Recommendation method and device for user generated works, storage medium and program product

Abstract

The invention discloses a recommendation method and device for a user-generated work, a storage medium and a program product, and the method comprises the steps: obtaining work features of the user-generated work, the work features comprising content features and intention features; and determining a target work in the user-generated works according to the work characteristics of the user-generated works and a player portrait of a game player, and pushing the target work to the game player. According to the technical scheme, the recommendation mode which can deeply insight into UGC content values and dynamically adapt to the interest requirements of players is constructed, and accurate recommendation of the works generated by the users is achieved.

Description

TECHNICAL FIELD

[0001] The application belongs to the field of data processing of games, and particularly relates to a user-generated work recommendation method and device, a storage medium and a program product. BACKGROUND

[0002] In the field of sandbox games, user-generated works (UGC) are the core support of platform ecology, and a recommendation system, as a key carrier connecting massive UGC content and game player needs, aims to accurately match content and needs. The current mainstream UGC content recommendation scheme has significant shortcomings, and has not formed an effective mechanism to adapt to the characteristics of UGC and the needs of game players, making it difficult to achieve accurate recommendations.

[0003] The prior art understands UGC content at the surface level, relying on limited information such as user interaction behavior, simple keywords or artificial tags, and fails to mine UGC content, resulting in problems such as burying high-quality content and excessive exposure of homogenized content. At the same time, the prior art lacks dynamic adaptation capability in modeling game player interests, assuming that interests remain constant. This leads to a lack of effective adaptation strategies for new content and new game players, further restricting the accuracy of recommendations.

[0004] Therefore, how to build a recommendation method that can deeply understand the value of UGC content and dynamically adapt to the interests and needs of game players to achieve accurate recommendations of user-generated works is a technical problem that needs to be solved. SUMMARY

[0005] The purpose of the present application is to build a recommendation method that can deeply understand the value of UGC content and dynamically adapt to the interests and needs of game players to achieve accurate recommendations of user-generated works.

[0006] Other characteristics and advantages of the present application will become apparent from the following detailed description, or will be learned by practice of the present application.

[0007] According to one aspect of an embodiment of the present application, a user-generated work recommendation method is provided, the method comprising: obtaining work features of a user-generated work, the work features including content features and intent features; determining a target work in the user-generated work according to the work features of the user-generated work and a player portrait of a game player, and pushing the target work to the game player.

[0008] According to one aspect of an embodiment of the present application, obtaining work features of a user-generated work comprises: extracting features of images in the user-generated work to obtain visual features of the user-generated work; Feature extraction is performed on the text in the user-generated work to obtain the text features of the user-generated work. The content features include the visual features and the text features.

[0009] According to one aspect of the embodiments of this application, obtaining the work features of a user-generated work further includes: Obtain the sequence of operation instructions used when creating the user-generated work; Based on the sequence of operation instructions, the intent characteristics of the user-generated work are determined.

[0010] According to one aspect of the embodiments of this application, the method further includes: Based on the player's historical game behavior within a set time period, a long-term profile of the player is constructed. Based on the player's current game behavior after this login, construct a short-term profile of the player; Based on the long-term and short-term profiles of the game players, the player profile of the game player is determined.

[0011] According to one aspect of the embodiments of this application, determining the player profile of the game player based on the long-term profile and short-term profile of the game player includes: Based on the current game mode of the player, determine the first weight corresponding to the long-term profile and the second weight corresponding to the short-term profile, and the sum of the first weight and the second weight is 1; The player profile is obtained by combining the long-term profile and the short-term profile based on the first weight and the second weight.

[0012] According to one aspect of the embodiments of this application, determining a target work among the user-generated works based on the work characteristics of the user-generated works and the player profile of the game player includes: The characteristics of the works and the characteristics of the player profiles are normalized to obtain normalized features; Based on the interaction between normalized features, a matching score is determined between the user-generated work and the player profile; The target work is determined from the user-generated works based on the matching score between each user-generated work and the player profile.

[0013] According to one aspect of the embodiments of this application, determining a target work among the user-generated works based on the matching score between each of the user-generated works and the player profile includes: The first number of user-generated works, ranked from highest to lowest matching score, are selected as the target works. After determining the first number of target works, the similarity between the user-generated works and the target works is determined sequentially according to the matching scores from largest to smallest to see if the similarity is greater than the similarity threshold. If the similarity between the user-generated work and the target work is greater than the similarity threshold, the user-generated work is excluded; if the similarity between the user-generated work and the target work is less than or equal to the similarity threshold, the user-generated work is taken as the target work; until the number of target works reaches the second number.

[0014] According to one aspect of the embodiments of this application, a recommendation device for user-generated works is provided, including a memory, a processor, and a readable program stored in the memory, wherein the processor executes the readable program to implement the method described in any of the above.

[0015] According to one aspect of the embodiments of this application, a readable storage medium is provided, on which a readable program / instruction is stored, which, when executed by a processor, implements the method described in any one of the above-described embodiments.

[0016] According to one aspect of the embodiments of this application, a program product is provided, including a readable program / instruction that, when executed by a processor, implements the method described in any one of the above.

[0017] This application first obtains the characteristics of user-generated works, including content characteristics and intent characteristics. Then, based on the characteristics of the user-generated works and the player profile of the game player, target works are identified from the user-generated works and pushed to the game player. In this embodiment, by simultaneously obtaining the content characteristics and intent characteristics of user-generated works, it breaks through the limitations of traditional recommendation methods that rely solely on single surface information (such as titles and simple tags), achieving a deep understanding of works from their external presentation to their intrinsic creative purpose. This effectively solves the problem of high-quality works being buried and clickbait content being overexposed due to superficial content understanding. Furthermore, by matching the characteristics of the works with the player profile of the game player, the recommendation decision is based on the alignment between the core value of the work and the real needs of the game player, avoiding the shortcomings of traditional static recommendations that cannot adapt to the interests of game players, significantly improving recommendation accuracy, and allowing game players to quickly obtain works that match their interests. The platform enhances the gaming experience by providing content tailored to individual player preferences. Furthermore, the platform allows for the identification of a work's intent and content characteristics without relying on historical interaction data. This enables newly released user-generated content, even lacking past exposure data, to gain recommendation opportunities by matching its own characteristics with player profiles. This effectively solves the cold start problem for new user-generated content, providing fair exposure opportunities for high-quality new works and creators with strong creative abilities but weak marketing skills. This, in turn, promotes the healthy development of the platform's UGC ecosystem and increases creators' enthusiasm. Simultaneously, the precise matching logic reduces ineffective push notifications of homogeneous content, lowers the cost for gamers to filter content, and further enhances player stickiness on the platform.

[0018] Other features and advantages of this application will become apparent from the following detailed description, or may be learned in part from practice of this application.

[0019] It should be understood that the above general description and the following detailed description are exemplary and explanatory only, and do not limit this application. Attached Figure Description

[0020] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this application and, together with the description, serve to explain the principles of this application. It is obvious that the drawings described below are merely some embodiments of this application, and those skilled in the art can obtain other drawings based on these drawings without any inventive effort.

[0021] Figure 1 A schematic diagram of a method for recommending user-generated works according to an embodiment of this application is shown.

[0022] Figure 2 A flowchart illustrating the acquisition of work features of a user-generated work according to an embodiment of this application is shown.

[0023] Figure 3 A flowchart illustrating the acquisition of work features of a user-generated work according to another embodiment of this application is shown.

[0024] Figure 4 A flowchart illustrating the determination of a player profile according to one embodiment of this application is shown.

[0025] Figure 5 A flowchart illustrating a process for determining a player profile based on a long-term and short-term profile of a game player, according to one embodiment of this application, is shown.

[0026] Figure 6 A flowchart illustrating a process for determining a target work from user-generated works based on the work characteristics of user-generated works and the player profile of game players, according to one embodiment of this application, is shown.

[0027] Figure 7 A flowchart illustrating a method for determining a target work among user-generated works based on a matching score between each user-generated work and a player profile, according to one embodiment of this application, is shown.

[0028] Figure 8 A block diagram of a computer device for implementing a method for recommending user-generated works, according to an embodiment of this application, is shown. Detailed Implementation

[0029] Exemplary embodiments will now be described more fully with reference to the accompanying drawings. However, these exemplary embodiments can be implemented in many forms and should not be construed as limited to the examples set forth herein; rather, these embodiments are provided to make this application more comprehensive and complete, and to fully convey the concept of the exemplary embodiments to those skilled in the art.

[0030] Furthermore, the described features, structures, or characteristics can be combined in any suitable manner in one or more embodiments. Numerous specific details are provided in the following description to give a thorough understanding of embodiments of this application. However, those skilled in the art will recognize that the technical solutions of this application can be practiced without one or more of the specific details, or other methods, components, apparatuses, steps, etc., can be employed. In other instances, well-known methods, apparatuses, implementations, or operations are not shown or described in detail to avoid obscuring various aspects of this application.

[0031] The block diagrams shown in the accompanying drawings are merely functional entities and do not necessarily correspond to physically independent entities. That is, these functional entities can be implemented in software, in one or more hardware modules or integrated circuits, or in different network and / or processor devices and / or microcontroller devices.

[0032] The flowcharts shown in the accompanying drawings are merely illustrative and do not necessarily include all content and operations / steps, nor do they necessarily have to be performed in the described order. For example, some operations / steps can be broken down, while others can be combined or partially combined; therefore, the actual execution order may change depending on the specific circumstances.

[0033] In the 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.

[0034] Please see Figure 1 , Figure 1 A schematic diagram of a method for recommending user-generated works according to an embodiment of this application is shown. This application embodiment provides the execution steps of a method for recommending user-generated works, including: Step S110: Obtain the work features of the user-generated work, which include content features and intent features; Step S120: Based on the characteristics of user-generated works and the player profiles of game players, identify target works among user-generated works and push them to game players.

[0035] The two steps above will be described in detail below.

[0036] In step S110, this application is used to accurately recommend user-generated works to corresponding game players. User-generated works refer to game-related content created independently by game players (creators), such as UGC content like maps, levels, and scene designs in sandbox games. Work features are a set of information describing the core attributes of user-generated works and are a key basis for matching game player needs; they can include two categories: content features and intent features.

[0037] Content features directly reflect the characteristics of the work's external presentation and surface information, such as the visual information of the images and the textual descriptions. Intent features reflect the core characteristics of the creator's creative purpose and reflect the original design intention of the work (such as the core gameplay orientation of puzzle-solving, architecture, parkour, etc.).

[0038] Comprehensive acquisition of the core attributes of user-generated works. On the one hand, collect the content characteristics of user-generated works, that is, relevant information about the external presentation of user-generated works (such as images, text, etc.), allowing the system to understand what the user-generated works are like; on the other hand, collect the intention characteristics of user-generated works, that is, the core purpose of the creator in creating the user-generated works (such as designing puzzle levels, building architectural scenes, etc.), determining what the user-generated works are used for. By collecting these two types of characteristics, a comprehensive understanding of the works from the surface to the depth can be achieved, avoiding misunderstandings caused by relying on only a single piece of information.

[0039] In step S120, game players can include two types of characters: first, players who are not creating content but simply playing the game (such as exploring and playing on a map); and second, game players who are creating content (such as creating maps in the game or creating map guides to publish on the platform). That is, game players can both browse user-generated content and create user-generated content.

[0040] First, the characteristics of the acquired works are correlated with the player profiles of game players to determine the degree of fit between user-generated works and game players' interests and preferences. Second, based on the degree of fit, the most suitable target works for the game player are selected, and works that do not match the game player's needs are excluded. Finally, the target works are pushed to game players through the game platform to complete the recommendation loop.

[0041] In this embodiment, both the content features and intent features of user-generated works are simultaneously acquired, overcoming the limitations of traditional recommendation methods that rely solely on superficial information (such as titles and simple tags). This achieves a deep understanding of the work, from its external presentation to its intrinsic creative purpose, effectively solving the problems of high-quality works being overlooked and clickbait content being overexposed due to superficial content understanding. Furthermore, by matching the work's features with the player profiles of gamers, recommendation decisions are based on the alignment between the core value of the work and the real needs of gamers. This avoids the shortcomings of traditional static recommendations that fail to match player interests, significantly improving recommendation accuracy and enabling gamers to quickly access content that suits their needs. The platform enhances the gaming experience by providing preferred content. Furthermore, the intent and content characteristics of works can be obtained without relying on historical interaction data. This allows newly released user-generated content, even lacking past exposure data, to gain recommendation opportunities by matching its own characteristics with game player profiles. This effectively solves the cold start problem for new user-generated content, giving high-quality new works and creators with strong creative abilities but weak marketing skills a fair exposure opportunity. This, in turn, promotes the healthy development of the platform's UGC ecosystem and enhances creators' enthusiasm. At the same time, the precise matching logic reduces ineffective pushes of homogeneous content, lowers the cost for game players to filter content, and further enhances player stickiness on the platform.

[0042] Please seeFigure 2 , Figure 2 A flowchart illustrating the process of obtaining work features of a user-generated work according to an embodiment of this application is shown. This application embodiment provides step S110 for obtaining work features of a user-generated work, including: Step S111a: Extract features from the images in the user-generated work to obtain the visual features of the user-generated work; Step S112a: Extract features from the text in the user-generated work to obtain the text features of the user-generated work. The content features include visual features and text features.

[0043] The two steps described above are described in detail below.

[0044] In step S111a, visual features refer to the features (such as image style, color, structure, etc.) extracted from the image of the user-generated work that characterize the visual attributes of the work.

[0045] Collect visual dimension information of user-generated works. The collection objects are images associated with user-generated works (such as cover images, key scene screenshots, etc.). Through specific technical means, the images are analyzed to extract visual features (such as style, color, structure, etc.) that reflect the core attributes of the images, determine the visual presentation characteristics of user-generated works, and avoid the omission of visual attributes caused by relying solely on text information.

[0046] In step S112a, text features refer to the features (such as theme, keywords, emotional tendencies, etc.) extracted from the text of the user-generated work that represent the textual connotation of the work.

[0047] Collect textual information from user-generated works. The collection targets are the text associated with user-generated works (such as titles, descriptions, comments, etc.), and use technical means to extract textual features (such as themes, keywords, etc.) that reflect the core meaning of the text; at the same time, clarify that visual features and textual features together constitute the content features of the work, achieving comprehensive coverage of the surface information of the work, and providing a comprehensive content basis for subsequent accurate matching.

[0048] This application's embodiments explicitly define content features as consisting of visual and textual features, extracting these two types of features from the images and text of user-generated works respectively. This breaks away from the comprehension biases caused by traditional recommendations relying solely on single information (such as simple tags or titles). Visual features can capture intuitive attributes such as the visual style and color distribution of user-generated works, while textual features can uncover the work's theme, core gameplay, and other intrinsic information. These two types of features complement each other to form comprehensive content features, enabling the system's understanding of user-generated works to move from superficial to multi-dimensional, avoiding misjudgments of clickbait content or the burying of high-quality user-generated works due to their simple expression. At the same time, image and textual features are inherent attributes of user-generated works themselves, and can be extracted without relying on historical interaction data, providing crucial support for the cold start of newly released works. This allows new user-generated works to enter the recommendation process based solely on their own visual and textual features. Furthermore, comprehensive content features provide a solid foundation for subsequent accurate matching with game player profiles, reducing recommendation biases caused by missing content information, significantly improving the scientific nature and accuracy of recommendation decisions, thereby optimizing the experience of game players in obtaining high-quality content, and providing fair exposure opportunities for creators with high creative skills but weak marketing capabilities.

[0049] Please see Figure 3 , Figure 3 A flowchart illustrating the process of obtaining work features of a user-generated work according to another embodiment of this application is shown. This application embodiment provides step S110 for obtaining work features of a user-generated work, including: Step S111b: Obtain the sequence of operation instructions when the creator generates the work; Step S112b: Determine the intent characteristics of the user-generated work based on the sequence of operation instructions.

[0050] The two steps above will be described in detail below.

[0051] In step S111b, the operation instruction sequence refers to a series of continuous operation records executed by the creator of the user-generated work in the game editor during the creation process, which is the core data reflecting the creation process.

[0052] By using technical means to obtain real-time records of every action of the creator (such as placing blocks, setting triggers, connecting circuits, etc.), a continuous and complete sequence of operation instructions is formed, which fully restores the creative process and provides original data support for subsequent exploration of creative intentions, making up for the shortcomings of only focusing on the finished work and ignoring the creative process.

[0053] In step S112b, the intent feature refers to the feature that represents the core creative purpose of the creator, obtained by mining the sequence of operation instructions in the creation process, and reflects the original design intention of the user-generated work (such as building an automated farm, designing a logical puzzle, etc.).

[0054] By analyzing the type, order, and logical connections of the operation instruction sequence (such as frequently executing operations to set triggers and connect circuits), the core features that reflect the creator's creative purpose (such as the intention to create a logic puzzle) can be extracted. This links the creative process with the creative purpose, allowing the characteristics of the work to extend from the surface presentation to the deeper intention, laying the foundation for accurately matching the real needs of game players.

[0055] This application's embodiments effectively solve the problems of superficial content understanding and lack of creative intent by acquiring the sequence of operation instructions during the creation process and determining intent features accordingly. This extends the understanding of user-generated works from merely focusing on the surface information of the finished product to uncovering the deeper purpose of the creation process, avoiding situations where clickbait content is misleadingly recommended due to superficial information, or high-quality works are buried due to simple expression. Intent features directly reflect the creator's core creative purpose, accurately resonating with the real interests and needs of game players, significantly improving the matching accuracy between works and game players, allowing game players to quickly obtain content that meets their needs and enhancing the gaming experience. Simultaneously, intent features are extracted based on the operation sequence of the creation process, without relying on the work's historical interaction data. Even newly released works lacking past exposure records can obtain recommendation opportunities by matching their own intent features with the target player profile, providing effective support for the cold start of new content.

[0056] In some embodiments, the intention features of the user-generated work are obtained by analyzing the sequence of operation instructions through a trained intelligent model.

[0057] If the sequence of operation instructions is greater than or equal to the length threshold, a trained Transformer model is used for analysis to obtain the intent features of the user-generated work. The Transformer model is a model based on a self-attention mechanism, which can quickly focus on key operations in long sequence data processing, improving the accuracy of analysis.

[0058] If the sequence of operation instructions is shorter than the length threshold, a trained LSTM or GRU model is used for analysis to obtain the intent features of the user-generated work. LSTM stands for Long Short-Term Memory network model, which excels at processing short sequence data and can effectively capture the logical connections between operations in a short sequence. GRU stands for Gated Recurrent Unit model, which is similar in function to LSTM and is also suitable for processing short sequence data, but with better computational efficiency.

[0059] By analyzing the sequence of operation instructions step by step in the above manner, the creative logic and core purpose reflected in the sequence are mined out, and the corresponding intent features are finally output to ensure the accuracy of intent mining of long sequence data.

[0060] This application's embodiments select an appropriate model for intent feature extraction based on the length of the operation instruction sequence, effectively solving the problems of insufficient accuracy and low efficiency of traditional single models when processing sequences of different lengths, and significantly improving the accuracy and adaptability of intent feature recognition. For short sequences, LSTM or GRU models can fully capture the long-term correlation logic between operations, avoiding the loss of intent information due to excessively short sequences, and accurately restoring the core purpose of complex creative processes. For long sequences, Transformer models can quickly focus on key operations, avoid redundant calculations, improve intent extraction efficiency, and ensure that core intents are not missed. Accurate and efficient intent feature extraction provides a reliable and in-depth basis for subsequent matching of work features with player profiles.

[0061] Please see Figure 4 , Figure 4 A flowchart illustrating the process of determining a player profile according to an embodiment of this application is shown. Embodiments of this application provide steps for determining a player profile, including: Step S201: Construct a long-term profile of the player based on their historical game behavior within a set time period; Step S202: Based on the player's current game behavior after this login, construct a short-term profile of the player. Step S203: Determine the player profile of the game player based on the long-term profile and short-term profile of the game player.

[0062] The three steps described above are described in detail below.

[0063] In step S201, the duration is set: a predefined time range for collecting historical game behavior data (e.g., 30 days, 90 days) to ensure that the long-term profile reflects stable preferences. Historical game behavior refers to various game-related behaviors (such as past downloads, games, likes, favorites, etc.) that game players have engaged in within the set duration, and is the basic data for extracting stable preferences.

[0064] Long-term profiles refer to virtual models built based on a player's historical gaming behavior within a set timeframe. These models reflect a player's stable interests, preferences, and habits, and possess continuity and stability. (Examples include gaming behavior within a set timeframe from the current time, or all gaming behavior within a set timeframe from the current time, excluding gaming behavior since the current login.) Acquire historical game behavior data (such as download records, playtime, favorite content, and interaction behavior) of game players within a set time period from the current time. By analyzing the patterns in this data (such as frequent downloads of building-related content and a long-term preference for puzzle gameplay), extract persistent interest characteristics to form a long-term profile. Ensure that the profile reflects the core and stable needs of game players and provide a basic direction for recommendations.

[0065] In some embodiments, the current time refers to the time when the technical solution described in this application is triggered, that is, the time when the triggering event occurs. The triggering event can trigger the technical solution described in this application. The triggering event may include a game player logging in, a game player changing game mode, a game player's current login duration reaching the trigger duration, receiving a recommendation list update instruction, or the number of user-generated works increasing by a set value since the last time the target work was determined. The triggering event can be customized.

[0066] In step S202, current game behavior refers to the real-time behavior generated by the player after logging into the game (such as current searches, recent browsing, and immediate actions), which is key data for capturing the player's real-time needs. Short-term profile refers to a virtual model constructed based on the player's current game behavior after logging into the game, reflecting the player's immediate needs and temporary interests, and possessing both real-time and dynamic characteristics.

[0067] This system acquires the player's current gaming behavior after this login (such as currently searching for parkour games, recently browsing 3 parkour-like games, and real-time actions in the editor). By analyzing these immediate behaviors, it uncovers the player's current temporary interests or immediate needs, forming a short-term profile to compensate for the shortcomings of long-term profiles in adapting to dynamic changes in interests. This login is the player's most recent login at the time of the current login, which is used to determine the short-term profile.

[0068] In step S203, the player profile refers to the complete model formed by integrating long-term and short-term profiles, which comprehensively represents the interests and preferences of game players (including stable preferences and real-time needs) and is the core basis for matching the characteristics of the work.

[0069] By integrating the stable interests reflected in long-term profiles with the real-time needs reflected in short-term profiles, a complete player profile is formed that neither deviates from the core preferences of game players nor fails to respond to immediate interests. This avoids recommendation bias caused by single-dimensional profiles and lays a comprehensive foundation for the accurate matching of subsequent game features with the needs of game players.

[0070] This application's embodiments construct long-term and short-term profiles separately and integrate them to form a complete player profile, breaking away from the one-sidedness of solely relying on historical player behavior or focusing only on immediate behavior. The long-term profile is built based on historical behavior over a set period, accurately capturing stable player interests and preferences, ensuring recommendations do not deviate from core player needs, and avoiding recommendation chaos caused by excessive pursuit of immediate behavior. The short-term profile is built based on current login behavior, capturing real-time changes in player interests and immediate needs, solving the deficiency of traditional static profiles in adapting to dynamic player interests, and making recommendations more aligned with current player needs. The complete player profile formed by the fusion of these two profiles retains the core preferences accumulated by players over a long period while incorporating real-time interest dynamics, providing a comprehensive and reliable basis for the accurate matching of subsequent content features with player needs, significantly improving recommendation accuracy, allowing players to quickly obtain content that matches both their long-term preferences and immediate needs, enhancing the gaming experience, and enabling creators' works to reach target players more accurately, especially works that align with players' real-time needs, gaining more exposure opportunities and promoting the healthy development of the platform's UGC ecosystem.

[0071] Please see Figure 5 , Figure 5 A flowchart illustrating a process for determining a player profile based on a long-term and short-term profile of a game player, according to an embodiment of this application, is shown. This embodiment provides step S203, which involves determining a player profile based on a long-term and short-term profile of a game player, including: Step S301: Based on the current game mode of the player, determine the first weight corresponding to the long-term profile and the second weight corresponding to the short-term profile, and the sum of the first weight and the second weight is 1. Step S302: Combine the long-term profile and the short-term profile according to the first weight and the second weight to obtain the player profile.

[0072] The two steps described above are described in detail below.

[0073] In step S301, the game mode refers to the core behavioral scenario in which the player is currently in the game (such as the creative mode or consumption mode in a sandbox game), which is the core basis for adjusting the profile weight.

[0074] The first weight refers to the weight value assigned to the long-term profile, which represents the importance of the long-term profile in the final player profile. The second weight refers to the weight value assigned to the short-term profile, which represents the importance of the short-term profile in the final player profile, and its sum with the first weight is fixed at 1.

[0075] First, the game mode (e.g., Creative Mode, Consumption Mode) currently in which the player is playing is identified. "Current time" refers to the time when the technical solution proposed in this application was triggered. Then, based on the characteristics of this mode, a first weight (long-term profile) and a second weight (short-term profile) are assigned, strictly ensuring that their values ​​add up to 1. For example, Creative Mode emphasizes immediate creative needs, so the second weight is higher than the first weight; Non-Creative Mode emphasizes stable preferences, so the first weight is higher than the second weight. This weight allocation achieves the adaptation of the profile to the scene. The first and second weights differ in different modes.

[0076] In step S302, the stable interests reflected in the long-term profile and the immediate needs reflected in the short-term profile are weighted and calculated. The features of the two types of profiles are integrated into a unified feature vector according to their importance. Finally, a complete player profile is formed that fits the current scene and does not deviate from the core preferences of game players, avoiding scene adaptation deviation caused by a single profile or fixed weight.

[0077] For example, the product of the long-term profile and the first weight is taken as the first product, the product of the short-term profile and the second weight is taken as the second product, and the sum of the first product and the second product is taken as the player profile.

[0078] For example, game players can be in two modes: First, players who are not creating content but simply playing the game (e.g., exploring the map) can be called in a "consumption mode." Second, players who are creating content (e.g., creating maps or map guides in the game and publishing them on a platform) can be called in a "creation mode." If a player is in creation mode, the first weight is less than the second weight; if a player is not in creation mode, the first weight is greater than the second weight.

[0079] In Creation mode, the first weight is 0.3 and the second weight is 0.7. In Consumption mode, the first weight is 0.6 and the second weight is 0.4.

[0080] In this embodiment, by combining the weights of long-term and short-term profiles based on the player's current game mode and then weighting and fusing them, the player profile is upgraded from a general type to a scenario-adaptive type. The game mode is directly related to the player's current core needs. The weights are adjusted based on the mode so that the ratio of the first and second weights accurately matches the scenario characteristics. For example, the second weight is higher in creative mode, which can highlight the player's immediate creative reference needs. In non-creative mode, the first weight is greater, which can ensure that the recommendations are in line with the player's stable preferences and significantly improve the fit between the player profile and the current needs. The weighted fusion method not only retains the core interests of the player reflected in the long-term profile, avoiding recommendations from deviating from the essential needs, but also strengthens the immediate needs orientation of the short-term profile through the second weight. It achieves a dual consideration of stable preferences and scenario needs, providing a scenario-based basis for the accurate matching of subsequent work features with user needs. This allows players to obtain content that fits their current needs in different modes, improving the game experience. At the same time, it allows creators' works to reach players in the corresponding modes more accurately, improving the scenario-based exposure efficiency of high-quality content and promoting the scenario-based healthy development of the platform's UGC ecosystem.

[0081] Please see Figure 6 , Figure 6 This document illustrates a flowchart illustrating how, according to an embodiment of this application, a target work is determined from user-generated works based on the work characteristics of user-generated works and the player profile of a game player. The embodiment of this application provides step 120 for determining a target work from user-generated works based on the work characteristics of user-generated works and the player profile of a game player, including: Step S121: Normalize the features of the work and the features of the player profile to obtain normalized features; Step S122: Determine the matching score between user-generated works and player profiles based on the interaction between normalized features; Step S123: Based on the matching score between each user-generated work and the player profile, determine the target work among the user-generated works.

[0082] The two steps described above are described in detail below.

[0083] In step S121, normalized features refer to the work features and player profile features with uniform numerical ranges after normalization processing, ensuring that different types of features are comparable.

[0084] Since the features of the work (such as visual features and text features) and the features of the player profile may be in different numerical ranges (such as some features having values ​​of 0-1000 and others having values ​​of 0-1), direct analysis would lead to the large-scale features dominating the results. Therefore, normalization is used to convert all features to a uniform scale to ensure that each type of feature has equal weight in subsequent analysis and to avoid matching bias caused by scale differences.

[0085] In step S122, the matching score refers to a numerical value (such as the range of 0-1) that quantifies the degree of fit between the features of the work and the features of the player profile. The higher the score, the more the work meets the needs of the game players.

[0086] By analyzing the interaction between normalized work features and player profile features (such as the fit between the work's puzzle-solving intention features and the player profile's puzzle-solving preference features, and the matching degree between the work's visual style features and the user's style preference features), the algorithm calculates the degree of fit between the two and presents it in the form of a matching score, making the fit quantifiable and comparable.

[0087] In step S123, the target work refers to the user-generated work that is highly compatible with the needs of game players and is selected based on the matching score.

[0088] The matching scores of all candidate user-generated works are sorted, and works with high scores are selected as target works to ensure that the selected works are highly consistent with the needs of game players, avoid the problem of insufficient accuracy caused by blind recommendations, and provide clear content basis for subsequent push processes.

[0089] For example, a set number of user-generated works are selected as target works in descending order of matching scores. That is, user-generated works with higher matching scores are selected as target works first.

[0090] This application's embodiments effectively solve the core problems of low recommendation accuracy and large deviations between content and demand by employing a process of feature normalization, matching score calculation based on feature interactions, and filtering target works according to the score. Feature normalization eliminates the scale differences between different types of features, avoids the defect of a single feature dominating the matching result, and establishes the analysis of work features and player profile features on a fair and comparable basis, thereby improving the scientific nature of the matching logic. Calculating the matching score based on normalized feature interactions can comprehensively capture the deep correlation between the work and the needs of game players, and compared with traditional matching methods that only rely on surface labels or single behaviors, it can more accurately reflect the degree of fit between the two.

[0091] Please see Figure 7 , Figure 7 This document illustrates a flowchart illustrating how, according to an embodiment of this application, a target work is determined from user-generated works based on a matching score between each user-generated work and a player profile. The embodiment of this application provides step S123, which involves determining a target work from user-generated works based on a matching score between each user-generated work and a player profile, including: Step S401: Select the first number of user-generated works as target works according to the matching scores from largest to smallest. Step S402: After determining the first number of target works, continue to judge whether the similarity between the user-generated works and the target works is greater than the similarity threshold in descending order of matching scores. Step S403: If the similarity between the user-generated work and the target work is greater than the similarity threshold, the user-generated work is excluded; if the similarity between the user-generated work and the target work is less than or equal to the similarity threshold, the user-generated work is taken as the target work; until the number of target works reaches the second number.

[0092] The above three steps are described in detail below.

[0093] In step S401, the matching score is used to quantify the degree to which user-generated works match the player profile, and is the core quantitative basis for selecting target works. The first quantity refers to the number of target works selected in the first stage based on the matching score (e.g., 5), which is used to ensure the core accuracy of the recommended content.

[0094] All user-generated works are sorted in descending order of matching score, and the top 5 works with the highest scores are selected as the target works to ensure that the core content of the recommendation list is highly consistent with the needs of game players and to avoid sacrificing core accuracy due to excessive pursuit of diversity.

[0095] In step S402, the similarity threshold refers to a preset numerical standard (such as 0.8) for judging the degree of similarity between the user-generated work to be screened and the target work. It is used to screen differentiated content and avoid duplication.

[0096] After the target works are determined in the first stage, the remaining user-generated works are still processed in descending order of matching scores. The similarity between each candidate work and the selected target work (such as the degree of matching of content features and intent features) is calculated and compared with the preset similarity threshold to determine whether there is excessive similarity.

[0097] In step S403, the second quantity refers to the total number of target works to be pushed to game players (e.g., 20), which is the preset length of the recommendation list.

[0098] For each user-generated work currently undergoing similarity assessment, if the similarity exceeds a threshold (i.e., the content is highly similar), the work is removed to avoid monotony and repetition in the recommendation list; if the similarity does not exceed the threshold (i.e., the content is different), it is included in the target works. This process continues until the number of target works reaches a preset second number, ultimately forming a recommendation list that is accurate in its core and diverse in its content.

[0099] It should be clarified that the first quantity, the second quantity, and the similarity threshold can be set by technical personnel according to the actual scenario or dynamically adjusted based on preset rules.

[0100] In this embodiment, a two-stage strategy of precision priority and similarity filtering is used to determine target works, effectively solving problems such as severe homogenization of recommended content and information cocoons, while balancing recommendation accuracy and content diversity. In the first stage, the first number of works are selected based on matching scores to ensure that the core recommended content is highly aligned with the needs of game players, avoiding a decrease in the quality of core recommendations due to the pursuit of diversity, and allowing game players to quickly obtain high-quality content that best suits their needs. In the second stage, content that is highly similar to the selected works is filtered through similarity thresholds, while retaining differentiated high-quality works. This avoids a decline in the game player experience caused by monotonous and repetitive recommendation lists, and provides exposure opportunities for more high-quality works of different types and styles (including long-tail content), breaking the ecological imbalance problem of traffic concentration on top-tier similar content. The final target works not only ensure core accuracy but also possess rich diversity, which not only enhances the content exploration fun and platform stickiness of game players but also provides fair exposure opportunities for creators with diverse creative styles and unique ideas, promoting the healthy and sustainable development of the platform's UGC ecosystem.

[0101] In some embodiments, this technical solution can be implemented in the following ways to complete the selection and recommendation of target works.

[0102] Firstly, by using a well-trained feature extraction model, the features of user-generated works are extracted.

[0103] Input UGC content (user-generated works) images (cover images, keyframes) and text (titles, descriptions, comments), as well as the creator's operation command sequence, and perform feature extraction to obtain content features and intent features in the work features. Such as visual features V (512 dimensions), text features T (768 dimensions), and intent features (50 dimensions).

[0104] Visual features: Image preprocessing, such as resizing and normalization, is performed. Feature extraction is then conducted: a pre-trained ResNet-50 is used to extract 2048-dimensional visual features, which are then reduced to 512 dimensions using PCA. Visual features from various image types are fused to obtain the visual features of the user-generated artwork, such as a weighted average of the cover image (weight a), keyframes (weight b), and thumbnails (weight c). The sum of a, b, and c is 1.

[0105] Text features: The text undergoes preprocessing, such as word segmentation, stop word removal, cleaning, and length limitation. Feature extraction is then performed, using BERT (BERT-base-chinese) or Word2Vec models to extract features for the title, description, and comments (768 dimensions). For comments, average pooling or an attention mechanism can be used to aggregate features from multiple comments. Feature fusion is then performed, such as weighted fusion of title (d), description (e), comment (f), and tag (g) using an attention mechanism to obtain the 768-dimensional text features. The sum of d, e, f, and g is 1.

[0106] Intent characteristics: The operation instruction sequence is preprocessed, invalid operations are filtered out, encoding is converted, and the length is standardized (e.g., padding with 0s if insufficient, truncating to 50 if excessive). Short sequences (less than 50): feature extraction is performed using an LSTM / GRU model. Long sequences (50 or more): feature extraction is performed using a Transformer model (6-layer encoder, 8-head attention). Intent features are obtained (e.g., 50 dimensions), with the dimension representing the number of intent labels.

[0107] In some embodiments, the output intent label probability distribution is used, and the top N (positive integer) intent labels with the highest probabilities are selected as intent features, while other intent labels are set to null values. In this case, the intent features still retain their original dimensionality, such as 50 dimensions.

[0108] The operation type system can include: Basic operations: placing, deleting, moving, editing, etc. Trigger operations: creating triggers, setting conditions, setting actions, etc. Circuit operations: placing redstone, connecting circuits, setting logic gates, etc. Script operations: writing scripts, setting variables, calling functions, etc. Creature operations: placing creatures, setting AI, setting attributes, etc. Terrain operations: terrain generation, terrain editing, terrain copying, etc. Building operations: quick construction, building tools, building decorations, etc. Gameplay settings: game rules, victory conditions, scoring system, etc.

[0109] Intent tags can include: Puzzle: logic puzzles, mechanism puzzles, code puzzles, etc.; Parkour: parkour challenges, jumping challenges, racing parkour, etc.; Building: building showcases, building tutorials, building replication, etc.; Combat: PVP battles, PVE challenges, tower defense, etc.; Creative: automated farms, redstone circuits, mechanical devices, etc.; Story: story maps, role-playing, narrative, etc.; Mini-games: puzzle games, casual games, etc.; Other: tutorial maps, demonstration maps, etc.

[0110] Secondly, player profiles are determined through trained models.

[0111] Input: Historical behavior data of game players, real-time behavior data, and current mode. Output: Player profile (128 dimensions).

[0112] Long-term portrait: This retrieves player behavior data from a specified timeframe (e.g., 30 days), including: download behavior (content ID and download time); play behavior (play duration, completion rate, and replay count); and interaction behavior (likes, favorites, shares, and comments). These are the historical game behaviors.

[0113] Feature extraction is performed to obtain a long-term profile (which can be 128-dimensional), which can be updated at set intervals and used directly when implementing this technical solution.

[0114] Short term portrait: Get the player's current search terms: search keywords. Recently viewed: the 3-5 most recently viewed items. Current actions: actions taken in the editor or while playing. These are the current game behaviors.

[0115] An attention mechanism is used to extract features from search terms, browsing history, and current actions, and these features are then weighted and aggregated, mapped to a 128-dimensional vector, and updated in real time to obtain a short-term profile.

[0116] Pattern awareness: Creative Mode (First Weight 0.3, Second Weight 0.7): Operate in the editor, browse tutorials, and import resources; Consumption Mode (First Weight 0.6, Second Weight 0.4): Play on maps, browse maps, and download maps; Exploration Mode (First Weight 0.5, Second Weight 0.5): Browse a large number of different types of maps; Learning Mode (First Weight 0.4, Second Weight 0.6): Browse tutorials but do not start creating.

[0117] Player profile fusion: Player profile = first weight × long-term profile + second weight × short-term profile, with the first and second weights dynamically adjusted according to the game mode.

[0118] Third, the matching score between player profiles and user-generated works is determined through a trained model.

[0119] Input: Game characteristics, player profile. Output: Target game recommendation list (e.g., 20 items).

[0120] Feature splicing involves normalizing and splicing together the characteristics of the work and the features of the intention.

[0121] Interactive features are calculated by multiplying the Hadamard product among various features. This includes calculating the Hadamard product among normalized visual features, text features, intent features, long-term profile features, short-term profile features, and player profile features. All features are then concatenated to obtain a composite feature.

[0122] The DeepFM model uses an FM layer to learn second-order feature interactions; then, a DNN layer with full connectivity pools the features to learn higher-order feature interactions. The output is a matching score (0-1) between user-generated artwork and player profiles. The matching score for each user-generated artwork and player profile is calculated.

[0123] Fourth, determine the target works (e.g., 20 target works are needed). The first 5: sort by matching score (to ensure quality). The last 15: sort by MMR score (to ensure diversity). MMR algorithm (higher similarity results in lower score): MMR(i) = λ × Score(i) - (1-λ) × MaxSim(i, R), where Score is the matching score, MaxSim is the maximum similarity with the target work, and λ is a balancing parameter (e.g., 0.7).

[0124] For example, in descending order of matching scores, the first number (5) of user-generated works are selected as target works; after determining the first number of target works, the MMR of user-generated works and target works is determined in descending order of matching scores to see if it is greater than the score threshold. If a user-generated work is less than or equal to the rating threshold, it is excluded; if it is greater than the rating threshold, it is selected as a target work; this continues until the number of target works reaches the second limit (15). Both the first and second limits are positive integers.

[0125] Figure 8 A block diagram of a computer device for implementing a method for recommending user-generated works, according to an embodiment of this application, is shown.

[0126] It should be noted that, Figure 8 The computer device 800 shown is merely an example and should not impose any limitation on the functionality and scope of use of the embodiments of this application.

[0127] like Figure 8 As shown, the computer device 800 includes a central processing unit (CPU) 801, which can perform various appropriate actions and processes based on programs stored in read-only memory (ROM) 802 or programs loaded from storage section 808 into random access memory (RAM). The RAM 803 also stores various programs and data required for device operation. The CPU 801, ROM 802, and RAM 803 are interconnected via a bus 804. An input / output interface 805 (I / O interface) is also connected to the bus 804.

[0128] The following components are connected to the input / output interface 805: an input section 806 including a keyboard, mouse, etc.; an output section 807 including a cathode ray tube (CRT), liquid crystal display (LCD), etc., and speakers, etc.; a storage section 808 including a hard disk, etc.; and a communication section 809 including a network interface card such as a local area network card, modem, etc. The communication section 809 performs communication processing via a network such as the Internet. A drive 810 is also connected to the input / output interface 805 as needed. A removable medium 811, such as a disk, optical disk, magneto-optical disk, semiconductor memory, etc., is installed on the drive 810 as needed so that computer programs read from it can be installed into the storage section 808 as needed.

[0129] 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 communication section 809, and / or installed from removable medium 811. When the computer program is executed by central processing unit 801, it performs the various functions defined in the device of this application.

[0130] It should be noted that the computer-readable medium shown in the embodiments of this application can be a computer-readable signal medium, a computer-readable storage medium, or any combination thereof. A computer-readable storage medium can be, for example,—but not limited to—an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor device, apparatus, or any combination thereof. More specific examples of a computer-readable storage medium may include, but are not limited to: an electrical connection having one or more wires, a portable computer disk, a hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM), flash memory, optical fiber, portable compact disk read-only memory (CD-ROM), optical storage device, magnetic storage device, or any suitable combination thereof. In this application, a computer-readable storage medium can be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution device, apparatus, or apparatus. In this application, a computer-readable signal medium can include a data signal propagated in baseband or as part of a carrier wave, carrying computer-readable program code. Such transmitted data signals can take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. The computer-readable signal medium can also be any computer-readable medium other than a computer-readable storage medium, which can send, propagate, or transmit a program for use by or in connection with an instruction execution device, apparatus, or apparatus. The program code contained on the computer-readable medium can be transmitted using any suitable medium, including but not limited to wireless, wired, etc., or any suitable combination thereof.

[0131] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of devices, methods, and computer program products according to various embodiments of this application. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in a block diagram or flowchart, and combinations of blocks in a block diagram or flowchart, may be implemented using a dedicated hardware-based device that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions.

[0132] It should be noted that although several modules or units for the device used to perform actions have been mentioned in the detailed description above, this division is not mandatory. In fact, according to the embodiments of this application, the features and functions of two or more modules or units described above can be embodied in one module or unit. Conversely, the features and functions of one module or unit described above can be further divided and embodied by multiple modules or units.

[0133] Through the above description of the embodiments, those skilled in the art will readily understand that the exemplary embodiments described herein can be implemented by software or by combining software with necessary hardware. Therefore, the technical solutions according to the embodiments of this application can be embodied in the form of a software product, which can be stored in a non-volatile storage medium (such as a CD-ROM, USB flash drive, external hard drive, etc.) or on a network, including several instructions to cause a computing device (such as a personal computer, server, touch terminal, or network device, etc.) to execute the methods according to the embodiments of this application.

[0134] Other embodiments of this application will readily occur to those skilled in the art upon consideration of the specification and practice of the invention disclosed herein. This application is intended to cover any variations, uses, or adaptations of this application that follow the general principles of this application and include common knowledge or customary techniques in the art not disclosed herein.

[0135] It should be understood that this application is not limited to the precise structure described above and shown in the accompanying drawings, and various modifications and changes can be made without departing from its scope. The scope of this application is limited only by the appended claims.

Claims

1. A method for recommending user-generated works, characterized in that, The method includes: Acquire the characteristics of user-generated works, including content characteristics and intent characteristics; Based on the characteristics of the user-generated works and the player profiles of the game players, target works are identified from the user-generated works and pushed to the game players.

2. The method according to claim 1, characterized in that, Obtain the characteristics of user-generated works, including: Feature extraction is performed on the images in the user-generated works to obtain the visual features of the user-generated works; Feature extraction is performed on the text in the user-generated work to obtain the text features of the user-generated work. The content features include the visual features and the text features.

3. The method according to claim 1, characterized in that, Obtaining the characteristics of user-generated works also includes: Obtain the sequence of operation instructions used when creating the user-generated work; Based on the sequence of operation instructions, the intent characteristics of the user-generated work are determined.

4. The method according to claim 1, characterized in that, The method further includes: Based on the player's historical game behavior within a set time period, a long-term profile of the player is constructed. Based on the player's current game behavior after this login, construct a short-term profile of the player; Based on the long-term and short-term profiles of the game players, the player profile of the game player is determined.

5. The method according to claim 4, characterized in that, Based on the long-term and short-term profiles of the game players, a player profile is determined, including: Based on the current game mode of the player, determine the first weight corresponding to the long-term profile and the second weight corresponding to the short-term profile, and the sum of the first weight and the second weight is 1; The player profile is obtained by combining the long-term profile and the short-term profile based on the first weight and the second weight.

6. The method according to claim 1, characterized in that, Based on the characteristics of the user-generated works and the player profiles of the game players, the target works are identified from the user-generated works, including: The characteristics of the works and the characteristics of the player profiles are normalized to obtain normalized features; Based on the interaction between normalized features, a matching score is determined between the user-generated work and the player profile; The target work is determined from the user-generated works based on the matching score between each user-generated work and the player profile.

7. The method according to claim 6, characterized in that, Based on the matching scores between each user-generated work and the player profile, target works are identified from the user-generated works, including: The first number of user-generated works, ranked from highest to lowest matching score, are selected as the target works. After determining the first number of target works, the similarity between the user-generated works and the target works is determined sequentially according to the matching scores from largest to smallest to see if the similarity is greater than the similarity threshold. If the similarity between the user-generated work and the target work is greater than the similarity threshold, the user-generated work is excluded; if the similarity between the user-generated work and the target work is less than or equal to the similarity threshold, the user-generated work is taken as the target work; until the number of target works reaches the second number.

8. A device for recommending user-generated works, comprising a memory, a processor, and a readable program stored in the memory, characterized in that, The processor executes the readable program to implement the method of any one of claims 1 to 7.

9. A readable storage medium, characterized in that, It stores a readable program / instruction that, when executed by a processor, implements the method of any one of claims 1 to 7.

10. A program product comprising a readable program / instructions, characterized in that, When the readable program / instruction is executed by the processor, the method of any one of claims 1 to 7 is implemented.

Images