent the situation in which all game users of the virtual game exchange game resource usage rights by consuming elements, and the third feature includes the first consumption feature.

[0159] S7-3, Obtain the second consumption feature corresponding to the virtual game, wherein the second consumption feature is used to represent the situation where the user exchanges elements for the right to use game resources, and the third feature includes the second consumption feature.

[0160] In an optional embodiment, a virtual game is a virtual world built on computer technology and network environment, which provides users with opportunities for entertainment, competition, social interaction, and other activities. Users complete various tasks and interactions in the game by operating virtual characters, such as role-playing games, strategy games, and competitive games.

[0161] In an optional embodiment, the third feature is a dataset used to comprehensively describe various characteristics of the virtual game, including the game's activity level and the resource consumption of users at different levels, in order to fully reflect the game's operation and user behavior.

[0162] In an optional embodiment, the activity feature is relevant data used to reflect the activity level of a virtual game within a preset time period. It is usually measured by indicators such as the number of online users, the number of game matches, and the frequency of user interaction, reflecting the popularity of the game and the degree of user participation within a specific time period.

[0163] In an optional embodiment, the preset time period is a pre-defined time interval for statistical analysis of game data, which can be set to a day, a week, a month, or a game season, etc., according to specific needs.

[0164] In an optional embodiment, consumable elements refer to virtual items or currencies that a user needs to consume in order to obtain the right to use game resources in a virtual game, such as game coins, diamonds, points, props, etc.

[0165] In an optional embodiment, the right to use game resources refers to the user's right to use and experience various resources (such as character skins, equipment, levels, functions, etc.) within the game during the use of the virtual game.

[0166] In an optional embodiment, the first consumption feature represents the relevant data of all game users in the virtual game exchanging elements for the right to use game resources, reflecting the overall situation of resource consumption by all users in the game.

[0167] In an optional embodiment, the second consumption feature represents data on how users actually use game resources to exchange elements for the right to use game resources, focusing on the resource consumption behavior of the core user group.

[0168] It should be noted that this embodiment revolves around acquiring the third characteristic corresponding to the virtual game, collecting data through three different dimensions. Acquiring the activity characteristic aims to understand the activity level of the virtual game within a preset time period, which is an important indicator for measuring the overall popularity of the game and user engagement. Acquiring the first consumption characteristic starts from the perspective of all game users, statistically analyzing how they exchange elements for the right to use game resources, reflecting the overall scale and trend of resource consumption within the game. Acquiring the second consumption characteristic focuses on users who actually use game resources, statistically analyzing how they exchange elements for the right to use resources, providing in-depth understanding of the resource consumption behavior of the core user group. Combining these three characteristics allows for a comprehensive and in-depth depiction of the virtual game's operational status and user behavior patterns, providing strong evidence for game operation, optimization, and decision-making.

[0169] Optionally, the content of this embodiment can be applied to other embodiments described above, or combined with other embodiments described above.

[0170] Through the embodiments provided in this application, after obtaining activity characteristics, the system can rationally arrange in-game operational activities based on the changing trends of activity levels. For example, it can launch limited-time offers and new gameplay during periods of low activity to attract more users and improve the overall activity level of the game. Obtaining the first consumption characteristic provides a clear understanding of the scale and preferences of resource consumption by all users in the game. The operations team can optimize resource allocation strategies based on this data to ensure the rational allocation and effective utilization of resources, thereby improving user satisfaction and retention rates. Obtaining the second consumption characteristic provides in-depth understanding of the resource consumption behavior and needs of the core user group. The system can then accurately push personalized game content, items, or services to these users, enhancing user stickiness and loyalty, and promoting the long-term stable development of the game.

[0171] As an optional approach, the activity characteristics corresponding to the virtual game can be obtained, including at least one of the following:

[0172] S8-1, Obtain the features used to represent the number of daily active users of the virtual game;

[0173] S8-2, Obtain the features used to represent the number of games in each mode of the virtual game.

[0174] In an optional embodiment, the activity feature is a set of relevant data features used to reflect the activity level of a virtual game, which measures user participation and popularity of the game within a specific time period through data from different dimensions.

[0175] In an optional embodiment, daily active users: the number of unique users who log in and participate in the virtual game in a single day, reflecting the game's daily appeal and user engagement.

[0176] In optional embodiments, each mode refers to a different game mode in a virtual game, categorized by factors such as gameplay, rules, and scenarios, such as team-based competitive mode, bomb defusal mode, and survival mode in a shooting game.

[0177] In an optional embodiment, the number of matches refers to the number of full-scale battles between players in various modes of the virtual game, reflecting the frequency and popularity of player participation in that mode.

[0178] It should be noted that this embodiment focuses on acquiring the activity characteristics of the virtual game, providing two different acquisition methods. It is not required that both methods be acquired simultaneously, but at least one must be acquired. Acquiring the daily active user count of the virtual game directly reveals how many different users log in and participate in the game each day, a key indicator for measuring the game's daily popularity and user engagement. Acquiring the number of matches played in each mode of the virtual game allows for a clear understanding of the frequency and intensity of player participation in different modes, helping to analyze the popularity of each mode and player preferences. Combining these two characteristics provides a more comprehensive and detailed picture of the virtual game's activity, offering strong data support for game operation decisions and gameplay optimization.

[0179] Optionally, the content of this embodiment can be applied to other embodiments described above, or combined with other embodiments described above.

[0180] By obtaining the characteristics of daily active users through the embodiments provided in this application, the system can analyze the patterns of user activity in different time periods based on the changing trends of this data, such as the differences between weekdays and weekends. This allows for the development of targeted daily operational strategies, such as launching limited-time events and updating content during peak user activity periods, thereby improving user engagement and retention rates. Obtaining the number of matches in each mode provides a clear understanding of the popularity of different gameplay modes. The operations team can use this data to further optimize and promote popular modes, and improve or adjust less popular modes, thereby enhancing the overall gameplay quality and player experience, and promoting the long-term healthy development of the game.

[0181] As an optional approach, the first consumption characteristic corresponding to the virtual game is obtained, including at least one of the following:

[0182] S9-1, Obtain the feature representing the proportion of users in all game users who exchange elements for the right to use game resources in a virtual game;

[0183] S9-2, Obtain the feature representing the number of users in a virtual game who exchange elements for the right to use game resources in each interval;

[0184] S9-3, Obtain the feature used to represent the average consumption of all users for the consumable element.

[0185] In an optional embodiment, the right to use game resources refers to the user's right to use and experience various resources within the game (such as character skins, equipment, levels, special functions, etc.) during the use of the virtual game.

[0186] In an optional embodiment, all game users: the entire group of users who have registered and are using the virtual game.

[0187] In an optional embodiment, the number of consumed elements in each interval is divided into different intervals according to certain rules, such as 0-10, 10-50, etc.

[0188] In an optional embodiment, average consumption is the value obtained by dividing the total consumption of all users on the consumable elements by the total number of users, reflecting the average resource consumption level of all users.

[0189] It should be noted that this embodiment revolves around obtaining the first consumption characteristic corresponding to the virtual game, providing three different acquisition methods. It is not required that all three be acquired simultaneously, but at least one must be acquired. By acquiring the percentage of users who exchange consumable elements for the right to use game resources among all game users, we can understand how many users participate in resource consumption behavior, reflecting the prevalence of this behavior. Acquiring the number of users who exchange consumable elements for the right to use game resources in each consumption range allows us to analyze the distribution of user groups from different consumption ranges, clarifying the user scale at different consumption levels. Acquiring the average consumption of consumable elements by all users allows us to grasp the average resource consumption level of all users. Combining these three characteristics provides a comprehensive and in-depth understanding of users' resource consumption within the virtual game, providing a strong basis for game operation strategy formulation and resource pricing.

[0190] Optionally, the content of this embodiment can be applied to other embodiments described above, or combined with other embodiments described above.

[0191] By obtaining the percentage of users who exchange consumable elements for game resource usage rights among all game users through the embodiments provided in this application, a low percentage indicates that the popularity of resource consumption behavior is not high, and the operator can increase promotional efforts, such as launching guidance activities and optimizing the resource acquisition process. Obtaining the number of users exchanging consumable elements for game resource usage rights in each range clearly shows the user scale at different consumption levels. High-end exclusive resources can be offered to high-spending users, while cost-effective resources can be provided to low-spending users. Obtaining the average consumption of consumable elements by all users provides insight into the overall consumption level, allowing for reasonable adjustments to resource pricing to ensure both revenue and maintain user spending enthusiasm, thus promoting the healthy operation and sustainable development of the game.

[0192] As an optional approach, the second consumption feature corresponding to the virtual game can be obtained, including at least one of the following:

[0193] S10-1, Obtain the feature representing the number of users who exchange the right to use game resources by consuming elements in each interval;

[0194] S10-2, Obtain the feature used to represent the average consumption of consumable elements by the user.

[0195] In an optional embodiment, the user refers to the user in the virtual game who actually exchanges elements for the right to use game resources.

[0196] In an optional embodiment, the number of consumed elements in each interval is divided into different ranges according to specific rules, such as 0-10, 10-50, 50-100, etc.

[0197] In an optional embodiment, consumable elements refer to virtual items or currencies that users need to consume in order to obtain the right to use game resources, such as game coins, diamonds, and points.

[0198] In an optional embodiment, the right to use game resources refers to the user's right to use and experience various resources within the game (such as character skins, equipment, special levels, functions, etc.) during the use of the virtual game.

[0199] In an optional embodiment, average consumption is a value obtained by dividing the total consumption of consumable elements by the number of users, reflecting the average resource consumption level of the user group.

[0200] It should be noted that this embodiment focuses on obtaining the second consumption characteristic corresponding to the virtual game, providing two acquisition methods, and not requiring simultaneous acquisition; at least one is sufficient. The second consumption characteristic mainly targets users, that is, the group that actually exchanges game resource usage rights. By obtaining the number of users exchanging game resource usage rights through consumption elements in each range, the distribution of users at different consumption levels can be clearly understood; by obtaining the average consumption of consumption elements by users, the average resource consumption level of the user group as a whole can be grasped. Combining these two characteristics, the resource consumption behavior patterns of the core user group can be more accurately grasped, providing key basis for the formulation of strategies targeting core users in game operation.

[0201] Optionally, the content of this embodiment can be applied to other embodiments described above, or combined with other embodiments described above.

[0202] By obtaining the characteristics of the number of users who exchange game resources for the right to use them by consuming elements in various ranges through the embodiments provided in this application, the distribution of core users at different consumption levels can be accurately analyzed. For high-spending users, exclusive high-end resources or services can be launched to enhance their game experience and loyalty; for low-spending users, cost-effective resource combinations can be provided to stimulate their consumption. By obtaining the characteristics of the average consumption of consumable elements by users, the overall resource consumption level of the core user group can be understood, and the amount of resources allocated and the pricing strategy can be adjusted accordingly to avoid resource waste or excessive pricing that leads to user churn, thereby improving the game's operational efficiency and user satisfaction.

[0203] As an optional approach, by combining the first feature, the second feature, and the third feature, the usage data of the first virtual appearance in the future time period is predicted to obtain a first prediction result, including:

[0204] S11-1, the first feature, the second feature, and the third feature are fused to obtain multi-dimensional features;

[0205] S11-2, input the multi-dimensional features into the data prediction model to obtain the first prediction result output by the data prediction model. The data prediction model is trained using multiple sample data and is a gradient regression model used to predict the usage data of appearance resources in the future time period.

[0206] In an optional embodiment, the fusion process is a process of integrating first, second, and third features of different sources and properties through a specific algorithm or method to form a multi-dimensional feature containing multiple aspects of information.

[0207] In an optional embodiment, multi-dimensional features: features obtained after fusion processing, which describe the factors affecting the use of the first virtual appearance from multiple different dimensions (such as game activity, user consumption behavior, appearance attributes, etc.).

[0208] In an optional embodiment, the data prediction model is a gradient regression model trained using multiple sample data, capable of outputting a first virtual appearance based on the multi-dimensional features of the input and using the data prediction results in the future time period.

[0209] In an optional embodiment, the sample data is historical data used to train the data prediction model, including usage data of the virtual appearance and information such as the corresponding first feature, second feature, and third feature.

[0210] In an optional embodiment, the gradient regression model is a regression model based on the gradient descent algorithm for parameter optimization. By learning from a large amount of sample data, it establishes a mapping relationship between input features and output prediction results, which is used to predict continuous target variables (here, the data used by the virtual appearance).

[0211] It should be noted that this embodiment is an overall process for predicting virtual appearance usage data based on multi-feature fusion. It first integrates the first, second, and third features (the specific content of the third feature is not mentioned above, but can be understood as other key features related to the virtual appearance), forming multi-dimensional features through fusion processing. These multi-dimensional features comprehensively reflect various factors affecting the use of virtual appearances from different perspectives. Then, the multi-dimensional features are input into a gradient regression model (data prediction model) trained with a large amount of sample data. Utilizing the model's powerful data analysis and prediction capabilities, it outputs the predicted usage data results of the first virtual appearance in the future time period, providing a forward-looking reference for game operation, resource planning, and other aspects.

[0212] Optionally, the content of this embodiment can be applied to other embodiments described above, or combined with other embodiments described above.

[0213] The embodiments provided in this application integrate the first, second, and third features to obtain multi-dimensional features, which comprehensively consider various factors affecting the use of the first virtual appearance, avoiding the limitations of a single feature. These multi-dimensional features are input into a gradient regression model trained with a large amount of sample data. The model, through learning, has mastered the inherent patterns and relationships between the input features and the virtual appearance usage data. Through the model's calculations and predictions, relatively accurate predictions of the first virtual appearance's usage data in the future can be obtained. This allows game operators to understand the usage trends of virtual appearances in advance, rationally allocate resources, and formulate marketing strategies. For example, based on the prediction results, they can decide whether to increase the acquisition channels for the appearance or conduct promotional activities, thereby improving the game's operational efficiency and user satisfaction.

[0214] As an optional approach, before inputting multi-dimensional features into the data prediction model and obtaining the first prediction result output by the data prediction model, the method also includes:

[0215] The initial data prediction model is trained multiple times until a well-trained data prediction model is obtained. The i-th training round is any one of the multiple training rounds, and the i-th training round is as follows:

[0216] S12-1, Obtain the sample data for the i-th round of training and the data prediction model for the i-th round of training;

[0217] S12-2, using the data prediction model trained in the i-th round, predict the sample data trained in the i-th round to obtain the first predicted value;

[0218] S12-3, using the residual between the true value corresponding to the sample data in the i-th round of training and the first predicted value as the objective, fit the initial decision tree;

[0219] S12-4: Maximize the information gain by splitting nodes of the initial decision tree until the stopping condition is met, and obtain the decision tree of the i-th round of training.

[0220] S12-5: Using the decision tree trained in the i-th round, update the first predicted value to obtain the second predicted value;

[0221] S12-6, if the second predicted value meets the convergence condition, the data prediction model trained in the i-th round is determined as the well-trained data prediction model.

[0222] In an optional embodiment, the initial data prediction model is an initial model framework that has not yet been trained and does not have accurate prediction capabilities. It is used for subsequent learning and optimization based on sample data to establish a mapping relationship between input features and output prediction results.

[0223] In an optional embodiment, multi-round training is a process of iteratively training the initial data prediction model. Each round of training is based on new sample data and the current model state, gradually adjusting the model parameters to improve the model's prediction accuracy.

[0224] In an optional embodiment, the i-th round of training refers to any round in a multi-round training process, where i is a positive integer representing the round number of the training. Each round of training has independent sample data and training steps.

[0225] In an optional embodiment, the sample data for the i-th round of training is a representative dataset used for the i-th round of training, which includes input features (such as multi-dimensional features) and corresponding real output values ​​(such as actual usage data of the first virtual appearance), and is used to guide the learning and adjustment of the model.

[0226] In an optional embodiment, the data prediction model for the i-th training round is the data prediction model used at the beginning of the i-th training round. It is the model obtained after the previous i-1 training rounds and is continuously optimized as the number of training rounds increases.

[0227] In an optional embodiment, the first predicted value is a preliminary prediction result obtained by using the data prediction model trained in the i-th round to predict the sample data in the i-th round, reflecting the predictive ability of the model in the current state.

[0228] In an optional embodiment, the true value is the actual output value corresponding to the sample data in the i-th training round. It is the benchmark for measuring the accuracy of the model prediction and the residual can be obtained by comparing it with the first predicted value.

[0229] In an optional embodiment, the residual is the difference between the true value and the first predicted value corresponding to the sample data in the i-th training round. It reflects the degree of deviation between the model's predicted value and the actual value and is the target for subsequent fitting of the decision tree.

[0230] In an optional embodiment, the initial decision tree is a decision tree model initially fitted with the residuals as the target. A decision tree is a model that makes decisions based on a tree structure and classifies or regresses data through a series of conditional judgments.

[0231] In an optional embodiment, the split node maximizes information gain: during the construction of the decision tree, the feature that maximizes information gain is selected as the split node. Information gain measures the degree of reduction in uncertainty of the data before and after the split. In this way, a more effective decision tree can be generated.

[0232] In an optional embodiment, the stopping condition is a condition used to determine whether to stop splitting nodes during the construction of the decision tree, such as reaching the maximum tree depth or the number of samples in a node being less than a threshold, to avoid overfitting of the decision tree.

[0233] In an optional embodiment, the decision tree trained in the i-th round is a decision tree obtained after maximizing information gain through split nodes until the stopping condition is met. It is an optimization and improvement of the initial decision tree and can better fit the residual.

[0234] In an optional embodiment, the second predicted value is a new predicted value obtained by updating the first predicted value using the decision tree trained in the i-th round. It combines the prediction results of the initial model and the decision tree and is closer to the true value.

[0235] In an optional embodiment, the convergence condition is a condition used to determine whether the model training has reached a stable state. For example, the change between the second predicted value and the previous predicted value is less than a threshold, the residual is less than a threshold, etc. When the convergence condition is met, the model is considered to have been trained.

[0236] In an optional embodiment, the trained data prediction model: after multiple rounds of training and meeting the convergence condition, the data prediction model has high prediction accuracy and can be used to predict the usage data of the first virtual appearance in the future time period.

[0237] It should be noted that this embodiment focuses on the training process of the data prediction model, which is a crucial step to ensure accurate prediction of the future usage data of the first virtual appearance. Before inputting multi-dimensional features into the data prediction model to obtain the first prediction result, the initial data prediction model needs to undergo multiple rounds of training. Each round of training has clear steps: first, obtain the sample data and the current data prediction model; use the model to predict the sample data to obtain the first predicted value; fit an initial decision tree with the residual between the actual sample value and the first predicted value as the objective; then, maximize the information gain by splitting nodes to generate the decision tree for this round; use the decision tree to update the predicted value to obtain the second predicted value; if the second predicted value meets the convergence condition, the model for this round is determined to be a well-trained model; otherwise, continue to the next round of training.

[0238] To further illustrate, optional examples include... Figure 8 As shown in the figure below, the specific steps are as follows:

[0239] S802, obtain the sample data of the i-th round and the current model;

[0240] S804, the first predicted value is obtained using the current model;

[0241] S806, use residuals to fit the initial decision tree;

[0242] S808, a decision tree obtained by splitting nodes to maximize information gain;

[0243] S810, update the first predicted value using the decision tree to obtain the second predicted value;

[0244] S812, determine whether the convergence condition is met. If yes, determine the current model as the trained model. Otherwise, continue to the next round of training, consistent with the training in the i-th round.

[0245] S814, obtain the trained data prediction model.

[0246] Optionally, the content of this embodiment can be applied to other embodiments described above, or combined with other embodiments described above.

[0247] Through the embodiments provided in this application, multi-round training iteratively optimizes the model, effectively improving the accuracy of the data prediction model. Each training round aims to fit a decision tree to the residuals, using the decision tree to update the predicted values, gradually narrowing the gap between the predicted and actual values. Maximizing information gain in split nodes ensures that the generated decision tree captures key features and patterns in the data, improving the model's generalization ability. By setting stopping and convergence conditions, overfitting is avoided, ensuring good predictive performance even with new data. The resulting well-trained data prediction model can more accurately predict the usage data of the first virtual appearance over future time periods, providing a reliable basis for game operation decisions.

[0248] As an optional approach, before correcting the first prediction result using at least one second virtual appearance to obtain the second prediction result, the method further includes:

[0249] S13-1, From the appearance resources of the virtual game, determine at least two appearance resource sets that have a relationship with the first virtual appearance, wherein different appearance resource sets correspond to different relationship types, and the appearance resource sets include the second virtual appearance;

[0250] S13-2, Based on the amount of resources in the appearance resource set, process each appearance resource set in at least two appearance resource sets using the corresponding recall method to obtain at least two recalled resource sets.

[0251] S13-3, assign corresponding fusion weights to each of the recall resource sets in at least two recall resource sets;

[0252] S13-4, according to the fusion weight, perform weighted fusion on at least two recall resource sets to obtain the first corrected data;

[0253] The method of correcting a first prediction result using at least one second virtual appearance to obtain a second prediction result includes: correcting the first prediction result using first correction data to obtain a second prediction result.

[0254] In an optional embodiment, the association is a certain connection between the first virtual appearance and other appearance resources, such as belonging to the same series, having a similar style, or being able to be used together in the game.

[0255] In an optional embodiment, the association type is a different category for classifying associations, used to distinguish associations of different natures, such as style association, function association, series association, etc.

[0256] In an optional embodiment, the appearance resource set is a set of appearance resources determined from virtual game appearance resources according to a specific association type, including at least one second virtual appearance.

[0257] In an optional embodiment, the second virtual appearance is a set of appearance resources other than the first virtual appearance, used to assist in correcting the prediction results of the first virtual appearance.

[0258] In an optional embodiment, resource quantity: the number of appearance resources contained in the appearance resource set, used to measure the size of the appearance resource set.

[0259] In an optional embodiment, the recall method is as follows: based on the amount of resources in the appearance resource set, different methods are used to extract resources from the appearance resource set to form a recall resource set. For example, random recall is used for appearance resource sets with a large amount of resources, and full recall is used for appearance resource sets with a small amount of resources.

[0260] In an optional embodiment, the recall resource set is the set of appearance resources extracted from the appearance resource set through the corresponding recall method, which is the basic data for subsequent fusion processing.

[0261] In an optional embodiment, the fusion weight is a numerical value assigned to each recall resource set to represent the importance of that recall resource set in the weighted fusion process. The larger the weight, the greater the impact of that recall resource set on the final corrected data.

[0262] In an optional embodiment, weighted fusion is the process of merging at least two recall resource sets according to the fusion weights corresponding to each recall resource set to obtain a new data set (first corrected data).

[0263] In an optional embodiment, the first corrected data is data obtained by weighted fusion of at least two recall resource sets, which is used to correct the first prediction result to improve the accuracy of the prediction.

[0264] In an optional embodiment, the second prediction result is the result obtained by correcting the first prediction result using the first correction data, which is a more accurate prediction of the future usage data of the first virtual appearance.

[0265] It should be noted that this embodiment revolves entirely around correcting the prediction results of the first virtual appearance usage data. Before correcting the first prediction result using at least one second virtual appearance, at least two appearance resource sets related to the first virtual appearance are first identified from the virtual game appearance resources. Different appearance resource sets correspond to different association types and contain second virtual appearances. Then, based on the resource quantity of each appearance resource set, a corresponding recall method is used to process each appearance resource set to obtain at least two recalled resource sets. Then, a fusion weight is assigned to each recalled resource set, and the weighted fusion is used to obtain the first corrected data. Finally, the first corrected data is used to correct the first prediction result to obtain the second prediction result, thereby improving the prediction accuracy.

[0266] Optionally, the content of this embodiment can be applied to other embodiments described above, or combined with other embodiments described above.

[0267] The embodiments provided in this application determine the set of appearance resources associated with the first virtual appearance, comprehensively considering various factors affecting its usage data. Employing different recall methods based on resource quantity ensures that the recalled resource set covers key information without excessive redundancy. Assigning fusion weights and performing weighted fusion highlights the role of important recalled resource sets, enabling the first corrected data to more accurately reflect the actual usage related to the first virtual appearance. Using the first corrected data to revise the first prediction result effectively reduces prediction errors, improves the accuracy and reliability of predicting future usage data for the first virtual appearance, and provides more valuable references for resource management and operational decisions in virtual games.

[0268] As an optional approach, based on the resource quantity in each of the at least two appearance resource sets, a corresponding recall method is applied to each appearance resource set in the remaining two appearance resource sets to obtain at least two recalled resource sets, including:

[0269] S14-1 For a first resource set with a resource quantity greater than or equal to a preset threshold, remove the highest first proportion and the lowest second proportion from the first resource set, and take the average value of the remaining samples as the representative value of the recall resource set to obtain the first recall resource set.

[0270] S14-2, For the second resource set whose resource quantity is less than the preset threshold and greater than 0, determine the mean of all samples in the second resource set, and adjust the mean according to the resource quantity to obtain the second recall resource set.

[0271] In an optional embodiment, the appearance resource set is a set of appearance resources determined from virtual game appearance resources that are associated with the first virtual appearance, and includes multiple sample data.

[0272] In an optional embodiment, resource quantity: the number of samples contained in the appearance resource set, used to measure the size of the appearance resource set.

[0273] In an optional embodiment, a preset threshold is defined as a pre-set value used to classify the size of the appearance resource set and serve as the basis for selecting different recall methods.

[0274] In an optional embodiment, the first resource set is an appearance resource set whose resource quantity is greater than or equal to a preset threshold.

[0275] In an optional embodiment, the first proportion, which is the proportion of the highest-value samples to be removed from the first resource set, is a preset value.

[0276] In an optional embodiment, the second ratio, which is the proportion of the lowest-value samples to be removed from the first resource set, is a preset value.

[0277] In an optional embodiment, the endpoints are the highest and lowest values ​​of the sample data in the first resource set.

[0278] In an optional embodiment, the first recall resource set is the average value of the remaining samples after removing the extreme values ​​from the first resource set, which is then used as the representative value.

[0279] In an optional embodiment, the second resource set is an appearance resource set whose resource quantity is less than a preset threshold and greater than 0.

[0280] In an optional embodiment, the sample mean is the average value of all sample data in the second resource set.

[0281] In an optional embodiment, the second recall resource set is a recall resource set obtained by adjusting the sample mean based on the resource quantity of the second resource set.

[0282] It should be noted that this embodiment processes at least two sets of appearance resources determined from virtual game appearance resources to obtain at least two recall resource sets, providing a data basis for subsequent correction of the prediction results of the first virtual appearance. Different recall methods are used for appearance resource sets with different resource quantities: for the first resource set with a resource quantity greater than or equal to a preset threshold, the highest and lowest proportions are first removed, and then the average value of the remaining samples is taken as the representative value to obtain the first recall resource set; for the second resource set with a resource quantity less than the preset threshold but greater than 0, the mean of all samples is first determined, and then the mean is adjusted according to the resource quantity to obtain the second recall resource set.

[0283] Optionally, the content of this embodiment can be applied to other embodiments described above, or combined with other embodiments described above.

[0284] Through the embodiments provided in this application, for a first resource set with a large amount of resources, removing extreme values ​​can effectively avoid the impact of abnormal data on the overall data. Taking the average of the remaining samples as a representative value can yield a more stable recall resource set that better reflects the overall characteristics. For a second resource set with a smaller amount of resources, first determining the sample mean as a basis, and then adjusting the mean according to the amount of resources, can make full use of the limited data, making the resulting second recall resource set more consistent with the actual situation. Through these two different recall methods, appearance resource sets of different sizes can be reasonably processed, providing accurate and reliable data for subsequent correction of prediction results, thereby improving the accuracy of prediction of the first virtual appearance usage data.

[0285] As an optional approach, a corresponding fusion weight is assigned to each of the recall resource sets in at least two recall resource sets, including at least one of the following:

[0286] S15-1, Based on the importance of each appearance resource in the recall resource set, assign a corresponding first weight to each appearance resource, wherein the importance is positively correlated with the first weight, and the fusion weight includes the first weight;

[0287] S15-2, based on the number of appearance resources in each recall resource set, assign a corresponding second weight to each appearance resource, wherein the number of resources and the second weight are positively correlated, and the fusion weight includes the second weight;

[0288] S15-3, based on the dispersion of appearance resources in each recall resource set, assign a corresponding third weight to each appearance resource, wherein the dispersion is positively correlated with the third weight, and the fusion weight includes the third weight.

[0289] In an optional embodiment, the recall resource set is the resource set obtained from the appearance resource set through different recall methods mentioned above, which is used for subsequent weighted fusion to correct the prediction results.

[0290] In an optional embodiment, the fusion weight is a numerical value assigned to each recall resource set to measure its importance in the weighted fusion process and determine its impact on the final fusion result.

[0291] In an optional embodiment, appearance resources are data used in virtual games to change the appearance of characters, items, etc., such as character skins and the appearance of attack items.

[0292] In an optional embodiment, importance is a comprehensive reflection of the appearance resource's value, popularity, and impact on the game experience in the virtual game. The higher the importance, the more critical the appearance resource is.

[0293] In an optional embodiment, the first weight is a weight assigned based on the importance of the appearance resource. The importance is positively correlated with the first weight, that is, the higher the importance, the greater the first weight.

[0294] In an optional embodiment, resource quantity: the number of appearance resources contained in the recall resource set, reflecting the size of the recall resource set.

[0295] In an optional embodiment, the second weight is a weight assigned based on the number of appearance resources. The number of resources is positively correlated with the second weight, that is, the more resources there are, the greater the second weight.

[0296] In an optional embodiment, the degree of dispersion: the degree of dispersion of appearance resource data in the recall resource set, which can be measured by statistical measures such as variance and standard deviation. The greater the degree of dispersion, the more dispersed the data.

[0297] In an optional embodiment, the third weight is a weight assigned according to the degree of dispersion of the appearance resource. The degree of dispersion is positively correlated with the third weight, that is, the greater the degree of dispersion, the greater the third weight.

[0298] It should be noted that this embodiment focuses on assigning fusion weights to each recall resource set in at least two recall resource sets, and provides three allocation methods, which can be used individually or in combination. The first method assigns a first weight based on the importance of the appearance resources in the recall resource set, with higher importance resulting in a larger weight; the second method assigns a second weight based on the number of appearance resources, with more resources resulting in a larger weight; the third method assigns a third weight based on the dispersion of appearance resources, with greater dispersion resulting in a larger weight. The final fusion weight may include one or more of these weights.

[0299] Optionally, the content of this embodiment can be applied to other embodiments described above, or combined with other embodiments described above.

[0300] The embodiments provided in this application allocate a first weight based on importance, enabling key appearance resources to play a greater role in the fusion process and highlighting the impact of important information on the final result. Allocating a second weight based on resource quantity ensures that a rich set of recalled resources contributes appropriately to the fusion result, avoiding unreasonable impacts caused by differences in resource quantity. Allocating a third weight based on dispersion recognizes that a highly dispersed set of recalled resources may contain more unique information, and assigning it a larger weight allows for full utilization of this information. By comprehensively utilizing these weight allocation methods, the fusion weights become more scientific and reasonable, improving the accuracy of weighted fusion and thus enhancing the reliability of correcting the first virtual appearance prediction result.

[0301] As an alternative approach, after identifying at least two sets of appearance resources that are associated with the first virtual appearance from the appearance resources of the virtual game, the method further includes:

[0302] Based on the amount of resources in the appearance resource set, each appearance resource set in at least two appearance resource sets is processed using the corresponding recall method. If 0 recall resource sets are obtained, the second correction data is obtained based on the usage data of the appearance resources in the virtual game that belong to the same appearance quality as the first virtual appearance.

[0303] The method of correcting a first prediction result using at least one second virtual appearance to obtain a second prediction result includes: correcting the first prediction result using second correction data to obtain a second prediction result.

[0304] In an optional embodiment, appearance quality: a classification standard for appearance resources in virtual games, used to distinguish appearances of different qualities, rarity, or popularity, where appearances of the same appearance quality are similar in certain characteristics.

[0305] In an optional embodiment, the data used is: usage data of appearance resources in virtual games, such as usage frequency, usage duration, usage scenarios, etc., which can reflect the popularity and usage patterns of appearance resources.

[0306] In an optional embodiment, the second correction data is data obtained from usage data that has the same appearance quality as the first virtual appearance when the recall resource set cannot be obtained through the recall method, and is used to correct the first prediction result.

[0307] It should be noted that this embodiment revolves entirely around correcting the prediction result of the first virtual appearance. After identifying at least two appearance resource sets related to the first virtual appearance from the virtual game appearance resources, if processing using a corresponding recall method based on the resource quantity of the appearance resource sets yields 0 recall resource sets (i.e., valid data cannot be obtained through conventional recall methods), the strategy is changed. Second corrected data is obtained based on usage data from the virtual game appearance resources that belong to the same appearance quality as the first virtual appearance. Finally, the first prediction result is corrected using the second corrected data to obtain the second prediction result, thereby improving the prediction accuracy.

[0308] Optionally, the content of this embodiment can be applied to other embodiments described above, or combined with other embodiments described above.

[0309] The embodiments provided in this application demonstrate that when a recall resource set cannot be obtained through conventional recall methods, a method of obtaining second correction data based on usage data of similar appearance quality can effectively utilize existing relevant data resources in the virtual game. Appearances of similar appearance quality share certain similarities in usage characteristics, and their usage data can provide valuable reference information, thereby avoiding situations where the inability to obtain a recall resource set leads to the inability to correct prediction results. Correcting the first prediction result using the second correction data can improve the consistency between the prediction result and the actual situation, enhance the accuracy and reliability of the prediction, and provide more effective support for resource management and operational decisions in the virtual game.

[0310] As an optional approach, for ease of understanding, the aforementioned method for predicting the use of virtual appearances is applied to the scenario of predicting the sales of skins (virtual appearances) in Multiplayer Online Battle Arena (MOBA) games (virtual games). This embodiment proposes a MOBA game skin sales prediction algorithm that integrates multi-dimensional features and strategy correction to solve the problem of predicting the sales of virtual items (skins) in MOBA games, especially for cold start scenarios such as new hero (virtual character) skins and new skin series.

[0311] Specifically, this embodiment uniformly constructs temporal features (such as holiday cycles), categorical features (such as quality and series), statistical features (such as historical hero skin sales), and behavioral features (such as hero usage frequency), establishing a high-dimensional sparse-dense hybrid vector space that supports automatic feature interaction via XGBoost. Through a systematic feature engineering method, it achieves a comprehensive characterization of multidimensional heterogeneous information affecting skin sales. This embodiment does not rely on a fixed similarity threshold but adaptively adjusts the weights based on the sample size of each recall candidate set. When the recall results of a certain path have sufficient samples, a truncated mean is used to reduce the impact of outliers; when the samples are sparse, a sample variance penalty term is introduced to reduce confidence. This mechanism effectively solves the stability problem of relevance similarity recommendation methods under uneven sample distribution.

[0312] This embodiment also uses the XGBoost model's predicted values ​​and the corrected values ​​from similar sample recall as inputs to train a linear regression model and output the final result. The fusion model can automatically learn the credibility weights of each input source, relying more on model predictions in scenarios with sufficient data and more on policy corrections in cold-start scenarios.

[0313] The entire framework retains intermediate outputs (model predictions and various correction values), allowing planners to trace the prediction results back to their attribution, meeting the interpretability requirements of "why this sales volume was predicted" in game operation scenarios, and facilitating business decisions and strategy adjustments.

[0314] It should be noted that this embodiment supports high-precision prediction of new skin sales in MOBA games, serving core business scenarios such as product pricing (optimizing skin pricing strategies based on predicted sales to balance revenue and sales), marketing strategies (predicting high-potential skins and rationally allocating promotional resources), inventory management (planning production and inventory in advance for physical co-branded products), R&D resources (identifying high-return heroes / series and guiding skin development priorities), and event planning (selecting event skins based on sales predictions to maximize event revenue).

[0315] To further illustrate, this embodiment provides a sales prediction method that integrates multi-dimensional features and similar sample strategy correction. It aims to achieve high-precision prediction of new skin sales in MOBA games through comprehensive analysis of various data features and strategy correction. Optional examples include... Figure 9 As shown, the specific steps are as follows:

[0316] S902 begins with the construction of a multi-dimensional feature vector. In this stage, a comprehensive collection of various data related to game skin sales is undertaken, covering skin attributes (such as appearance design, associated hero, rarity, etc.), game operation data (such as hero usage rate, win rate, and related event information), market environment data (such as the market performance of similar game skins and player spending trends), and user behavior data (such as players' purchase history and collection preferences for skins of similar styles). This data from different dimensions is then cleaned, transformed, and integrated to construct a multi-dimensional feature vector that comprehensively reflects the factors influencing skin sales, providing a rich data foundation for subsequent model training.

[0317] S904, next we proceed with the training of the XGBoost basic model. Using the constructed multi-dimensional feature vectors as input data, the XGBoost algorithm is employed for model training. XGBoost, as an efficient ensemble learning algorithm, has advantages such as handling large-scale data, automatically handling missing values, and supporting regularization. It can effectively uncover complex relationships between features, making preliminary predictions of skin sales and laying the foundation for subsequent corrections and optimizations.

[0318] S906, and then a multi-path recall strategy is implemented for correction. Considering the differences in characteristics among different skins and the dynamic changes in the market environment, the prediction of a single model may have certain biases. Therefore, a multi-path recall strategy is introduced to recall sample data with similar characteristics to the skin to be predicted from multiple perspectives (such as historical sales data of similar skins, popularity of skins of the same type of hero, etc.). Through in-depth analysis of these similar sample data, the prediction results of the XGBoost basic model are corrected to improve the accuracy and adaptability of the prediction.

[0319] S908, followed by adaptive weight calculation. Based on the multi-path recall strategy correction, an adaptive weight calculation method is adopted to more reasonably integrate the impact of information from different sources on the final prediction result. Weights are dynamically assigned to different recall results based on factors such as the correlation and reliability of each recalled sample data with the current skin to be predicted, ensuring that the advantages of each source's information are fully utilized during the fusion process, further improving the accuracy of the prediction.

[0320] S910, finally completing the hierarchical fusion prediction. The results from each path, after adaptive weight calculation, are hierarchically fused, integrating information from multiple dimensions, similar sample correction, and adaptive weight allocation to arrive at the final prediction result for new game skin sales. This result provides game operations teams with a scientific and reliable basis for decision-making in multiple core business scenarios such as product pricing, marketing strategy formulation, inventory management, R&D resource allocation, and event planning, contributing to the sustainable development of the game business.

[0321] Optionally, in this embodiment, for the construction of multi-dimensional feature vectors, firstly, through multi-dimensional feature engineering, a systematic integration of various structured data such as skin release time information, basic skin attributes, hero-related features, and player spending behavior is carried out, mainly including:

[0322] 1. Skin release time characteristics: This includes the specific date, day of the week, month, and quarter in which the skin is released, as well as whether it covers statutory holidays or important game event cycles, in order to capture the impact of time periodicity and holiday effects on skin sales.

[0323] 2. Basic attributes and characteristics of the skin: including the hero's lane, role, skin series, quality, price, and whether it is a low-cost / partial skin, reflecting the skin's own attractiveness and characteristics.

[0324] 3. Historical Statistical Characteristics of Hero Skins: Before the skin is released, the number of skins that have been released for this hero, the historical average sales of skins of each quality, and the average number of skins of each quality owned by active players for this hero in the past month are used to measure the hero's popularity and the scarcity and market saturation of the skins.

[0325] 4. Hero activity characteristics: For example, segmenting modes and statistics on the win rate, usage frequency, and number of players who have used the hero corresponding to the skin in the past month to measure the hero's popularity and player attention.

[0326] 5. Overall market activity characteristics: such as the number of daily active users (DAU) in the past month, the number of matches in each mode, etc., reflecting the overall game activity and market size.

[0327] 6. Player spending characteristics: For example, the monthly spending percentage of all players up to the day before the skin is released, the number of players with different spending abilities, and the average spending per player, which reflect the market's spending power and spending structure.

[0328] 7. Hero consumption characteristics: For example, among players who have used this hero in the past month, the number of players with different spending abilities and the average spending per person can be used to measure the consumption potential and willingness to pay of the hero's user group.

[0329] It should be noted that these seven types of data are highly consistent with the current state of data tracking in MOBA games. They are all standardized data that naturally exists in the log system and is automatically calculated daily, making them inexpensive to obtain.

[0330] Furthermore, multi-dimensional cross-referencing can significantly improve prediction accuracy. Skin sales are not determined by a single factor, and the strength of tree models such as XGBoost lies in handling such high-order feature cross-referencing. These 7 types of features cover macro (overall DAU) to micro (single hero payment rate), static (skin tags) to dynamic (win rate in the past month), which can approximate the real player's purchasing decision logic as closely as possible.

[0331] To provide guidance on "strategy adjustment" and "interpretability," this embodiment mentions "multi-path recall" and "SHAP value attribution." If the feature dimensions are not rich enough, the recall strategy cannot be implemented (for example, the fifth path recall requires "same DAU range"). Subdividing these 7 types of features allows business personnel to analyze the principles of sales forecasting during post-mortem analysis.

[0332] Optionally, in this embodiment, for training the XGBoost base model, the training data is first prepared by collecting feature vectors of historical skins and 14-day cumulative sales data to construct a training set. ,in For the first Feature vector of the skin For the first The cumulative sales of this skin in the 14 days after its release.

[0333] The core optimization objective of the XGBoost regression model is:

[0334]

[0335] in, It is the predicted value of the i-th sample. This is the prediction value of the k-th decision tree for the i-th sample, where K is the total number of decision trees. For regularization, y is the sum of the sample weights of the smallest leaf nodes in the tree (controlling the complexity of the tree). It is the number of leaf nodes in the k-th decision tree. It is the L2 regularization coefficient (which controls the smoothness of the weights of leaf nodes).

[0336] Huber Loss is used to enhance robustness to outlier sales values:

[0337]

[0338] in, This is the threshold parameter, set to 1.0 by default, and can be adjusted according to the data distribution. y is the actual value. This is a predicted value.

[0339] Optionally, the hyperparameter configuration in this embodiment is as follows: Figure 10As shown, n_estimators is set to 500, representing the number of decision trees in the model; max_depth is 6, used to limit the maximum depth of the trees; learning_rate is 0.05, controlling the step size of model updates in each iteration; subsample is 0.8, meaning that 80% of the samples are randomly selected for training when building each decision tree; colsample_bytree is 0.8, meaning that 80% of the features are randomly selected for node splitting when building each decision tree; reg_alpha is set to 0.1, which is the coefficient for applying L1 regularization constraints to the model parameters; reg_lambda is 1.0, which is the L2 regularization coefficient; min_child_weight is 5, and node splitting stops when the sum of the sample weights of the leaf nodes is less than this value.

[0340] Optionally, in this embodiment, for the training process, the data is first divided into a training set (80%), a validation set (10%), and a test set (10%) according to the time sequence; then the model is trained, the XGBoost model is fitted using the training set, and an early stopping strategy is implemented using the validation set.

[0341] Then, feature importance is calculated, and the SHAP value of each feature is output. The calculation formula is Φ_j=Σ_{S N{j}}|S|!(|N|-|S|-1)! / |N|!×[f(S∪{j})-f(S)],where, j is the SHAP value of feature j, N is the set of all features, S is a feature subset that does not contain feature j, f(S) is the model's prediction on feature subset S, and f(S∪{j}) is the model's prediction after adding feature j to feature subset S.

[0342] Next, the model is evaluated by calculating metrics such as MAPE, RMSE, and R² on the test set; finally, the model is saved by serializing the model parameters for online prediction.

[0343] Optionally, in this embodiment, for the multi-path recall strategy correction, for the target new skin, multi-path recall rules are executed in parallel to filter historically similar skins, providing empirical anchor points to correct model predictions. The recall strategy is as follows: Figure 11As shown, the recall strategy definition includes multiple recall rules and their corresponding business implications: The first recall rule is for the same hero, meaning all historical skins of the same hero will be recalled; the second recall rule is for the same skin series, meaning all historical skins of the same skin series will be recalled; the third recall rule is for the same quality, meaning all historical skins of the same quality level will be recalled; the fourth recall rule is for the same price range, meaning historical skins with price differences within ±10% will be recalled; the fifth recall rule is for the same DAU range, meaning historical skins with DAU differences within ±5% will be recalled upon launch; the sixth recall rule is for the same hero and the same quality, meaning historical skins of the same hero and the same quality will be recalled; the seventh recall rule is for the same series and the same quality, meaning historical skins of the same series and the same quality will be recalled.

[0344] Optionally, for processing the recall results, for each recall result set... Different statistics are used depending on the sample size:

[0345] Case 1: For samples ≥ 10, use the truncated mean, removing the highest and lowest 10% of samples and then taking the average:

[0346]

[0347] in, is the truncated mean of the k-th recall (the statistic in case 1, used for large sample stability estimation), m is the total number of samples in the k-th recall (must satisfy m≥10, otherwise use case 2), 0.1m is the number of one-sided extreme samples (remove the highest / lowest 0.1m samples, such as removing 1 sample when m=15). It is the i-th sample value after sorting (the core of truncation: removing extreme values ​​at both ends and keeping 80% of the samples in the middle), and the summation range is from i... arrive (Index of the remaining samples after truncation), the denominator is the number of valid samples after truncation (total number of samples minus the 20.1m samples removed from both sides).

[0348] Case 2: For sample sizes between 1 and 9, use the ordinary mean multiplied by the sample size penalty factor:

[0349]

[0350] in, is the adjusted mean of the k-th recall (the statistic in case 2, used for small sample penalty), and m is the total number of samples in the k-th recall (must satisfy 1≤m≤9, otherwise use case 1 or 3). It is the i-th original sample value of the k-th recall (unsorted, directly taking the original score of the recall result). It is the ordinary arithmetic mean of the k-th recall (the average of all samples). It is the sample size penalty coefficient (which increases with increasing m: 0.1 when m=1, 0.9 when m=9, penalizing the random fluctuations of small samples).

[0351] Case 3: If the number of samples is 0, the recall is invalid and will not participate in subsequent fusion.

[0352] Optionally, in this embodiment, a fusion weight is calculated for each valid policy correction result, and the weight consists of three parts.

[0353] 1. Prior weights based on feature importance:

[0354] Based on the feature importance (SHAP value) output by the XGBoost model, calculate the sum of the importance of the features involved in each recall loop:

[0355]

[0356] in, It is the prior weight of the k-th path recall (the "basic weight" based on the importance of the feature; the more important the feature, the higher the weight). It is the feature set involved in the k-th recall rule (e.g., if the k-th recall uses "price + sales volume"), then ={price, sales volume}). It is the SHAP value of feature j (the feature importance output by the XGBoost model, reflecting the marginal contribution of feature j to the prediction result), ∑j∈ It is the sum of the importance strengths of all features involved in the k-th path (the more important / more features the path depends on, the greater the prior weight).

[0357] 2. Log-weighted weights based on sample size:

[0358] The larger the sample size, the higher the confidence level.

[0359]

[0360] in, Rk is the sample size weight of the k-th recall (the larger the sample size, the higher the weight, reflecting the "statistical confidence"), and Rk is the result set of the k-th recall (containing the original scores / predicted values ​​of m samples, i.e. ={y1,y2,…,ym}). It is the sample size recalled by the k-th path ( The number of elements in the middle, i.e., m).

[0361] 3. Penalty weights based on sample dispersion:

[0362] The more dispersed the sample (the greater the variance), the lower the confidence level.

[0363]

[0364] in, =μk / σk (coefficient of variation) is the dispersion penalty weight of the k-th recall (the more dispersed the samples, the lower the weight, reflecting "unstable results"), CVk is the coefficient of variation of the k-th recalled samples (relative dispersion, σk is the standard deviation, μk is the mean), σk is the standard deviation of the k-th recalled samples (absolute dispersion, measuring the magnitude of sample fluctuation), and μk is the mean of the k-th recalled samples (μtrim(k) or μadj(k) calculated by the "Recall Result Processing" module). It is a penalty function for discreteness ( The larger the value, the larger the denominator and the smaller the weight. When the value is 0, the weight is 1, and there is no penalty.

[0365] Normalize after multiplying the three weights:

[0366]

[0367] Where wk is the final fusion weight of the k-th recall (the product of the three weights, which is the sum of the weights of all valid paths after normalization), vj is the validity indicator of the j-th recall (vj=1 indicates validity, vj=0 indicates invalidity; only valid paths participate in the summation of the denominator), the numerator is the product of the three weights of the k-th path (prior weight × sample size weight × dispersion weight, which comprehensively reflects "reliability + importance"), and the denominator is the sum of the products of the three weights of all valid recall paths (normalization operation).

[0368] Optionally, if all recall circuits are invalid ( If the prediction is downgraded, the median sales volume of all products of the same quality will be used as the strategy adjustment value, and the prediction will be marked as low confidence.

[0369]

[0370] in, It is the downgrade strategy correction value (the "safety net" value when all recall loops are invalid, marked as low confidence). It is the quality attribute of the i-th sample (such as the "superior / qualified / inferior" grade of the product). It is the target quality (the quality level that needs to be predicted at present, such as "sales volume of high-quality products"), { t} is the set of all samples whose quality is equal to the target quality (filtering out historical samples that are consistent with the current prediction target), and median is the median function (taking the middle value after sorting the set, which is more robust than the mean and less susceptible to extreme values). It is the sales volume / predicted value of the i-th sample (of which the quality is 1 in historical samples). (sales volume).

[0371] Optionally, in this embodiment, the predicted values ​​of the XGBoost model and the policy correction values ​​are used as meta-features to train a secondary linear regression model to output the final prediction.

[0372] The input feature vector for constructing the fusion model:

[0373]

[0374] in These are predictions from the XGBoost model. It is the weighted strategy adjustment value. It is the model prediction confidence (based on the model's historical error on similar samples). It is the strategy correction confidence (number of effective recall loops).

[0375] Perform linear fusion:

[0376]

[0377] in, yes, It is the intercept term (the constant term in linear regression, which captures the base prediction when "all features are 0", such as the default sales when there is no model / no strategy). It is the coefficient of the XGBoost prediction (measures the influence of the base model's prediction on the final result: the larger β1 is, the higher the weight of the base model). It is the coefficient of the strategy correction value (measures the degree of influence of the multi-path recall fusion result on the final result: the larger the β2, the higher the weight of the strategy correction). It is the model confidence coefficient (measures the degree to which the reliability of the base model adjusts the final result: the higher the confxgb, the greater the positive impact of β3). It is the coefficient of policy confidence (measures the degree to which the reliability of policy modification adjusts the final result: the higher the confidence, the greater the positive impact of β4).

[0378] The model was trained using the least squares method.

[0379]

[0380] in, The optimization objective is to find the optimal coefficient vector β that minimizes the loss function. It is the sum of squared residuals (the sum of the true value yi of the i-th sample and the model prediction). The sum of squared differences measures how well the model fits the training data. It is the model prediction value of the i-th sample (the inner product of the feature vector Zi and the coefficient vector β, i.e., the prediction value of linear regression). It is the regularization coefficient (a hyperparameter that needs to be tuned through cross-validation: when λ=0, it degenerates into ordinary least squares; when λ is too large, the model is underfitting).

[0381] Optionally, in this embodiment, regarding the adaptability of the fusion strategy, the model performance and strategy performance vary in different scenarios, and the fusion results will also differ accordingly. For example... Figure 12 As shown, when there is sufficient historical data, the model exhibits high confidence and the strategy has reference value. In this case, the fusion mainly relies on the model. In the cold start scenario, the model exhibits low confidence, while the strategy has high confidence, and the fusion mainly relies on the strategy. If the data is balanced, both the model and the strategy exhibit moderate performance, and the fusion result will adopt a balanced fusion approach.

[0382] Optionally, in this embodiment, regarding the online prediction process, firstly, real-time feature calculation is performed, receiving basic information about the new skin and capturing dynamic features in real time; then, model inference is carried out, inputting the feature vector into the XGBoost model to obtain the predicted value y_xgb; then, policy correction is implemented, performing multi-path recall and calculating the policy correction value μ_strategy; then, fusion output is performed, inputting relevant data into the fusion model to obtain the final predicted value y_final; finally, the results are packaged, outputting the predicted value, confidence interval, and attribution information.

[0383] The feedback loop mechanism is activated after a new skin is launched. Real sales data S_1, S_2, ..., S_14 are collected daily. On the 14th day, the prediction error is calculated as error = |y_final - S_14| / S_14. If the error exceeds a set threshold (e.g., 30%), anomaly detection is triggered, and the errors of model prediction and strategy correction are analyzed to pinpoint the main sources of deviation. Subsequently, sample weighting is performed, and failed samples are added to the high-weight training set with weight w = 1 + error. Finally, the model is retrained periodically (e.g., monthly) to achieve continuous optimization.

[0384] The model iteration strategy adopts corresponding processing methods according to different triggering conditions: when the cumulative samples increase by 10%, full retraining is performed; if the error of three consecutive samples is greater than 30%, incremental learning is used for updating; at quarterly time nodes, retraining and parameter tuning are forced; when encountering major version updates, the features are re-engineered.

[0385] In an optional embodiment, multidimensional feature modeling refers to the systematic extraction and encoding of spatiotemporal features, static attribute features, and dynamic behavioral features that affect the sales of virtual items, and the construction of a structured feature vector space to characterize cross-domain heterogeneous information such as skin release time, skin basic attributes, hero activity, and player consumption behavior.

[0386] In an optional embodiment, the strategy is modified by screening a set of historical samples that are highly similar to the target skin through a multi-path recall mechanism, and using the actual sales statistics (mean, quantiles, etc.) of the samples in the set to weight and calibrate the preliminary prediction results of the machine learning model in order to make up for the data sparsity problem in the cold start scenario.

[0387] In an optional embodiment, the cold start problem refers to the phenomenon that the generalization ability of the relevant supervised learning model is reduced due to feature distribution shift when there are no historical training samples, such as the first skin of a new hero, a brand new quality level, or an attribute combination that has never appeared before.

[0388] In an optional embodiment, multi-path recall is a method that generates multiple candidate historical skin sets by executing a screening strategy based on one or more feature dimensions (such as hero ID, skin series, price range) in parallel.

[0389] In an optional embodiment, sales forecast fusion: The output value of the machine learning model and the correction value of the multi-way recall strategy are used as meta-features and input into the secondary linear regression model for interpretable feature importance: The marginal contribution of each dimension feature to the sales forecast target is quantified through the built-in evaluation mechanism of the XGBoost model or SHAP value calculation, which is used to guide the priority ranking of recall conditions in the strategy correction stage.

[0390] In an optional embodiment, XGBoost is an ensemble learning algorithm based on decision trees. Its objective function is: , where L ( ) is the overall objective function of the XGBoost model (the loss to be minimized). Represents all parameters of the model, such as the structure of the decision tree, the weights of the leaf nodes, etc.), ∑i is the summation of all training samples (i is the sample index, covering all n training samples), l ( (yi) is a sample-level loss function (measuring the predicted value of the i-th sample). The error from the true value yi, such as Huber Loss, squared loss, etc.), ∑k is the summation of all decision trees (k is the decision tree index, covering all K base learners of XGBoost). For loss function, This is a regularization term.

[0391] In an optional embodiment, Huber Loss: a loss function robust to outliers, defined as: when hour, ;when hour, , where Lδ(y, ) is the Huber loss function (a regression loss robust to outliers, where δ is the threshold parameter). δ is the norm of the prediction error, and δ is the robust threshold parameter, which controls the "switching point" of the loss function. The default value is usually 1.0, which can be adjusted according to the data distribution.

[0392] In an optional embodiment, the SHAP value is a feature importance interpretation method based on game theory Shapley values, which quantifies the marginal contribution of each feature to the prediction result.

[0393] In an optional embodiment, the truncated mean is the mean calculated after removing a certain percentage of the highest and lowest extreme values ​​in the dataset, used to reduce the impact of outliers on the statistics.

[0394] In an optional embodiment, Daily Active Users (DAU) is a core metric for measuring the number of daily active players on a gaming platform.

[0395] In an optional embodiment, skin: a cosmetic item that enhances the appearance of a hero character in the game.

[0396] In an optional embodiment, quality tier: a classification of skin rarity, typically including epic, legendary, limited, and other tiers.

[0397] In an optional embodiment, a skin series is a group of skins that share the same theme.

[0398] In an optional embodiment, hero lanes: the hero's position in the game, including top lane, mid lane, marksman, jungler, and support.

[0399] In an optional embodiment, a companion skin is a skin that a hero first obtains in the game or that is released at the same time as the hero.

[0400] Optionally, the content of this embodiment can be applied to other embodiments described above, or combined with other embodiments described above.

[0401] The embodiments provided in this application significantly improve the accuracy of new skin sales prediction, especially in cold start and data-sparse scenarios. This embodiment combines data-driven approaches with experience-based adjustments, allowing for flexible adaptation to dynamic changes in the market environment and player behavior. Furthermore, its prediction results are highly interpretable, facilitating understanding and strategy adjustments by business personnel. In addition, it possesses good scalability and versatility, enabling migration to other virtual goods prediction scenarios. Evaluation using real data shows that the MAPE (Mean Absolute Percentage Error) of the fusion model in this embodiment is approximately 25%.

[0402] It is understood that in the specific embodiments of this application, data such as user information are involved. When the above embodiments of this application are applied to specific products or technologies, user permission or consent is required, and the collection, use and processing of related data must comply with the relevant laws, regulations and standards of the relevant countries and regions.

[0403] It should be noted that, for the sake of simplicity, the foregoing method embodiments are all described as a series of actions. However, those skilled in the art should understand that this application is not limited to the described order of actions, as some steps may be performed in other orders or simultaneously according to this application. Furthermore, those skilled in the art should also understand that the embodiments described in the specification are preferred embodiments, and the actions and modules involved are not necessarily essential to this application.

[0404] According to another aspect of the embodiments of this application, a virtual appearance usage prediction apparatus for implementing the above-described virtual appearance usage prediction method is also provided. For example... Figure 13 As shown, the device includes:

[0405] The first acquisition unit 1302 is used to acquire the first feature corresponding to the first virtual appearance, wherein the first virtual appearance is used to change the appearance of the virtual character in the virtual game, and the first feature is used to represent the attribute data and time information associated with the first virtual appearance.

[0406] The second acquisition unit 1304 is used to acquire the second feature corresponding to the virtual character, wherein the second feature is used to represent the appearance resources of the virtual character and the usage data of the virtual character;

[0407] The third acquisition unit 1306 is used to acquire the third feature corresponding to the virtual game, wherein the third feature is used to represent the activity data of the virtual game and the usage information of game resources by game users in the virtual game. Game users include users who use virtual characters, and game resources include the appearance resources of virtual characters.

[0408] The prediction unit 1308 is used to combine the first feature, the second feature, and the third feature to predict the usage data of the first virtual appearance in a future time period and obtain the first prediction result.

[0409] The correction unit 1310 is used to correct the first prediction result by using at least one second virtual appearance to obtain a second prediction result when at least one second virtual appearance is determined from the appearance resources of the virtual game based on the first feature, the second feature, and the third feature, and satisfies the relevant conditions with the first virtual appearance.

[0410] For specific implementation examples, please refer to the examples shown in the above virtual appearance usage prediction method, which will not be repeated here.

[0411] As an optional solution, the first acquisition unit 1302 includes:

[0412] The first acquisition module is used to acquire the attribute features corresponding to the first virtual appearance, wherein the attribute features are used to represent the first attribute data of the first virtual appearance and the second attribute data of the virtual character, and the first feature includes the attribute features.

[0413] The second acquisition module is used to acquire the time feature corresponding to the first virtual appearance, wherein the time feature is used to represent the opening time of the right to use the first virtual appearance, and the first feature includes the time feature.

[0414] For specific implementation examples, please refer to the examples shown in the above virtual appearance usage prediction method, which will not be repeated here.

[0415] As an optional approach, the first acquisition module includes at least one of the following:

[0416] The first acquisition submodule is used to acquire features that represent the appearance series to which the first virtual appearance belongs;

[0417] The second acquisition submodule is used to acquire features that represent the appearance quality of the first virtual appearance.

[0418] The fourth acquisition submodule is used by the third acquisition submodule to acquire features that represent the appearance type to which the first virtual appearance belongs;

[0419] The fifth acquisition submodule is used to acquire the feature representing the amount of elements required to obtain the right to use the first virtual appearance.

[0420] For specific implementation examples, please refer to the examples shown in the above virtual appearance usage prediction method, which will not be repeated here.

[0421] As an optional solution, the second acquisition module includes at least one of the following:

[0422] The sixth submodule is used to acquire features representing the time period in which the opening time falls;

[0423] The seventh acquisition submodule is used to acquire features for representing the preset time points of open time coverage.

[0424] For specific implementation examples, please refer to the examples shown in the above virtual appearance usage prediction method, which will not be repeated here.

[0425] As an optional solution, the second acquisition unit 1304 includes:

[0426] The third acquisition module is used to acquire the first usage feature corresponding to the virtual character, wherein the first usage feature is used to represent the first usage data of the virtual character's appearance resources before the development and use rights of the first virtual appearance, and the second feature includes the first usage feature.

[0427] The fourth acquisition module is used to acquire the second usage feature corresponding to the virtual character, wherein the second usage feature is used to represent the second usage data of the virtual character in a recent time period, and the second feature includes the second usage feature.

[0428] For specific implementation examples, please refer to the examples shown in the above virtual appearance usage prediction method, which will not be repeated here.

[0429] As an optional solution, the third acquisition module includes at least one of the following:

[0430] The eighth acquisition submodule is used to acquire features representing the quantity of historical appearance resources of a virtual character before the first virtual appearance usage rights were granted;

[0431] The ninth submodule is used to acquire features representing the usage of historical appearance resources under different appearance qualities.

[0432] The tenth acquisition submodule is used to acquire features that represent the usage of historical appearance resources within a specific time period.

[0433] For specific implementation examples, please refer to the examples shown in the above virtual appearance usage prediction method, which will not be repeated here.

[0434] As an optional solution, the fourth acquisition module includes at least one of the following:

[0435] The eleventh submodule is used to obtain features representing the virtual character's win rate in recent time periods.

[0436] The twelfth acquisition submodule is used to acquire features representing the usage of virtual characters in a recent time period;

[0437] The thirteenth submodule is used to obtain features representing the number of users of a virtual character in a recent time period.

[0438] For specific implementation examples, please refer to the examples shown in the above virtual appearance usage prediction method, which will not be repeated here.

[0439] As an optional solution, the third acquisition unit 1306 includes:

[0440] The fifth acquisition module is used to acquire the activity characteristics corresponding to the virtual game, wherein the activity characteristics are used to reflect the activity level of the virtual game within a preset time period;

[0441] The sixth acquisition module is used to acquire the first consumption feature corresponding to the virtual game. The first consumption feature is used to represent the situation in which all game users of the virtual game exchange game resources for the right to use by consuming elements. The third feature includes the first consumption feature.

[0442] The seventh acquisition module is used to acquire the second consumption feature corresponding to the virtual game. The second consumption feature is used to represent the situation where a user exchanges elements for the right to use game resources. The third feature includes the second consumption feature.

[0443] For specific implementation examples, please refer to the examples shown in the above virtual appearance usage prediction method, which will not be repeated here.

[0444] As an optional solution, the fifth acquisition module includes at least one of the following:

[0445] The fourteenth submodule is used to obtain features representing the number of daily active users of the virtual game;

[0446] The fifteenth submodule is used to obtain features representing the number of matches in each mode of the virtual game.

[0447] For specific implementation examples, please refer to the examples shown in the above virtual appearance usage prediction method, which will not be repeated here.

[0448] As an optional solution, the sixth acquisition module includes at least one of the following:

[0449] The sixteenth submodule is used to obtain the characteristic representing the proportion of users in the total number of game users who exchange elements for the right to use game resources in the virtual game.

[0450] The seventeenth submodule is used to obtain the feature representing the number of users who exchanged elements in each interval for the right to use game resources in the virtual game.

[0451] The eighteenth submodule is used to obtain features representing the average consumption of all users for consumable elements.

[0452] For specific implementation examples, please refer to the examples shown in the above virtual appearance usage prediction method, which will not be repeated here.

[0453] As an optional solution, the seventh acquisition module includes at least one of the following:

[0454] The nineteenth submodule is used to obtain the feature representing the number of users who exchanged the right to use game resources by consuming elements in each interval.

[0455] The twentieth submodule is used to obtain features representing the average consumption of consumable elements by users.

[0456] For specific implementation examples, please refer to the examples shown in the above virtual appearance usage prediction method, which will not be repeated here.

[0457] As an optional solution, prediction unit 1308 includes:

[0458] The fusion module is used to fuse the first feature, the second feature, and the third feature to obtain multi-dimensional features;

[0459] The first prediction module is used to input multi-dimensional features into the data prediction model and obtain the first prediction result output by the data prediction model. The data prediction model is trained using multiple sample data and is a gradient regression model used to predict the usage data of appearance resources in the future time period.

[0460] For specific implementation examples, please refer to the examples shown in the above virtual appearance usage prediction method, which will not be repeated here.

[0461] As an optional solution, the device also includes:

[0462] The training module is used to train the initial data prediction model multiple times before obtaining the first prediction result output by the data prediction model after inputting multi-dimensional features, until a well-trained data prediction model is obtained. The i-th training round is any one of the multiple training rounds, and the i-th training round is as follows:

[0463] The eighth acquisition module is used to acquire the sample data of the i-th round of training and the data prediction model of the i-th round of training;

[0464] The second prediction module is used to predict the sample data trained in the i-th round using the data prediction model trained in the i-th round, and obtain the first prediction value.

[0465] The fitting module is used to fit an initial decision tree with the residual between the true value corresponding to the sample data in the i-th round of training and the first predicted value as the objective.

[0466] The splitting module is used to maximize the information gain by splitting nodes of the initial decision tree until the stopping condition is met, thus obtaining the decision tree for the i-th round of training.

[0467] The update module is used to update the first predicted value using the decision tree trained in the i-th round, and obtain the second predicted value.

[0468] The first determining module is used to determine the data prediction model trained in the i-th round as the trained data prediction model when the second predicted value meets the convergence condition.

[0469] For specific implementation examples, please refer to the examples shown in the above virtual appearance usage prediction method, which will not be repeated here.

[0470] As an optional solution, the device also includes:

[0471] The determining unit is configured to determine at least two sets of appearance resources that are associated with the first virtual appearance from the appearance resources of the virtual game before correcting the first prediction result using at least one second virtual appearance to obtain the second prediction result, wherein different sets of appearance resources correspond to different types of association, and the sets of appearance resources include the second virtual appearance.

[0472] The recall unit is used to process each appearance resource set in at least two appearance resource sets according to the resource quantity in the appearance resource set before correcting the first prediction result using at least one second virtual appearance to obtain the second prediction result, thereby obtaining at least two recalled resource sets.

[0473] The allocation unit is used to allocate corresponding fusion weights to each recall resource set in at least two recall resource sets before correcting the first prediction result using at least one second virtual appearance to obtain the second prediction result.

[0474] The fusion unit is used to perform weighted fusion of at least two recall resource sets according to fusion weights to obtain first corrected data before correcting the first prediction result using at least one second virtual appearance to obtain a second prediction result.

[0475] The correction unit 1310 includes: a first correction module, used to correct the first prediction result using first correction data to obtain a second prediction result.

[0476] For specific implementation examples, please refer to the examples shown in the above virtual appearance usage prediction method, which will not be repeated here.

[0477] As an optional solution, the recall unit includes:

[0478] The removal module is used to remove the highest first proportion and the lowest second proportion from the first resource set with a resource quantity greater than or equal to a preset threshold, and then take the average value of the remaining samples as the representative value of the recall resource set to obtain the first recall resource set.

[0479] The second determination module is used to determine the mean of all samples in the second resource set for a second resource set whose resource quantity is less than a preset threshold and greater than 0, and adjust the mean according to the resource quantity to obtain the second recall resource set.

[0480] For specific implementation examples, please refer to the examples shown in the above virtual appearance usage prediction method, which will not be repeated here.

[0481] As an optional scheme, the allocation unit includes at least one of the following:

[0482] The first allocation module is used to allocate a first weight to each appearance resource according to the importance of the appearance resources in each recall resource set. The importance and the first weight are positively correlated, and the fusion weight includes the first weight.

[0483] The second allocation module is used to allocate a corresponding second weight to each appearance resource based on the number of appearance resources in each recall resource set. The number of resources and the second weight are positively correlated, and the fusion weight includes the second weight.

[0484] The third allocation module is used to assign a corresponding third weight to each appearance resource based on the dispersion of appearance resources in each recall resource set. The dispersion is positively correlated with the third weight, and the fusion weight includes the third weight.

[0485] For specific implementation examples, please refer to the examples shown in the above virtual appearance usage prediction method, which will not be repeated here.

[0486] As an optional solution, the device also includes:

[0487] The fourth acquisition unit is used to, after determining at least two sets of appearance resources that are related to the first virtual appearance from the appearance resources of the virtual game, process each of the at least two sets of appearance resources using the corresponding recall method according to the amount of resources in the appearance resource sets, and obtain 0 recalled resource sets, and then obtain the second correction data based on the usage data of the appearance resources of the virtual game that are of the same appearance quality as the first virtual appearance.

[0488] The correction unit 1310 includes: a second correction module, used to correct the first prediction result using second correction data to obtain a second prediction result.

[0489] For specific implementation examples, please refer to the examples shown in the above virtual appearance usage prediction method, which will not be repeated here.

[0490] According to another aspect of the embodiments of this application, an electronic device for implementing the above-described virtual appearance usage prediction method is also provided. This electronic device may, but is not limited to, […]. Figure 1 The user equipment 102 or server 112 shown in the figure, in this embodiment, is taken as an example of an electronic device, namely user equipment 102. Further, as shown in the figure... Figure 14 As shown, the electronic device includes a memory 1402 and a processor 1404. The memory 1402 stores a computer program, and the processor 1404 is configured to execute the steps of any of the above method embodiments via the computer program.

[0491] In an optional embodiment, the aforementioned electronic device may be located in at least one of a plurality of network devices in a computer network.

[0492] In an optional embodiment, the processor described above may be configured to perform the following steps via a computer program:

[0493] S1, obtain the first feature corresponding to the first virtual appearance, wherein the first virtual appearance is used to change the appearance of the virtual character in the virtual game, and the first feature is used to represent the attribute data and time information associated with the first virtual appearance;

[0494] S2, obtain the second feature corresponding to the virtual character, wherein the second feature is used to represent the appearance resources of the virtual character and the usage data of the virtual character;

[0495] S3, obtain the third feature corresponding to the virtual game, wherein the third feature is used to represent the activity data of the virtual game and the usage information of game resources by game users in the virtual game. Game users include users who use virtual characters, and game resources include the appearance resources of virtual characters.

[0496] S4, combining the first feature, the second feature, and the third feature, predict the usage data of the first virtual appearance in the future time period to obtain the first prediction result;

[0497] S5, based on the first feature, the second feature, and the third feature, if at least one second virtual appearance is determined from the appearance resources of the virtual game that satisfies the relevant conditions with the first virtual appearance, the first prediction result is corrected using the at least one second virtual appearance to obtain the second prediction result.

[0498] Alternatively, as those skilled in the art will understand, Figure 14 The structure shown is for illustrative purposes only. Figure 14 This does not limit the structure of the aforementioned electronic devices. For example, the electronic device may also include components that are more... Figure 14 The more or fewer components shown (such as network interfaces, etc.), or having the same Figure 14 The different configurations shown.

[0499] The memory 1402 can be used to store software programs and modules, such as the program instructions / modules corresponding to the virtual appearance usage prediction method and apparatus in this embodiment. The processor 1404 executes various functional applications and data processing by running the software programs and modules stored in the memory 1402, thereby realizing the aforementioned virtual appearance usage prediction method. The memory 1402 may include high-speed random access memory and may also include non-volatile memory, such as one or more magnetic storage devices, flash memory, or other non-volatile solid-state memory. In some instances, the memory 1402 may further include memory remotely located relative to the processor 1404, and these remote memories can be connected to electronic devices via a network. Examples of such networks include, but are not limited to, the Internet, corporate intranets, local area networks, mobile communication networks, and combinations thereof. Specifically, the memory 1402 may be used, but is not limited to, to store information such as a first feature, a second feature, a third feature, a first prediction result, and a second prediction result. As an example, such as... Figure 14 As shown, the memory 1402 may include, but is not limited to, the first acquisition unit 1302, the second acquisition unit 1304, the third acquisition unit 1306, the prediction unit 1308, and the correction unit 1310 of the virtual appearance usage prediction device. Furthermore, it may include, but is not limited to, other module units of the virtual appearance usage prediction device, which will not be elaborated upon in this example.

[0500] Optionally, the transmission device 1406 described above is used to receive or send data via a network. Specific examples of the network described above may include wired networks and wireless networks. In one example, the transmission device 1406 includes a Network Interface Controller (NIC), which can be connected to other network devices and a router via a network cable to communicate with the Internet or a local area network. In another example, the transmission device 1406 is a Radio Frequency (RF) module, used for wireless communication with the Internet.

[0501] In addition, the above-mentioned electronic device also includes: a display 1408 for displaying information such as the first feature, the second feature, the third feature, the first prediction result, and the second prediction result; and a connection bus 1410 for connecting the various module components in the above-mentioned electronic device.

[0502] In other embodiments, the aforementioned user equipment or server can be a node in a distributed system, wherein the distributed system can be a blockchain system, which is a distributed system formed by connecting multiple nodes through network communication. The nodes can form a peer-to-peer network, and any form of computing device, such as a server, user equipment, or other electronic device, can become a node in the blockchain system by joining this peer-to-peer network.

[0503] According to one aspect of this application, a computer program product is provided, comprising a computer program / instructions containing program code for performing the methods shown in the flowchart. In such embodiments, the computer program can be downloaded and installed from a network via a communication component, and / or installed from a removable medium. When the computer program is executed by a central processing unit, it performs various functions provided in embodiments of this application.

[0504] The sequence numbers of the embodiments in this application are for descriptive purposes only and do not represent the superiority or inferiority of the embodiments.

[0505] It should be noted that the computer system of the electronic device is merely an example and should not impose any limitations on the functionality and scope of use of the embodiments of this application.

[0506] A computer system includes a Central Processing Unit (CPU), which performs various appropriate actions and processes based on programs stored in Read-Only Memory (ROM) or loaded from RAM. ROM also stores various programs and data required for system operation. The CPU, ROM, and RAM are interconnected via a bus. Input / output interfaces (I / O interfaces) are also connected to the bus.

[0507] The following components are connected to the input / output interface: input sections including keyboards, mice, etc.; output sections including cathode ray tubes (CRTs), liquid crystal displays (LCDs), and speakers; storage sections including hard drives; and communication sections including network interface cards such as LAN cards and modems. The communication section performs communication processing via a network such as the Internet. Drives are also connected to the input / output interface as needed. Removable media, such as disks, optical discs, magneto-optical discs, semiconductor memories, etc., are installed on the drive as needed so that computer programs read from them can be installed into the storage section as required.

[0508] Specifically, according to embodiments of this application, the processes described in the various method flowcharts can be implemented as computer software programs. For example, embodiments of this application include a computer program product comprising a computer program carried on a computer-readable medium, the computer program containing program code for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via a communication component, and / or installed from a removable medium. When the computer program is executed by a central processing unit, it performs various functions defined in the system of this application.

[0509] According to one aspect of this application, a computer-readable storage medium is provided, wherein a processor of a computer device reads computer instructions from the computer-readable storage medium, and executes the computer instructions, causing the computer device to perform the methods provided in the various alternative implementations described above.

[0510] In an optional embodiment, the computer-readable storage medium described above may be configured to store a computer program for performing the following steps:

[0511] S1, obtain the first feature corresponding to the first virtual appearance, wherein the first virtual appearance is used to change the appearance of the virtual character in the virtual game, and the first feature is used to represent the attribute data and time information associated with the first virtual appearance;

[0512] S2, obtain the second feature corresponding to the virtual character, wherein the second feature is used to represent the appearance resources of the virtual character and the usage data of the virtual character;

[0513] S3, obtain the third feature corresponding to the virtual game, wherein the third feature is used to represent the activity data of the virtual game and the usage information of game resources by game users in the virtual game. Game users include users who use virtual characters, and game resources include the appearance resources of virtual characters.

[0514] S4, combining the first feature, the second feature, and the third feature, predict the usage data of the first virtual appearance in the future time period to obtain the first prediction result;

[0515] S5, based on the first feature, the second feature, and the third feature, if at least one second virtual appearance is determined from the appearance resources of the virtual game that satisfies the relevant conditions with the first virtual appearance, the first prediction result is corrected using the at least one second virtual appearance to obtain the second prediction result.

[0516] Optionally, in embodiments of this application, the terms "module" or "unit" refer to a computer program or part of a computer program that has a predetermined function and works with other related parts to achieve a predetermined goal, and can be implemented wholly or partially using software, hardware (such as processing circuitry or memory), or a combination thereof. Similarly, a processor (or multiple processors or memory) can be used to implement one or more modules or units. Furthermore, each module or unit can be part of an overall module or unit that includes the functionality of that module or unit.

[0517] In optional embodiments, those skilled in the art will understand that all or part of the steps in the various methods of the above embodiments can be implemented by a program instructing related hardware of an electronic device. The program can be stored in a computer-readable storage medium, which may include: a flash drive, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk, etc.

[0518] The sequence numbers of the embodiments in this application are for descriptive purposes only and do not represent the superiority or inferiority of the embodiments.

[0519] If the integrated units in the above embodiments are implemented as software functional units and sold or used as independent products, they can be stored in the aforementioned computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause one or more computer devices (which may be personal computers, servers, or network devices, etc.) to execute all or part of the steps of the methods of the various embodiments of this application.

[0520] In the above embodiments of this application, the descriptions of each embodiment have different focuses. For parts not described in detail in a certain embodiment, please refer to the relevant descriptions of other embodiments.

[0521] In the several embodiments provided in this application, it should be understood that the disclosed user equipment can be implemented in other ways. The device embodiments described above are merely illustrative; for example, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection of units or modules may be electrical or other forms.

[0522] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0523] Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0524] The above description is only a preferred embodiment of this application. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the principle of this application, and these improvements and modifications should also be considered within the scope of protection of this application.

Claims

1. A method for predicting the use of virtual appearances, characterized in that, include: Obtain the first feature corresponding to the first virtual appearance, wherein the first virtual appearance is used to change the appearance of the virtual character in the virtual game, and the first feature is used to represent the attribute data and time information associated with the first virtual appearance; Obtain the second feature corresponding to the virtual character, wherein the second feature is used to represent the appearance resources of the virtual character and the usage data of the virtual character; Obtain the third feature corresponding to the virtual game, wherein the third feature is used to represent the activity data of the virtual game and the usage information of game resources by game users in the virtual game, wherein the game users include users who use the virtual character, and the game resources include the appearance resources of the virtual character; Combining the first feature, the second feature, and the third feature, the usage data of the first virtual appearance in a future time period is predicted to obtain a first prediction result, including: fusing the first feature, the second feature, and the third feature to obtain multi-dimensional features; inputting the multi-dimensional features into a data prediction model to obtain the first prediction result output by the data prediction model, wherein the data prediction model is a gradient regression model trained using multiple sample data to predict the usage data of the appearance resource in a future time period; Based on the first feature, the second feature, and the third feature, at least one second virtual appearance is determined from the appearance resources of the virtual game that satisfies the relevant conditions for similarity with the first virtual appearance. The first prediction result is then corrected using the usage data of the at least one second virtual appearance to obtain a second prediction result.

2. The method according to claim 1, characterized in that, The step of obtaining the first feature corresponding to the first virtual appearance includes: Obtain the attribute features corresponding to the first virtual appearance, wherein the attribute features are used to represent the first attribute data of the first virtual appearance and the second attribute data of the virtual character, and the first feature includes the attribute features; Obtain the time feature corresponding to the first virtual appearance, wherein the time feature is used to represent the opening time of the right to use the first virtual appearance, and the first feature includes the time feature.

3. The method according to claim 2, characterized in that, The step of obtaining the attribute features corresponding to the first virtual appearance includes at least one of the following: Obtain the features used to represent the appearance series to which the first virtual appearance belongs; Obtain features that represent the appearance quality of the first virtual appearance; Obtain the features used to represent the appearance type to which the first virtual appearance belongs; Obtain the feature representing the amount of elements required to obtain the right to use the first virtual appearance.

4. The method according to claim 2, characterized in that, The step of obtaining the time feature corresponding to the first virtual appearance includes at least one of the following: Obtain features representing the time period in which the opening time falls; Obtain features that represent the preset time points covered by the open time.

5. The method according to claim 1, characterized in that, The step of obtaining the second feature corresponding to the virtual character includes: Obtain the first usage feature corresponding to the virtual character, wherein the first usage feature is used to represent the first usage data of the appearance resource of the virtual character before the first virtual appearance development and usage right, and the second feature includes the first usage feature; Obtain the second usage feature corresponding to the virtual character, wherein the second usage feature is used to represent the second usage data of the virtual character in a recent time period, and the second feature includes the second usage feature.

6. The method according to claim 5, characterized in that, The step of obtaining the first usage feature corresponding to the virtual character includes at least one of the following: Obtain features representing the quantity of historical appearance resources of the virtual character before the right to use the first virtual appearance was granted; Obtain features representing the usage of the historical appearance resources under different appearance qualities; Obtain features that represent the usage of the historical appearance resource within a specific time period.

7. The method according to claim 5, characterized in that, The step of obtaining the second usage feature corresponding to the virtual character includes at least one of the following: Obtain features representing the virtual character's win rate in matches during the recent time period; Obtain features representing the usage of the virtual character during the recent time period; Obtain features representing the number of users of the virtual character within the recent time period.

8. The method according to claim 1, characterized in that, The step of obtaining the third feature corresponding to the virtual game includes: Obtain the activity characteristics corresponding to the virtual game, wherein the activity characteristics are used to reflect the activity level of the virtual game within a preset time period; Obtain the first consumption feature corresponding to the virtual game, wherein the first consumption feature is used to represent the situation in which all game users of the virtual game exchange the right to use the game resources by consuming elements, and the third feature includes the first consumption feature; Obtain the second consumption feature corresponding to the virtual game, wherein the second consumption feature is used to indicate the situation where the user exchanges the right to use the game resources through the consumption element, and the third feature includes the second consumption feature.

9. The method according to claim 8, characterized in that, The acquisition of the activity characteristics corresponding to the virtual game includes at least one of the following: Obtain features representing the number of daily active users of the virtual game; Obtain features used to represent the number of matches in each mode of the virtual game.

10. The method according to claim 8, characterized in that, The acquisition of the first consumption feature corresponding to the virtual game includes at least one of the following: Obtain a feature representing the percentage of all game users who exchange the virtual game for the right to use game resources by consuming elements. Obtain a feature representing the number of users who exchange the right to use game resources for the virtual game through the consumed elements in each interval; Obtain features representing the average consumption of the consumed element by all users.

11. The method according to claim 8, characterized in that, The acquisition of the second consumption feature corresponding to the virtual game includes at least one of the following: Obtain a feature representing the number of users who exchanged the right to use the game resources for the consumed elements in each interval; Obtain features representing the average consumption of the consumable element by the user.

12. The method according to claim 1, characterized in that, Before inputting the multi-dimensional features into the data prediction model and obtaining the first prediction result output by the data prediction model, the method further includes: The initial data prediction model is trained multiple times until a well-trained data prediction model is obtained. The i-th training round is any one of the multiple training rounds, and the i-th training round is as follows: Obtain the sample data from the i-th training round and the data prediction model from the i-th training round; Using the data prediction model trained in the i-th round, a prediction is made on the sample data trained in the i-th round to obtain a first prediction value; Using the residual between the true value corresponding to the sample data in the i-th round of training and the first predicted value as the target, fit an initial decision tree; By splitting nodes of the initial decision tree to maximize information gain until the stopping condition is met, the decision tree of the i-th training round is obtained. Using the decision tree trained in the i-th round, update the first predicted value to obtain the second predicted value; If the second predicted value meets the convergence condition, the data prediction model trained in the i-th round is determined as the trained data prediction model.

13. The method according to any one of claims 1 to 12, characterized in that, Before correcting the first prediction result using the at least one second virtual appearance to obtain the second prediction result, the method further includes: From the appearance resources of the virtual game, at least two appearance resource sets that are associated with the first virtual appearance are determined, wherein different appearance resource sets correspond to different association types, and the appearance resource sets include the second virtual appearance; Based on the amount of resources in the appearance resource set, each appearance resource set in the at least two appearance resource sets is processed using the corresponding recall method to obtain at least two recalled resource sets. Assign corresponding fusion weights to each of the at least two recall resource sets; According to the fusion weight, the at least two recall resource sets are weighted and fused to obtain the first corrected data; The step of correcting the first prediction result using the at least one second virtual appearance to obtain a second prediction result includes: correcting the first prediction result using the first correction data to obtain the second prediction result.

14. The method according to claim 13, characterized in that, The step involves processing each appearance resource set in the at least two appearance resource sets using a corresponding recall method based on the resource quantity in the appearance resource set, to obtain at least two recalled resource sets, including: For a first resource set whose resource quantity is greater than or equal to a preset threshold, after removing the highest first proportion and the lowest second proportion from the first resource set, the average value of the remaining samples is taken as the representative value of the recall resource set to obtain the first recall resource set. For a second resource set whose resource quantity is less than the preset threshold and greater than 0, the mean of all samples in the second resource set is determined, and the mean is adjusted according to the resource quantity to obtain a second recall resource set.

15. The method according to claim 13, characterized in that, Assigning corresponding fusion weights to each of the at least two recall resource sets includes at least one of the following: Based on the importance of each appearance resource in the recall resource set, a first weight is assigned to each appearance resource, wherein the importance is positively correlated with the first weight, and the fusion weight includes the first weight; Based on the number of appearance resources in each of the recalled resource sets, a corresponding second weight is assigned to each appearance resource, wherein the number of resources is positively correlated with the second weight, and the fusion weight includes the second weight; Based on the dispersion of appearance resources in each recall resource set, a corresponding third weight is assigned to each appearance resource, wherein the dispersion is positively correlated with the third weight, and the fusion weight includes the third weight.

16. The method according to claim 13, characterized in that, After determining at least two sets of appearance resources that are associated with the first virtual appearance from the appearance resources of the virtual game, the method further includes: if, based on the amount of resources in the appearance resource sets, each appearance resource set in the at least two appearance resource sets is processed using the corresponding recall method to obtain 0 recalled resource sets, then, based on the usage data of the appearance resources of the virtual game that belongs to the same appearance quality as the first virtual appearance, second correction data is obtained; The step of correcting the first prediction result using the at least one second virtual appearance to obtain a second prediction result includes: correcting the first prediction result using the second correction data to obtain the second prediction result.

17. A device for predicting the use of a virtual appearance, characterized in that, include: The first acquisition unit is used to acquire the first feature corresponding to the first virtual appearance, wherein the first virtual appearance is used to change the appearance of the virtual character in the virtual game, and the first feature is used to represent the attribute data and time information associated with the first virtual appearance. The second acquisition unit is used to acquire a second feature corresponding to the virtual character, wherein the second feature is used to represent the appearance resources of the virtual character and the usage data of the virtual character; The third acquisition unit is used to acquire the third feature corresponding to the virtual game, wherein the third feature is used to represent the activity data of the virtual game and the usage information of game resources by game users in the virtual game, the game users include users who use the virtual character, and the game resources include the appearance resources of the virtual character; A prediction unit is configured to combine the first feature, the second feature, and the third feature to predict the usage data of the first virtual appearance in a future time period, and obtain a first prediction result. The unit includes: fusing the first feature, the second feature, and the third feature to obtain multi-dimensional features; inputting the multi-dimensional features into a data prediction model to obtain the first prediction result output by the data prediction model, wherein the data prediction model is a gradient regression model trained using multiple sample data to predict the usage data of the appearance resource in a future time period. The correction unit is configured to, based on the first feature, the second feature, and the third feature, determine at least one second virtual appearance from the appearance resources of the virtual game that satisfies the relevant conditions for similarity with the first virtual appearance, and then use the usage data of the at least one second virtual appearance to correct the first prediction result to obtain a second prediction result.

18. A computer-readable storage medium, characterized in that, The computer-readable storage medium includes a stored program, wherein the program is executed by an electronic device to perform the method according to any one of claims 1 to 16.

19. A computer program product comprising a computer program / instructions, characterized in that, When the computer program / instructions are executed by the processor, they implement the steps of the method described in any one of claims 1 to 16.

20. An electronic device comprising a memory and a processor, characterized in that, The memory stores a computer program, and the processor is configured to execute the method described in any one of claims 1 to 16 through the computer program.

Images