Artificial intelligence based game development dynamic content generation method and system

Abstract

The application provides a game development dynamic content generation method and system based on artificial intelligence, and relates to the technical field of game development. First, the dynamic content generation requirements corresponding to game development are acquired, and requirement features are extracted to form a requirement feature spectrum. Then, adaptive elements are screened from a game development resource pool based on the requirement feature spectrum, and a content evolution factor is generated by calculating the adaptability. A pre-trained game content self-evolution AI model is called, the requirement feature spectrum and the content evolution factor are inputted to generate initial dynamic content of the game. Interaction marks of game test users with the initial content are collected, and game evolution dynamic content is generated in combination with the content evolution factor. Finally, a content evolution completeness index is calculated based on the evolution dynamic content, and a game dynamic content package is outputted when the requirements are met, so that efficient, intelligent and personalized game dynamic content generation is achieved.

Description

Technical Field

[0001] This invention relates to the field of game development technology, and more specifically, to a method and system for generating dynamic content for game development based on artificial intelligence. Background Technology

[0002] In game development, dynamic content generation is a key factor in enhancing game appeal and player retention. Traditional methods of dynamic content generation primarily rely on manual design and planning. Game developers need to pre-conceive of character behavior patterns, scene change rules, and task triggering conditions, and then manually write code to implement these functions. However, this approach has several limitations. On the one hand, manually designed content is often limited by the developer's personal experience and creativity, making it difficult to meet diverse player needs and constantly changing market trends. On the other hand, as games grow in scale and complexity, the cost and time investment in manually designing and maintaining dynamic content rises sharply, resulting in slow game updates and iterations, and an inability to respond promptly to player feedback and market changes.

[0003] In recent years, while some dynamic content generation methods based on rule engines and simple algorithms have emerged, these methods typically generate only relatively fixed and limited content, lacking flexibility and innovation. They cannot dynamically adjust and optimize based on players' real-time interactions, making it difficult to achieve a truly personalized gaming experience. Furthermore, existing methods also fall short in terms of the completeness and iterative efficiency of content generation, failing to ensure that the generated content fully covers game requirements and completes iterative updates within a specified timeframe. Summary of the Invention

[0004] In view of the aforementioned problems, and in conjunction with the first aspect of the present invention, embodiments of the present invention provide a method for generating dynamic content for game development based on artificial intelligence, the method comprising:

[0005] Obtain the dynamic content generation requirements corresponding to game development, extract the content style orientation, interaction depth requirements and content iteration frequency requirements from the dynamic content generation requirements, and form a requirement feature spectrum;

[0006] Based on the aforementioned demand feature spectrum, suitable basic game elements are selected from the game development resource pool, the compatibility degree between each basic game element and the aforementioned demand feature spectrum is calculated, and a content evolution factor is generated. The basic game elements include game character elements, game scene elements, and game task elements.

[0007] The pre-trained game content self-evolution AI model is invoked, and the demand feature spectrum and the content evolution factor are input to generate the initial dynamic content of the game. The initial dynamic content of the game includes character behavior flow, scene change flow and task trigger flow.

[0008] Collect content interaction imprints during the interaction process between game test users and the initial dynamic content of the game, and combine them with the content evolution factors to perform evolution processing on the initial dynamic content of the game to generate game evolution dynamic content. The content interaction imprints include character operation trajectory, scene stay trajectory and task execution trajectory.

[0009] Based on the game's dynamic evolution content, the coverage of the content evolution factors is calculated, and a content evolution completeness index is generated. When the content evolution completeness index meets the content iteration frequency requirements in the demand feature spectrum, a game dynamic content package is output.

[0010] In another aspect, embodiments of the present invention also provide a game development dynamic content generation system based on artificial intelligence, including a processor and a machine-readable storage medium connected to the processor. The machine-readable storage medium is used to store programs, instructions, or code, and the processor is used to execute the programs, instructions, or code in the machine-readable storage medium to implement the above-described method.

[0011] Based on the above, this invention achieves effective integration and utilization of game resources by comprehensively acquiring dynamic content generation requirements and forming a requirement feature spectrum. It then selects suitable basic game elements based on this spectrum and calculates the fit to generate content evolution factors. By calling a pre-trained game content self-evolution AI model and combining the requirement feature spectrum and content evolution factors, it generates initial dynamic game content, fully leveraging the powerful computing and learning capabilities of artificial intelligence to quickly generate diverse and demand-compliant content. Interaction imprints between game test users and the initial dynamic content are collected and combined with content evolution factors for evolutionary processing to generate evolving dynamic game content. This allows the content to be dynamically adjusted and optimized based on players' actual behavior, greatly enhancing the game's personalization and interactivity. Finally, based on the evolving dynamic content, the coverage of content evolution factors is calculated, and a completeness index is generated. When the index meets the requirements, a dynamic game content package is output, ensuring the completeness and iteration efficiency of the generated content. This effectively shortens the game development cycle, reduces development costs, and provides players with a richer, more novel, and personalized gaming experience, enhancing the game's market competitiveness and player stickiness. Attached Figure Description

[0012] Figure 1 This is a schematic diagram of the execution flow of the AI-based game development dynamic content generation method provided in this embodiment of the invention.

[0013] Figure 2 This is a schematic diagram of exemplary hardware and software components of the AI-based game development dynamic content generation system provided in this embodiment of the invention. Detailed Implementation

[0014] The present invention will now be described in detail with reference to the accompanying drawings. Figure 1 This is a flowchart illustrating an embodiment of the AI-based dynamic content generation method for game development. The following is a detailed description of this AI-based dynamic content generation method for game development.

[0015] Step S110: Obtain the dynamic content generation requirements corresponding to game development, extract the content style orientation, interaction depth requirements and content iteration frequency requirements from the dynamic content generation requirements, and form a requirement feature spectrum.

[0016] In this embodiment, taking the development of an open-world adventure game as an example, the dynamic content generation requirements corresponding to the game development are first obtained. These requirements originate from the game development team's planning scheme, which specifies that the game scene needs to cover various terrains, such as mountains, forests, plains, and rivers, and different terrains need to have unique visual styles. Simultaneously, the player's interaction methods in the scene should be diversified, including interaction with environmental objects and triggering hidden quests. Furthermore, the game scene content is required to be updated and iterated every two months. Based on the above dynamic content generation requirements, corresponding features are extracted to form a requirement feature spectrum.

[0017] Step S111: Analyze the dynamic content generation requirements corresponding to game development, extract the text information describing the content presentation style in the dynamic content generation requirements, and divide the content style dimensions, which include visual style dimension, narrative style dimension and interaction style dimension.

[0018] A detailed analysis of the acquired game development requirements was conducted, extracting key information from the text describing the presentation style. For example, the text mentions that "mountain scenes should present a majestic and rugged visual effect, with clear rock textures and snow cover; forest scenes should showcase a lush and verdant landscape, with tall trees and dappled light filtering through the leaves," which falls under the visual style dimension. "The narrative in the game scene unfolds through environmental details, such as ancient stone tablets recording historical events and abandoned camps hinting at past activities," which falls under the narrative style dimension. "Players can open hidden passages by interacting with switches in the scene, and obtain items by interacting with specific plants," which falls under the interaction style dimension. Following this approach, the content style is divided into visual style, narrative style, and interaction style dimensions.

[0019] Step S112: Extract feature description items under each content style dimension from the text information, convert each feature description item into quantifiable style feature values, and form a content style guide.

[0020] Within the visual style dimension, features such as "clarity of mountain rock texture," "density of forest trees," and "speed of river flow" are extracted. For "clarity of mountain rock texture," a quantifiable range is set based on the degree of "texture clarity" in the description, converting it into corresponding style feature values. For example, the clearest texture is set as a higher value, and a less clear texture as a lower value, with specific values ​​determined based on the actual description. Similarly, features such as "completeness of inscriptions on stone tablets" and "number of items left in campsites" under the narrative style dimension, and "susceptibility of switch responses" and "success rate of plant interactions" under the interaction style dimension, are processed in a similar way, converting them all into quantifiable style feature values. These style feature values ​​collectively constitute the content style guide.

[0021] Step S113: Extract the text information describing the degree of interaction between the user and the game content from the dynamic content generation requirements, and divide the interaction depth dimension. The interaction depth dimension includes the operation interaction dimension, the decision interaction dimension, and the emotional interaction dimension.

[0022] Information describing the degree of user interaction with game content is extracted from the requirements text to define the interaction depth dimension. For example, "Players can move stones in the scene by clicking and dragging to build temporary bridges," which belongs to the operational interaction dimension. "When faced with a fork in the road, the player's decision will affect the subsequent development of the scene; choosing the left path will lead to a dangerous area, while choosing the right path will lead to a friendly NPC," which belongs to the decision-making interaction dimension. "After the player helps an injured animal, the animal will provide guidance to the player in subsequent scenes, enhancing the player's sense of immersion and emotional connection," which belongs to the emotional interaction dimension. Based on the above classification method, the operational interaction dimension, decision-making interaction dimension, and emotional interaction dimension are determined.

[0023] Step S114: Extract the interaction requirements under each interaction depth dimension from the text information, and convert each interaction requirement into a quantifiable interaction feature value to form the interaction depth requirements.

[0024] In the operational interaction dimension, interaction requirements such as "precision requirements for stone movement" and "stability requirements for bridge construction" were extracted. For "precision requirements for stone movement," based on the requirement of "accurate placement in the designated location," quantifiable standards were set and converted into interaction feature values. In the decision-making interaction dimension, interaction requirements such as "the significance of the impact of path selection on the scene" and "the magnitude of scene differences caused by different choices" were also converted into quantifiable interaction feature values ​​according to corresponding standards. In the emotional interaction dimension, interaction requirements such as "the degree of animal response to player assistance" and "the intensity of player emotional engagement" were also quantified to form interaction feature values. These interaction feature values ​​collectively constitute the interaction depth requirements.

[0025] Step S115: Extract the text information describing the game content update cycle from the dynamic content generation requirements, divide the content iteration dimensions, extract the cycle requirement items under each content iteration dimension from the text information, convert each cycle requirement item into a quantifiable iteration feature value, and form the content iteration frequency requirement.

[0026] Information related to the game's content update cycle was extracted from the requirements document, and content iteration dimensions were divided, such as "scene terrain update cycle," "scene quest update cycle," and "scene NPC dialogue update cycle." For the "scene terrain update cycle," the requirement states "major terrain features are updated every two months," which is treated as a cycle requirement. A quantifiable standard was set according to the length of the time cycle, and this was converted into iterative feature values. The "scene quest update cycle" requires "5 new quests updated monthly," which was also converted into corresponding iterative feature values. The "scene NPC dialogue update cycle" is "dialogue content is updated every two weeks," which was also converted into iterative feature values. These iterative feature values ​​collectively form the content iteration frequency requirements.

[0027] Step S116: The content style guidance, the interaction depth requirement, and the content iteration frequency requirement are spliced ​​together in a preset dimension order to generate a demand feature spectrum. The number of dimensions of the demand feature spectrum is consistent with the sum of the number of dimensions of the content style dimension, the interaction depth dimension, and the content iteration dimension.

[0028] The preset dimension order is: content style orientation, interaction depth requirements, and content iteration frequency requirements. The style feature values ​​included in the previously obtained content style orientation are arranged in the order of visual style, narrative style, and interaction style; the interaction feature values ​​included in the interaction depth requirements are arranged in the order of operational interaction, decision-making interaction, and emotional interaction; and the iteration feature values ​​included in the content iteration frequency requirements are arranged in the order of scene terrain update cycle, scene task update cycle, and scene NPC dialogue update cycle. These three parts are then concatenated in the preset order to form a multi-dimensional requirement feature spectrum. The number of dimensions in this requirement feature spectrum is the sum of the three content style dimensions, the three interaction depth dimensions, and the three content iteration dimensions, i.e., nine dimensions.

[0029] Step S120: Based on the demand feature spectrum, select suitable basic game elements from the game development resource pool, calculate the fit degree between each basic game element and the demand feature spectrum, and generate content evolution factors. The basic game elements include game character elements, game scene elements, and game task elements.

[0030] Based on the generated demand feature spectrum, suitable basic game elements are selected from the game development resource pool. The game development resource pool contains a large number of candidate elements, including various game character elements, such as characters with different professions, appearances, and skills; game scene elements, such as various terrain modules, vegetation models, and building models; and game quest elements, such as main quest templates, side quest templates, and hidden quest templates. By calculating the fit between these candidate elements and the demand feature spectrum, suitable elements are selected as basic game elements, and content evolution factors are generated based on the fit.

[0031] Step S121: Extract all candidate game elements from the game development resource pool. The candidate game elements include candidate character elements, candidate scene elements, and candidate task elements.

[0032] The game development resource pool is a database storing a large number of elements needed for game development. All candidate game elements are extracted from it. Candidate character elements include character models for different classes such as warriors, mages, and archers, each with corresponding attribute information such as appearance, skill characteristics, and dialogue style. Candidate scene elements include modules for different terrains such as mountains, forests, plains, and rivers, as well as scene object models such as trees, rocks, and houses. Each model has detailed attribute descriptions such as size, material, and visual effects. Candidate task elements include various task templates, such as item collection tasks, monster killing tasks, and escort tasks. Each task template includes information such as task objectives, task flow, and task rewards.

[0033] Step S122: Extract the character style attributes, character interaction attributes, and character iteration attributes of the candidate character elements to form a character element feature vector.

[0034] For each candidate character element, extract its relevant attributes. Character style attributes include the character's appearance style, such as a warrior's heavy armor style or a mage's flowing robe style; the character's dialogue style, such as humorous or serious. Character interaction attributes include the character's interaction methods with other characters, such as cooperative skill release and dialogue choice branches; the character's ability to interact with scene objects, such as whether they can destroy specific rocks or activate specific mechanisms. Character iteration attributes include the frequency of character skill updates and the cycle of character appearance adjustments. These attributes are converted into quantifiable feature values ​​and arranged in the order of character style attributes, character interaction attributes, and character iteration attributes to form a character element feature vector.

[0035] Step S123: Extract the scene style attributes, scene interaction attributes, and scene iteration attributes of the candidate scene elements to form scene element feature vectors.

[0036] For each candidate scene element, its attributes are extracted. Scene style attributes include the scene's terrain style, such as the ruggedness of mountains or the seclusion of forests; and the scene's lighting style, such as bright daytime or dim nighttime. Scene interaction attributes include the number of interactive objects in the scene, such as the number of harvestable plants or openable treasure chests; and the interaction methods of interactive objects, such as click interaction or drag-and-drop interaction. Scene iteration attributes include the frequency of scene terrain changes and the update cycle of objects in the scene. After quantifying the above attributes, they are arranged in the order of scene style attributes, scene interaction attributes, and scene iteration attributes to form a scene element feature vector.

[0037] Step S124: Extract the task style attribute, task interaction attribute, and task iteration attribute of the candidate task element to form a task element feature vector.

[0038] For each candidate task element, its attributes are extracted. Task style attributes include the narrative style (e.g., suspenseful mystery, action-packed combat) and difficulty style (e.g., easy to complete, complex and challenging). Task interaction attributes include the number of interactions with NPCs and the number of interactions with scene objects. Task iteration attributes include the frequency of task updates and the cycle of task reward adjustments. After quantifying these attributes, they are arranged in the order of task style attributes, task interaction attributes, and task iteration attributes to form a task element feature vector.

[0039] Step S125: Extract feature components related to the character element from the demand feature spectrum to generate a character demand sub-spectrum, and calculate the cosine similarity between the character element feature vector and the character demand sub-spectrum as the character element fit.

[0040] Feature components related to character elements are selected from the requirement feature spectrum. These include style features related to character appearance and dialogue in the content style orientation, interaction features related to character interaction in the interaction depth requirement, and iteration features related to character updates in the content iteration frequency requirement. These feature components collectively constitute the character requirement sub-spectrum. Then, the cosine similarity between the character element feature vector of each candidate character element and the character requirement sub-spectrum is calculated. This is done by measuring the consistency of the directions of the two vectors; the closer the directions of the two vectors are, the higher the cosine similarity value, which is then used as the character element fit.

[0041] Step S126: Extract feature components related to scene elements from the demand feature spectrum to generate a scene demand sub-spectrum, and calculate the cosine similarity between the scene element feature vector and the scene demand sub-spectrum as the scene element fit.

[0042] Feature components related to scene elements are extracted from the demand feature spectrum. These include style feature values ​​related to scene visuals and narrative in the content style guidance, interaction feature values ​​related to scene interaction in the interaction depth requirement, and iteration feature values ​​related to scene updates in the content iteration frequency requirement. These feature components constitute the scene demand sub-spectrum. The cosine similarity between the scene element feature vector of each candidate scene element and the scene demand sub-spectrum is calculated, and this is used as the scene element fit. Similarly, the higher the cosine similarity, the higher the fit between the candidate scene element and the scene requirements.

[0043] Step S127: Extract feature components related to task elements from the demand feature spectrum to generate a task demand sub-spectrum, and calculate the cosine similarity between the task element feature vector and the task demand sub-spectrum as the task element fit.

[0044] Feature components related to task elements are extracted from the requirement feature spectrum. These include style features related to task narrative and difficulty in content style guidance, interaction features related to task interaction in interaction depth requirements, and iteration features related to task updates in content iteration frequency requirements. These feature components form the task requirement sub-spectrum. The cosine similarity between the task element feature vector of each candidate task element and the task requirement sub-spectrum is calculated and used as the task element fit. The level of cosine similarity reflects the fit between the candidate task element and the task requirements.

[0045] Step S128: Select candidate character elements, candidate scene elements, and candidate task elements whose character element adaptation meets the character adaptation threshold, scene element adaptation meets the scene adaptation threshold, and task element adaptation meets the task adaptation threshold, and use them as basic game elements.

[0046] Character, scene, and task adaptation thresholds are set, determined based on specific game development requirements and historical experience. For candidate character elements, only those with a character element adaptation degree greater than or equal to the character adaptation threshold are considered compliant. Similarly, the scene element adaptation degree of candidate scene elements must be greater than or equal to the scene adaptation threshold, and the task element adaptation degree of candidate task elements must be greater than or equal to the task adaptation threshold. After screening, the candidate character elements, candidate scene elements, and candidate task elements that meet the criteria are determined as the basic game elements.

[0047] Step S129: Combine the character element adaptability, the scene element adaptability, and the task element adaptability in the order of character, scene, and task to generate a content evolution factor, and associate and record the identification information of the character requirement sub-spectrum, scene requirement sub-spectrum, and task requirement sub-spectrum in the content evolution factor.

[0048] The suitability of the selected basic game elements for character elements, scene elements, and task elements is combined in the order of character, scene, and task to form a comprehensive vector, namely the content evolution factor. Simultaneously, to facilitate subsequent tracing and association, the content evolution factor records the identification information of each of the character requirement sub-spectrum, scene requirement sub-spectrum, and task requirement sub-spectrum. This identification information can be a unique code, allowing for quick retrieval of the corresponding sub-spectrum content.

[0049] Step S130: Call the pre-trained game content self-evolution AI model, input the demand feature spectrum and the content evolution factor, and generate the initial dynamic content of the game. The initial dynamic content of the game includes character behavior flow, scene change flow and task trigger flow.

[0050] The pre-trained game content self-evolution AI model is invoked. This model, trained with a large amount of game development data, can generate corresponding game content based on input features. The previously obtained demand feature spectrum and content evolution factors are input into the model. After internal processing and calculation, the model generates the initial dynamic content of the game, including the character's behavior flow in the game; the scene change flow as time and player behavior change; and the task triggering flow.

[0051] Step S131: Input the demand feature spectrum into the feature parsing layer of the game content self-evolution AI model, extract the feature components related to character behavior in the demand feature spectrum, and obtain the character behavior guidance features.

[0052] The demand feature spectrum is input into the feature parsing layer of the game content self-evolution AI model. The main function of this feature parsing layer is to analyze and extract the input features. During the parsing process, the feature parsing layer identifies feature components in the demand feature spectrum that are related to character behavior, such as interaction feature values ​​related to character operation in interaction depth requirements, and style feature values ​​related to character action style in content style guidance. These feature components are integrated and processed to obtain character behavior guidance features, which clarify the character's behavioral direction and style in the game.

[0053] Step S132: Extract the feature components related to scene changes from the demand feature spectrum to obtain scene change-oriented features.

[0054] The feature parsing layer continues to analyze the requirement feature spectrum, extracting feature components related to scene changes. These components include style feature values ​​related to visual changes in the content style guidance, iteration feature values ​​related to the scene update cycle in the content iteration frequency requirement, and interaction feature values ​​related to scene changes due to player behavior in the interaction depth requirement. After processing these feature components, the scene change guidance feature is obtained, which specifies the direction and pattern of scene changes in the game.

[0055] Step S133: Extract the feature components related to task triggering from the demand feature spectrum to obtain task triggering guidance features.

[0056] The feature parsing layer also extracts feature components related to task triggering from the requirement feature spectrum. For example, style feature values ​​related to task narrative triggering in content style guidance, interaction feature values ​​related to task triggering conditions in interaction depth requirements, and iteration feature values ​​related to task update triggering in content iteration frequency requirements. These feature components are then integrated to obtain task triggering guidance features, which clarify the triggering methods and conditions for tasks in the game.

[0057] Step S134: Input the character behavior guidance feature and the character element adaptation in the content evolution factor into the character content generation layer of the game content self-evolution AI model to construct character behavior logic and generate a character behavior flow containing continuous character actions and character dialogues.

[0058] The character content generation layer of the game content self-evolution AI model is input into the character behavior guidance features and the character element adaptability from the content evolution factors. Based on these inputs, the character content generation layer constructs the character behavior logic. During the construction process, the overall direction and style of the character's behavior can be determined according to the character behavior guidance features, and the details and intensity of the behavior can be adjusted by combining the character element adaptability. Finally, a character behavior flow is generated, which includes a series of continuous actions of the character in the game, such as walking, attacking, and jumping, as well as the content and order of dialogue between characters or between characters and NPCs.

[0059] Step S135: Input the scene change guidance feature and the scene element adaptation in the content evolution factor into the scene content generation layer of the game content self-evolution AI model to construct the scene change logic and generate a scene change flow that includes continuous scene terrain transformation and scene weather transformation.

[0060] The scene change guidance features and scene element adaptability from the content evolution factors are input into the scene content generation layer. Based on these inputs, the scene content generation layer constructs scene change logic, determines the overall trend and type of scene change according to the scene change guidance features, and adjusts the magnitude and frequency of change in conjunction with the scene element adaptability. The generated scene change stream includes continuous transformations of scene terrain, such as a gradual transition from plains to mountains, and changes in scene weather, such as from sunny to rainy, or from day to night.

[0061] Step S136: Input the task triggering guidance feature and the task element adaptation in the content evolution factor into the task content generation layer of the game content self-evolution AI model to construct the task triggering logic and generate a task triggering flow that includes continuous task condition triggering and task stage transition.

[0062] The task triggering guidance features and task element adaptability from the content evolution factors are input into the task content generation layer. Based on these inputs, the task content generation layer constructs task triggering logic, determines the overall conditions and methods of task triggering according to the task triggering guidance features, and adjusts the difficulty level of the triggering and the rhythm of stage transitions by combining the task element adaptability. The generated task trigger flow includes the continuous triggering of tasks under different conditions, such as triggering a task when the player arrives at a specific location, and the transition between different stages of the task, such as transitioning from the item collection stage to the item delivery stage.

[0063] Step S137: Perform timeline alignment processing on the character behavior flow, the scene change flow, and the task trigger flow to generate initial dynamic content for the game. The initial dynamic content for the game includes the character behavior flow, the scene change flow, and the task trigger flow.

[0064] After acquiring the character behavior flow, scene change flow, and quest trigger flow, these three need to be time-aligned. Specifically, this involves uniformly sorting the events within each flow according to their occurrence time. For example, in the character behavior flow, a character begins moving at a certain point in time; in the scene change flow, weather changes occur at the same point in time; and in the quest trigger flow, quest clues appear near that point in time. By adjusting the time stamps of each event, they are made consistent on the timeline. After this processing, the three are integrated to form the initial dynamic content of the game, which fully includes the time-aligned character behavior flow, scene change flow, and quest trigger flow.

[0065] Step S140: Collect content interaction imprints during the interaction process between the game test user and the initial dynamic content of the game, and combine them with the content evolution factor to perform evolution processing on the initial dynamic content of the game to generate game evolution dynamic content. The content interaction imprints include character operation trajectory, scene stay trajectory and task execution trajectory.

[0066] After the initial dynamic content of the game is generated, test users are invited to experience it. During the interaction between the test users and this initial dynamic content, a corresponding data collection mechanism is activated to record various user interactions, forming content interaction imprints. Subsequently, combined with the previously generated content evolution factors, the initial dynamic content of the game is adjusted and optimized, that is, evolved, to generate game evolution dynamic content that better conforms to user interaction habits. Specifically, the content interaction imprints include the trajectory of the user's character, the trajectory of staying in the scene, and the trajectory of performing tasks.

[0067] Step S141: Load the initial dynamic content of the game into the game runtime environment, start the interaction imprint collection program, record the character control operations of the game test user when interacting with the initial dynamic content of the game, and form a continuous character operation coordinate sequence as the character operation trajectory.

[0068] The initial dynamic content of the game is loaded into a dedicated game runtime environment, which has various functions to simulate the actual operation of the game. Then, an interaction imprint collection program is launched, which monitors and records the actions of the test user in real time. When the test user performs control actions on the character, such as moving, attacking, or jumping, the program records the coordinate information corresponding to these actions and arranges them in chronological order to form a continuous sequence of character action coordinates. This sequence of character action coordinates is the character's action trajectory.

[0069] Step S1411: Set up a character operation acquisition node in the game running environment. The character operation acquisition node runs synchronously with the character behavior flow in the initial dynamic content of the game.

[0070] Within the game's runtime environment, multiple character action capture nodes are set up according to preset distribution rules. The runtime of these capture nodes is consistent with the runtime of the character behavior flow in the initial dynamic content of the game, ensuring accurate capture of the user's control operations at each stage of the character's behavior. For example, when the character is walking in the character behavior flow, the corresponding capture node is also active, ready to record the user's walking control operations at any time.

[0071] Step S1412: When the game test user performs a character control operation, the character operation acquisition node records the trigger time point and the corresponding character operation type for each operation.

[0072] Whenever a game test user performs a control action on a character, such as pressing a direction key to move the character or clicking a skill button to release a skill, the character action data collection node will respond immediately. It will accurately record the specific time when the action is triggered, and at the same time identify and record the type of the action, whether it is a movement action, an attack action, or another type of action.

[0073] Step S1413: Based on the trigger time point, arrange each character operation type in chronological order to form a character operation time sequence.

[0074] After collecting the trigger times and corresponding role operation types for all operations, these role operation types are sorted based on the trigger times. The operation types are arranged sequentially from earliest to latest time, forming an ordered role operation time sequence. This sequence clearly shows the order in which users performed actions on their roles.

[0075] Step S1414: Assign corresponding operation coordinate values ​​to each type of character operation. The operation coordinate values ​​are related to the direction and range of the character's movement in the game scene.

[0076] For different character operation types, corresponding operation coordinate values ​​are pre-set. For example, moving forward corresponds to a specific range of coordinate values, moving backward corresponds to another range of coordinate values, and the coordinate values ​​of attack operations are related to the direction and force of the attack. The setting of the above operation coordinate values ​​is directly related to the actual direction and range of the character's actions in the game scene. The larger the range of the action, the further the corresponding coordinate value deviates from the middle value within its range.

[0077] Step S1415: Replace each type of character operation in the character operation time sequence with the corresponding operation coordinate value to form a coordinate sequence containing timestamps and operation coordinate values.

[0078] Based on the operation coordinate values ​​assigned to each character operation type previously, the character operation time series is transformed. Each character operation type in the sequence is replaced with its corresponding operation coordinate value, while retaining the original trigger time of each operation as a timestamp. This creates a coordinate sequence that includes both timestamps and corresponding operation coordinate values.

[0079] Step S1416: Perform continuous processing on the coordinate sequence, fill in the blank coordinate values ​​between adjacent timestamps, so that the coordinate sequence remains continuous in the time dimension, and mark the continuous coordinate sequence as the character operation trajectory, the character operation trajectory containing continuous timestamps and corresponding operation coordinate values.

[0080] Because there are time intervals between the trigger times of adjacent operations, the coordinate sequence is blank within these intervals. To maintain the continuity of the coordinate sequence in the time dimension, continuous processing is required. This processing involves calculating the coordinate values ​​for the blank time intervals between two adjacent known operation coordinate values ​​and their timestamps through interpolation, thus filling in these gaps. After this processing, the resulting coordinate sequence is continuous in time and is marked as the character's operation trajectory. This trajectory fully contains consecutive timestamps and the corresponding operation coordinate values ​​for each timestamp.

[0081] Step S142: Record the sequence of scene position coordinates when the game test user moves in the game scene, forming a continuous sequence of scene dwell coordinates as the scene dwell trajectory.

[0082] During game testing, as the user controls the character to move within the game scene, the character's location coordinates are recorded in real time. These coordinates are arranged chronologically to form a sequence of scene location coordinates. After further processing, a continuous sequence of scene dwell coordinates is formed. This sequence of scene dwell coordinates is the scene dwell trajectory, which reflects the user's dwell time at various locations within the scene.

[0083] Step S1421: Divide the game scene into multiple scene location areas, each scene location area corresponding to a unique area coordinate identifier.

[0084] Based on the terrain and layout of the game scene, the entire game scene is divided into several appropriately sized scene areas. For example, the forest scene is divided into densely wooded areas, forest clearings, and streamside areas. Each divided area is assigned a unique coordinate identifier, which can be accurately mapped to a specific location within the game scene.

[0085] Step S1422: When the game test user moves in the game scene, track the current position of the user-controlled character in real time, and determine the scene location area to which the current position belongs and the corresponding area coordinates.

[0086] By utilizing the position tracking function in the game's runtime environment, the system acquires the character's current position information in real time while the user controls the character's movement during game testing. Based on preset scene location area division rules, it determines which scene location area the current position belongs to and finds the corresponding area coordinates.

[0087] Step S1423: Record the entry time of the user-controlled character when entering each scene location area and the exit time when leaving the scene location area.

[0088] When a user-controlled character enters a specific scene location area, the exact time of entry is recorded. Similarly, when the character leaves the area, the exact time of departure is recorded. These time points are recorded with millisecond precision to ensure data accuracy.

[0089] Step S1424: Calculate the difference between the departure time and the entry time for each scene location area, and use it as the dwell time for that scene location area.

[0090] For each scene location area, the difference between the recorded departure time and the entry time is the duration the user-controlled character stays in that area. For example, if the entry time is 10:00:00 and the departure time is 10:00:30, then the duration of stay in that area is 30 seconds.

[0091] Step S1425: Arrange the area coordinates, entry time, exit time, and duration of stay for each scene location area in chronological order of entry time to form a scene location record sequence.

[0092] The system collects information such as the area coordinates, entry time, exit time, and duration of stay for all scene locations. Using the entry time as the sorting criterion, this information is arranged in ascending order to form a scene location record sequence. This sequence clearly shows the order in which users engage in different scene areas.

[0093] Step S1426: Perform time continuity verification on the scene location recording sequence, and convert the scene location recording sequence that passes the verification into a sequence containing continuous area coordinates and corresponding dwell time to form a scene dwell trajectory.

[0094] Temporal continuity verification primarily checks whether the departure and arrival times of two adjacent areas in the scene location recording sequence are consecutive, and whether there is any temporal overlap or large time intervals. If overlap exists, the records need to be verified and corrected; if large intervals exist, it needs to be confirmed whether they are due to missing records. After verifying and ensuring temporal continuity, the scene location recording sequence is converted into a sequence containing only continuous area coordinates and their corresponding dwell times; this sequence is the scene dwell trajectory.

[0095] Step S143: Record the sequence of task operation steps when the game test user performs the task, forming a continuous sequence of task execution steps as the task execution trajectory.

[0096] When users perform tasks in the initial dynamic content of the game during game testing, each of their operation steps is recorded. These steps are arranged in chronological order to form a sequence of task operation steps. After processing, a continuous sequence of task execution steps is formed. This sequence of task execution steps is the task execution trajectory, which can reflect the process of the user completing the task.

[0097] Step S1431: Analyze the task trigger flow in the initial dynamic content of the game to determine all task stages included in the task and the task operation requirements corresponding to each task stage.

[0098] A detailed analysis of the task triggering flow in the initial dynamic content of the game is conducted to clarify the various stages of the task. For example, a task to collect items may include three stages: finding the item, collecting the item, and delivering the item. Simultaneously, the specific requirements for user actions in each task stage are determined, such as the requirement for the user to reach a designated area in the item finding stage, and the requirement for the user to click on the item to collect it in the item collecting stage.

[0099] Step S1432: Assign a unique stage identifier to each task stage, and assign a unique operation identifier to each task operation requirement.

[0100] To facilitate differentiation and recording, each task stage is assigned a unique stage identifier, which can be a combination of letters and numbers. Similarly, each task operation requirement is also assigned a unique operation identifier, allowing for accurate mapping between specific task stages and operation requirements.

[0101] Step S1433: When the game test user performs a task, monitor the user's operation behavior in real time and determine whether each operation behavior meets the task operation requirements of the current task stage.

[0102] During game testing, user actions are monitored in real time as they perform tasks. Each action is compared to the requirements of the current task stage to determine if it meets the requirements. For example, in the item collection stage, clicking on an item meets the requirements, while attacking a monster does not.

[0103] Step S1434: If the operation behavior meets the task operation requirements, then record the operation identifier, operation time point and task stage identifier corresponding to the operation behavior.

[0104] When a user's action is detected to meet the task operation requirements of the current task stage, the operation identifier, the time point of the operation, and the task stage identifier to which the operation belongs are immediately recorded to ensure that every valid operation is accurately recorded.

[0105] Step S1435: Arrange all the operation identifiers, operation time points and task stage identifiers that meet the requirements in chronological order of operation time points to form a task operation record sequence.

[0106] Collect relevant information on all eligible operations, and arrange the operation identifier, operation time point, and task stage identifier in chronological order from earliest to latest to form a task operation record sequence. This task operation record sequence reflects the order in which the user performed effective operations during the task execution process.

[0107] Step S1436: Check whether the task operation record sequence covers all task stage identifiers. If there are uncovered task stage identifiers, supplement the default operation record corresponding to the corresponding task stage identifier.

[0108] Examine the task operation record sequence to see if it contains stage identifiers for all task stages. If a stage identifier is not covered, it means the user did not perform the required operation at that stage. In this case, the default operation record corresponding to that stage identifier needs to be added. The default operation record can be set as the initial operation requirements for that stage and a default operation time point.

[0109] Step S1437: Convert the supplemented task operation record sequence into a continuous task execution step sequence. Each step in the task execution step sequence includes an operation identifier, an operation time point, and a task stage identifier, forming a task execution trajectory.

[0110] The supplemented task operation record sequence is processed to maintain its continuity in the time dimension. If there is a large time interval between adjacent operations, default operation records are added to fill the gap. The resulting continuous sequence is the task execution step sequence, where each step includes an operation identifier, an operation time point, and a task stage identifier. This task execution step sequence is marked as the task execution trajectory.

[0111] Step S144: Integrate the character operation trajectory, the scene stay trajectory, and the task execution trajectory to form a content interaction imprint.

[0112] The character operation trajectory, scene dwell trajectory, and task execution trajectory obtained earlier are integrated. During the integration process, based on the timestamp, information from the same or similar time points in the three trajectories is matched to form content interaction imprints, which fully reflect the interaction between the game test user and the game's initial dynamic content.

[0113] Step S145: Extract the operation frequency distribution of the character operation trajectory in the content interaction imprint, and adjust the interval of character actions and the number of character dialogues in the character behavior flow according to the statistical characteristics of the operation frequency distribution to obtain the evolved character behavior flow.

[0114] The operation frequency distribution of character operation trajectories is extracted from content interaction imprints, that is, the number of times different operation types occur per unit time. The statistical characteristics of this frequency distribution are analyzed, such as the occurrence patterns of high-frequency operation types, and the peaks and troughs of operation frequency. Based on these characteristics, the character behavior flow is adjusted: for character actions corresponding to high-frequency operations, the interval time is shortened; for character actions corresponding to low-frequency operations, the interval time is lengthened. Simultaneously, based on the overall operation frequency, the amount of character dialogue is adjusted: when the operation frequency is high, the amount of dialogue is appropriately reduced to avoid affecting operations; when the operation frequency is low, the amount of dialogue is appropriately increased to enrich content. After these adjustments, the evolved character behavior flow is obtained.

[0115] Step S146: Extract the dwell time distribution of the scene dwell trajectory in the content interaction imprint, and adjust the interval of scene terrain transformation and the duration of scene weather transformation in the scene change flow according to the statistical characteristics of the dwell time distribution itself to obtain the evolution scene change flow.

[0116] Extracting the dwell time distribution of scene dwell trajectories from content interaction imprints involves statistically analyzing the distribution of user dwell time in different scene locations. Analyzing the statistical characteristics of this distribution, such as which areas users spend more time in, which areas spend less time in, and the central trend of dwell time, allows for adjustments to the scene change flow. For areas where users spend more time, the interval between scene terrain transitions is extended to allow users more time to experience the scene; for areas where users spend less time, the transition interval is shortened. Simultaneously, based on the overall distribution of dwell time, the duration of scene weather transitions is adjusted, with weather transition durations appropriately extended for areas with longer dwell times and shortened for areas with shorter dwell times. This adjustment yields the evolved scene change flow.

[0117] Step S147: Extract the execution step distribution of the task execution trajectory in the content interaction imprint, and adjust the order of task condition triggering and the steps of task stage transition in the task triggering flow according to the statistical characteristics of the execution step distribution itself, so as to obtain the evolved task triggering flow.

[0118] Extracting the execution step distribution of the task execution trajectory from content interaction imprints involves statistically analyzing the frequency and order of each step during task execution. Analyzing the statistical characteristics of this distribution, such as which steps are executed frequently and the smoothness of transitions between steps, allows for adjustments to the task trigger flow. This involves advancing the trigger order of frequently executed tasks and appropriately delaying those with lower execution frequencies. Simultaneously, optimizing the steps for task stage transitions makes them more aligned with user execution habits, reducing unnecessary steps, resulting in an evolved task trigger flow.

[0119] Step S148: Integrate the evolved character behavior flow, the evolved scene change flow, and the evolved task trigger flow to generate dynamic game evolution content.

[0120] The game integrates the evolution of character behavior flow, the evolution of scene change flow, and the evolution of task trigger flow. During integration, the timeline is used as a benchmark to ensure that the three flows remain consistent in time, so that character behavior, scene changes, and task triggers can cooperate with each other to form dynamic content of game evolution.

[0121] Step S150: Based on the game's dynamic evolution content, calculate the coverage of the content evolution factors, generate a content evolution completeness index, and output the game dynamic content package when the content evolution completeness index meets the content iteration frequency requirements in the demand feature spectrum.

[0122] Based on the dynamic content of game evolution, the coverage of content evolution factors is calculated, which measures the degree to which these factors are reflected in the dynamic content. A content evolution completeness index is generated based on this coverage, reflecting the degree to which the dynamic content meets the initial requirements. When this index reaches the standard corresponding to the content iteration frequency requirement in the requirement feature spectrum, the dynamic content is packaged and output to form a dynamic game content package.

[0123] Step S151: Extract the evolutionary character behavior flow, evolutionary scene change flow, and evolutionary task trigger flow from the game's evolutionary dynamic content.

[0124] The game's dynamic evolution content is broken down into evolutionary character behavior flow, evolutionary scene change flow, and evolutionary task trigger flow, so that they can be analyzed and calculated separately.

[0125] Step S152: Analyze the evolved character behavior flow, count the number of character action types and the number of character dialogue topics contained therein, and calculate the ratio of the number of character action types and the number of character dialogue topics to the number of preset action types and the number of preset dialogue topics corresponding to the character element adaptation in the content evolution factor, respectively, as the character evolution coverage.

[0126] A detailed analysis of the evolving character behavior flow was conducted, counting the number of different character action types and the number of different topics involved in character dialogues. The number of preset action types and preset dialogue topics corresponding to character element adaptability were obtained from the content evolution factors; these two preset numbers were set based on initial requirements. The ratio of the statistically obtained number of character action types to the number of preset action types, and the ratio of the number of character dialogue topics to the number of preset dialogue topics, were calculated separately. These two ratios were then averaged, and the result was taken as the character evolution coverage.

[0127] Step S153: Analyze the evolution scene change flow, count the number of scene terrain types and scene weather types contained therein, and calculate the ratio of the number of scene terrain types and scene weather types to the preset number of terrain types and preset number of weather types corresponding to the scene element adaptation degree in the content evolution factor, respectively, as the scene evolution coverage.

[0128] Analyze the evolutionary scene change flow, and count the number of different scene terrain types and different scene weather types. Extract the preset number of terrain types and preset number of weather types corresponding to scene element adaptability from the content evolution factors. These two preset numbers are determined based on the initial demand feature spectrum. Calculate the ratio of the statistically obtained scene terrain type number to the preset terrain type number, and the ratio of the scene weather type number to the preset weather type number, respectively. Average these two ratios, and the result is taken as the scene evolution coverage.

[0129] Step S154: Analyze the evolution task trigger flow, count the number of task stage types and task reward types contained therein, and calculate the ratio of the number of task stage types and task reward types to the preset number of stage types and preset number of reward types corresponding to the task element adaptation degree in the content evolution factor, respectively, as the task evolution coverage.

[0130] The evolution task trigger flow is analyzed to count the number of different task stage types, such as collection stage, combat stage, and puzzle-solving stage, as well as the number of different task reward types, such as equipment rewards, experience rewards, and item rewards. The number of preset stage types and preset reward types corresponding to the task element adaptation degree is obtained from the content evolution factors; these two preset numbers are also set based on the initial demand feature spectrum. The ratio of the statistically obtained number of task stage types to the preset number of stage types, and the ratio of the number of task reward types to the preset number of reward types are calculated separately. These two ratios are averaged, and the result is used as the task evolution coverage.

[0131] Step S155: Calculate the average value of the character evolution coverage, the scene evolution coverage, and the task evolution coverage as the overall coverage of the content evolution factor.

[0132] The previously obtained character evolution coverage, scene evolution coverage, and task evolution coverage are summarized, and the arithmetic mean of these three coverages is calculated. Specifically, the values ​​of the three coverages are added together and then divided by three. The result is the overall coverage of the content evolution factor, which comprehensively reflects the coverage of the content evolution factor in terms of characters, scenes, and tasks.

[0133] Step S156: The overall coverage is correlated with the content iteration frequency requirement in the demand feature spectrum to generate a content evolution completeness index. The value of the content evolution completeness index is positively correlated with the overall coverage and negatively correlated with the iteration cycle of the content iteration frequency requirement.

[0134] Content iteration frequency requirements are extracted from the demand feature spectrum. These requirements include iteration cycle information for different content types, such as the update cycle of scene terrain and the update cycle of tasks. The overall coverage is correlated with these iteration cycles by dividing the overall coverage by the combined value of the iteration cycles. The combined value of the iteration cycles is obtained by weighted averaging of the iteration cycles of each content type, with the weights set according to the importance of each content type in the game. Since a higher overall coverage results in a higher content evolution completeness index, while a longer iteration cycle results in a lower index, this index is positively correlated with overall coverage and negatively correlated with iteration cycle. The final value generated is the content evolution completeness index.

[0135] For example, the method further includes:

[0136] Step S157: Extract the content iteration frequency requirement from the demand feature spectrum, and determine the completeness threshold corresponding to the content iteration frequency requirement. The completeness threshold is related to the iteration cycle of the content iteration frequency requirement. The shorter the iteration cycle, the lower the completeness threshold.

[0137] Specific information regarding content iteration frequency requirements is extracted from the demand feature spectrum, including the iteration cycle of each content item. Based on these iteration cycles, corresponding completeness thresholds are determined. The rule is that a shorter iteration cycle indicates a need for faster content updates, resulting in a relatively lower completeness requirement and thus a lower completeness threshold; conversely, a longer iteration cycle indicates a higher completeness requirement and a higher completeness threshold. For example, content with a one-month iteration cycle has a lower completeness threshold than content with a three-month iteration cycle. The final completeness threshold is determined using this method.

[0138] Step S158: Compare the content evolution completeness index with the completeness threshold. If the content evolution completeness index is greater than or equal to the completeness threshold, extract the evolutionary character behavior flow, evolutionary scene change flow, and evolutionary task trigger flow from the game evolution dynamic content.

[0139] The calculated content evolution completeness index is compared with the determined completeness threshold. When the content evolution completeness index is greater than or equal to the completeness threshold, it indicates that the current game evolution dynamic content has met the completeness standard corresponding to the content iteration frequency requirement. At this point, the evolution character behavior flow, evolution scene change flow, and evolution task trigger flow are extracted from the game evolution dynamic content to prepare for subsequent packaging processing.

[0140] Step S159: Perform timeline synchronization processing on the evolved character behavior flow, evolved scene change flow, and evolved task trigger flow, and add content identification information to the synchronized evolved character behavior flow, evolved scene change flow, and evolved task trigger flow. The content identification information includes generation time, demand feature spectrum identifier, and content evolution factor identifier.

[0141] The extracted evolutionary character behavior flow, evolutionary scene change flow, and evolutionary task trigger flow are synchronized along a timeline to ensure accurate temporal correspondence and avoid content inconsistencies caused by time misalignment. After synchronization, content identification information is added to these three flows. The generation time refers to the final determination time of the game's dynamic evolutionary content; the demand feature spectrum identifier is a unique code for the demand feature spectrum corresponding to the content; and the content evolution factor identifier is a unique code for the corresponding content evolution factor. Adding this identification information facilitates content management, traceability, and version control.

[0142] Step S160: Integrate and package the evolved character behavior flow, evolved scene change flow, and evolved task trigger flow after adding the tags according to the preset format to generate a game dynamic content package.

[0143] Following the preset format established during game development, the evolved character behavior flow, evolved scene change flow, and evolved task trigger flow, all with added identification information, are integrated and packaged. The preset format specifies the storage structure and data compression methods for each flow within the package; for example, the three flows are stored in different subdirectories within the package and compressed using a specific algorithm to reduce storage space. After integration and packaging, a complete game dynamic content package is generated, which can be directly used for game updates or releases.

[0144] Step S161: If the content evolution completeness index is less than the completeness threshold, return to the step of collecting content interaction imprints, re-collect content interaction imprints and re-evolve the game evolution dynamic content until the content evolution completeness index meets the completeness threshold, and then execute the above packaging step to output the game dynamic content package.

[0145] When the content evolution completeness index is less than the completeness threshold, it indicates that the current game evolution dynamic content has not yet met the required completeness standard and needs further optimization. At this point, the process returns to the step of collecting content interaction imprints, i.e., the collection phase in step S140, where game test users are invited to interact again to collect new content interaction imprints. Based on the new content interaction imprints and content evolution factors, the game evolution dynamic content undergoes further evolution processing, repeating the relevant processes in steps S140 to S150 until the calculated content evolution completeness index is greater than or equal to the completeness threshold. Afterward, the packaging step is executed according to steps S158 to S160, finally outputting the game dynamic content package.

[0146] The method further includes step S210: pre-training a game content self-evolution AI model, wherein the game content self-evolution AI model includes a feature parsing layer, a character content generation layer, a scene content generation layer, and a task content generation layer.

[0147] Before applying the game content self-evolution AI model, it needs to be pre-trained. The structure of this game content self-evolution AI model includes a feature parsing layer, a character content generation layer, a scene content generation layer, and a task content generation layer. The feature parsing layer is responsible for parsing the input feature spectrum and extracting feature components related to the character, scene, and task; the character content generation layer is specifically used to generate character behavior flows based on relevant features; the scene content generation layer is used to generate scene change flows; and the task content generation layer is used to generate task trigger flows.

[0148] Step S211: Collect historical game development data, which includes historical demand feature spectrum, historical content evolution factors and corresponding historical game dynamic content.

[0149] Collect a large amount of historical data accumulated during past game development. This data includes historical demand feature spectra, which are descriptions and quantified feature spectra of various demand features during past game development; historical content evolution factors, which are the adaptability of basic game elements selected based on historical demand feature spectra and related sub-spectrum identifiers; and corresponding historical game dynamic content, which are the final game content generated based on historical demand feature spectra and historical content evolution factors. The above historical data needs to cover different types and styles of games to ensure the comprehensiveness of training.

[0150] Step S212: Preprocess the historical game development data, including data cleaning, data standardization, and data partitioning.

[0151] The collected historical game development data undergoes preprocessing. Data cleaning removes noise and outliers, such as deleting illogical demand feature spectra or content evolution factors. Data standardization transforms feature values ​​of different magnitudes into a uniform range, enabling the model to learn data patterns more effectively. Data partitioning divides the preprocessed data into training, validation, and test sets. The training set is used for model parameter learning, the validation set is used to adjust hyperparameters, and the test set is used to evaluate the model's final performance. The partitioning ratio can be set according to the data volume, for example, 70% for the training set, 15% for the validation set, and 15% for the test set.

[0152] Step S213: Construct the network structure of the game content self-evolution AI model, set the initial parameters of each layer of the network structure, the feature parsing layer adopts a convolutional neural network structure, and the character content generation layer, scene content generation layer and task content generation layer all adopt a recurrent neural network structure.

[0153] The specific network structure for constructing the game content self-evolution AI model is as follows: The feature parsing layer adopts a convolutional neural network (CNN) structure, which can effectively extract local and global features from the input feature spectrum, and gradually refine key features through multiple convolutional and pooling operations. The character content generation layer, scene content generation layer, and task content generation layer all adopt a recurrent neural network (RNN) structure because these layers need to process sequential data and generate temporally continuous character behavior flows, scene change flows, and task trigger flows. RNNs can effectively capture the temporal dependencies in sequential data. Initial parameters for each layer are set, including convolutional kernel size, number of convolutional layers, hidden layer dimension of the RNN, and learning rate. These initial parameters can be set based on empirical values ​​or relevant literature.

[0154] Step S214: Input the training set into the constructed game content self-evolution AI model, use historical game dynamic content as the expected output, and adjust the model parameters through the backpropagation algorithm to minimize the error between the game content self-evolution AI model output and the expected output.

[0155] The pre-defined training set is input into the constructed game content self-evolution AI model. The model processes the historical demand feature spectrum and historical content evolution factors of the input to generate corresponding output content. The generated output content is compared with the historical dynamic game content (i.e., the expected output) in the training set, and the error between the two is calculated. The backpropagation algorithm is used to propagate the error from the output layer to the input layer. The parameters of each layer of the model are adjusted according to the error, such as the convolution kernel weights of the convolutional neural network and the hidden layer weights of the recurrent neural network. Through multiple iterations of training, the parameters are continuously adjusted so that the error between the model output and the expected output gradually decreases and reaches the preset error threshold.

[0156] Step S215: After each preset training round, the model is validated using the validation set. Based on the validation results, the hyperparameters of the game content self-evolution AI model are adjusted. The hyperparameters include the learning rate, hidden layer dimension, and number of iterations.

[0157] During model training, after each preset training epoch, such as every 10 epochs, the model is validated using a validation set. The historical demand feature spectrum and historical content evolution factors from the validation set are input into the model in the current training state to obtain the model's validation output. The error between the validation output and the historical game dynamic content in the validation set is calculated. The model's hyperparameters are adjusted based on changes in the validation error. If the validation error continues to decrease, the current learning rate can be maintained or appropriately increased; if the validation error stops decreasing or begins to increase, the learning rate is decreased. Simultaneously, hyperparameters such as the hidden layer dimension and the number of iterations are adjusted based on the validation results to improve the model's generalization ability.

[0158] Step S216: When the game content self-evolution AI model training reaches the preset number of iterations or the error reaches the preset error threshold, the test set is used to evaluate the performance of the game content self-evolution AI model. The evaluation indicators include the accuracy, smoothness and diversity of the generated content. If the evaluation results meet the preset evaluation criteria, the training of the game content self-evolution AI model is completed.

[0159] The training process stops when the AI ​​model for game content self-evolution reaches its preset maximum number of iterations, or when the model's error on the training set reaches a preset error threshold. At this point, a test set is used to perform a final performance evaluation of the model. Evaluation metrics include: accuracy of generated content (the degree to which the generated content matches historical game dynamic content in the test set); fluency (the temporal and logical coherence of generated character behavior flows, scene change flows, and task trigger flows); and diversity (the model's ability to generate content of different styles and types). If all these evaluation metrics meet the preset evaluation criteria, such as an accuracy rate of 85% or higher, and fluency and diversity scores of 4 or higher (out of 5), the model is considered trained and ready for use in subsequent game dynamic content generation. If not, the model structure or training parameters need to be readjusted, and the training and evaluation repeated until the criteria are met.

[0160] The method may further include:

[0161] Step S310: Maintain and update the game development resource pool, including adding new basic game elements, deleting outdated basic game elements, and updating the attribute information of basic game elements.

[0162] To ensure that the elements in the game development resource pool meet the ever-changing needs of game development, they require regular maintenance and updates. Adding new basic game elements involves adding newly designed character models, scene modules, quest templates, etc., to the resource pool and improving their attribute information. Deleting outdated basic game elements means removing elements that no longer conform to the current game style, technical standards, or player needs, such as old, low-precision character models or quest templates that don't fit the new storyline. Updating the attribute information of basic game elements involves adjusting the attributes of existing elements according to new requirements of game development, such as modifying character skill parameters or adjusting scene visual effect parameters, ensuring that the elements in the resource pool always remain effective and applicable.

[0163] Step S311: During the process of collecting the content interaction imprints of game test users, the user privacy data involved is encrypted. The user's operation trajectory data, dwell trajectory data and execution trajectory data are encrypted and stored using a symmetric encryption algorithm.

[0164] When collecting user interaction data for game testing, it may involve data such as user operation trajectories, dwell times, and execution trajectories. This data may contain private information such as user habits. To protect user privacy and prevent data leakage, this data needs to be encrypted. A symmetric encryption algorithm is used, meaning the same key is used for both encryption and decryption. All types of user trajectory data are encrypted before storage. During encryption, secure key management is ensured, and a key rotation mechanism is used to periodically change the key, further improving data security. Simultaneously, encrypted data requires authorized verification before decryption, restricting access permissions and preventing unauthorized personnel from obtaining user privacy data.

[0165] Step S312: Regularly manage the version of the generated game dynamic content package, record the update content, update time and corresponding demand feature spectrum of each version, so as to facilitate subsequent content traceability and version rollback.

[0166] Regular version management is implemented for the generated dynamic game content packages, assigning a unique version number to each package. Detailed information for each version is recorded, including updated content (modifications and additions to character behavior flow, scene change flow, and task trigger flow compared to the previous version); update time (the time the content package was generated and released); and the corresponding demand feature spectrum (the identifier of the demand feature spectrum upon which the version's content is based). By establishing version management archives, content changes for each version can be clearly traced. When issues arise with new version content, a quick rollback to a previous stable version can be implemented, ensuring normal game operation and a positive user experience.

[0167] Figure 2 The illustration shows exemplary hardware and software components of an AI-based game development dynamic content generation system 100 that can implement the ideas of this application, according to some embodiments of this application. For example, a processor 120 can be used in the AI-based game development dynamic content generation system 100 and to perform the functions described in this application.

[0168] The AI-based game development dynamic content generation system 100 can be a general-purpose server or a special-purpose server; both can be used to implement the AI-based game development dynamic content generation method of this application. Although only one server is shown in this application, for convenience, the functions described in this application can be implemented in a distributed manner on multiple similar platforms to balance the load.

[0169] For example, an AI-based game development dynamic content generation system 100 may include a network port 110 connected to a network, one or more processors 120 for executing program instructions, a communication bus 130, and various forms of storage media 140, such as a disk, ROM, or RAM, or any combination thereof. Exemplarily, the AI-based game development dynamic content generation system 100 may also include program instructions stored in ROM, RAM, or other types of non-transitory storage media, or any combination thereof. The methods of this application can be implemented according to these program instructions. The AI-based game development dynamic content generation system 100 also includes an I / O interface 150 between the computer and other input / output devices.

[0170] For ease of explanation, only one processor is described in the AI-based game development dynamic content generation system 100. However, it should be noted that the AI-based game development dynamic content generation system 100 of this application may also include multiple processors, and therefore the steps performed by one processor as described in this application may also be performed jointly or individually by multiple processors. For example, if the processor of the AI-based game development dynamic content generation system 100 performs steps A and B, it should be understood that steps A and B may also be performed jointly by two different processors or individually by one processor. For example, the first processor performs step A, the second processor performs step B, or the first processor and the second processor jointly perform steps A and B.

[0171] Furthermore, this embodiment of the invention also provides a readable storage medium, wherein computer-executable instructions are preset in the readable storage medium, and when the processor executes the computer-executable instructions, the above-mentioned method for generating dynamic content for game development based on artificial intelligence is implemented.

[0172] It should be noted that, in order to simplify the description of the present invention and thus help to understand one or more embodiments of the invention, multiple features may sometimes be grouped into one embodiment, drawing or description thereof in the foregoing description of the embodiments of the present invention.

Claims

1. A method for generating dynamic content for game development based on artificial intelligence, characterized in that, The method includes: Obtain the dynamic content generation requirements corresponding to game development, extract the content style orientation, interaction depth requirements and content iteration frequency requirements from the dynamic content generation requirements, and form a requirement feature spectrum; Based on the aforementioned demand feature spectrum, suitable basic game elements are selected from the game development resource pool, the compatibility degree between each basic game element and the aforementioned demand feature spectrum is calculated, and a content evolution factor is generated. The basic game elements include game character elements, game scene elements, and game task elements. The pre-trained game content self-evolution AI model is invoked, and the demand feature spectrum and the content evolution factor are input to generate the initial dynamic content of the game. The initial dynamic content of the game includes character behavior flow, scene change flow and task trigger flow. Collect content interaction imprints during the interaction process between game test users and the initial dynamic content of the game, and combine them with the content evolution factors to perform evolution processing on the initial dynamic content of the game to generate game evolution dynamic content. The content interaction imprints include character operation trajectory, scene stay trajectory and task execution trajectory. Based on the game's dynamic evolution content, the coverage of the content evolution factors is calculated, and a content evolution completeness index is generated. When the content evolution completeness index meets the content iteration frequency requirements in the demand feature spectrum, a game dynamic content package is output. The process of selecting suitable basic game elements from the game development resource pool based on the demand feature spectrum, calculating the fit degree between each basic game element and the demand feature spectrum, and generating content evolution factors includes: Extract all candidate game elements from the game development resource pool, including candidate character elements, candidate scene elements, and candidate task elements; Extract the character style attributes, character interaction attributes, and character iteration attributes of the candidate character elements to form a character element feature vector; Extract the scene style attributes, scene interaction attributes, and scene iteration attributes of the candidate scene elements to form scene element feature vectors; Extract the task style attributes, task interaction attributes, and task iteration attributes of the candidate task elements to form a task element feature vector; Extract feature components related to the character elements from the demand feature spectrum to generate a character demand sub-spectrum, and calculate the cosine similarity between the character element feature vector and the character demand sub-spectrum as the character element fit. Extract feature components related to scene elements from the demand feature spectrum to generate scene demand sub-spectrum, and calculate the cosine similarity between the scene element feature vector and the scene demand sub-spectrum as the scene element fit degree; Extract feature components related to task elements from the demand feature spectrum to generate a task demand sub-spectrum, and calculate the cosine similarity between the task element feature vector and the task demand sub-spectrum as the task element fit. Candidate character elements, candidate scene elements, and candidate task elements that meet the character adaptation threshold, scene element adaptation threshold, and task element adaptation threshold are selected as basic game elements. The character element fit, scene element fit, and task element fit are combined in the order of character, scene, and task to generate a content evolution factor, and the identification information of the character requirement sub-spectrum, scene requirement sub-spectrum, and task requirement sub-spectrum are associated and recorded in the content evolution factor. The step of calculating the coverage of the content evolution factors based on the game's dynamic evolution content and generating a content evolution completeness index includes: Extract the evolutionary character behavior flow, evolutionary scene change flow, and evolutionary task trigger flow from the game's dynamic evolution content; Analyze the evolutionary character behavior flow, count the number of character action types and the number of character dialogue topics contained therein, and calculate the ratio of the number of character action types and the number of character dialogue topics to the preset number of action types and the preset number of dialogue topics corresponding to the character element adaptation in the content evolution factor, respectively, as the character evolution coverage. Analyze the evolution scene change flow, count the number of scene terrain types and scene weather types contained therein, and calculate the ratio of the number of scene terrain types and scene weather types to the preset number of terrain types and preset number of weather types corresponding to the scene element adaptation degree in the content evolution factor, respectively, as the scene evolution coverage. Analyze the evolution task trigger flow, count the number of task stage types and task reward types contained therein, and calculate the ratio of the number of task stage types and task reward types to the preset number of stage types and preset number of reward types corresponding to the task element adaptation degree in the content evolution factor, respectively, as the task evolution coverage. Calculate the average of the character evolution coverage, the scene evolution coverage, and the task evolution coverage as the overall coverage of the content evolution factor; The overall coverage is correlated with the content iteration frequency requirement in the demand feature spectrum to generate a content evolution completeness index. The value of the content evolution completeness index is positively correlated with the overall coverage and negatively correlated with the iteration cycle of the content iteration frequency requirement.

2. The method for generating dynamic content for game development based on artificial intelligence according to claim 1, characterized in that, The process involves obtaining the dynamic content generation requirements corresponding to game development, extracting the content style orientation, interaction depth requirements, and content iteration frequency requirements from these requirements, and forming a requirement feature spectrum, including: Analyze the dynamic content generation requirements corresponding to game development, extract the text information describing the content presentation style from the dynamic content generation requirements, and divide the content style dimensions, which include visual style dimension, narrative style dimension and interaction style dimension; Extract feature descriptions under each content style dimension from the text information, convert each feature description into quantifiable style feature values, and form a content style guide; Extract the text information describing the degree of interaction between the user and the game content from the dynamic content generation requirements, and divide the interaction depth dimensions, which include the operation interaction dimension, the decision interaction dimension, and the emotional interaction dimension. The interaction requirements under each interaction depth dimension are extracted from the text information, and each interaction requirement is converted into a quantifiable interaction feature value to form the interaction depth requirements. Extract the text information describing the game content update cycle from the dynamic content generation requirements, divide the content iteration dimensions, extract the cycle requirement items under each content iteration dimension from the text information, and convert each cycle requirement item into a quantifiable iteration feature value to form the content iteration frequency requirement. The content style guidance, the interaction depth requirement, and the content iteration frequency requirement are concatenated in a preset dimension order to generate a demand feature spectrum. The number of dimensions of the demand feature spectrum is consistent with the sum of the number of dimensions of the content style dimension, the interaction depth dimension, and the content iteration dimension.

3. The method for generating dynamic content for game development based on artificial intelligence according to claim 1, characterized in that, The process of calling a pre-trained game content self-evolution AI model, inputting the demand feature spectrum and the content evolution factor, and generating initial dynamic game content includes: The demand feature spectrum is input into the feature parsing layer of the game content self-evolution AI model to extract the feature components related to character behavior in the demand feature spectrum and obtain the character behavior guidance features. Extract the feature components related to scene changes from the demand feature spectrum to obtain scene change-oriented features; Extract the feature components related to task triggering from the demand feature spectrum to obtain task triggering guidance features; The character behavior guidance features and the character element adaptation in the content evolution factor are input into the character content generation layer of the game content self-evolution AI model to construct character behavior logic and generate a character behavior flow containing continuous character actions and character dialogues. The scene change guidance feature and the scene element adaptation in the content evolution factor are input into the scene content generation layer of the game content self-evolution AI model to construct scene change logic and generate a scene change flow that includes continuous scene terrain transformation and scene weather transformation. The task triggering guidance feature and the task element adaptation in the content evolution factor are input into the task content generation layer of the game content self-evolution AI model to construct the task triggering logic and generate a task triggering flow that includes continuous task condition triggering and task stage transition. The character behavior flow, the scene change flow, and the task trigger flow are aligned on the timeline to generate the initial dynamic content of the game, which includes the character behavior flow, the scene change flow, and the task trigger flow.

4. The method for generating dynamic content for game development based on artificial intelligence according to claim 1, characterized in that, The process of collecting interaction imprints between test users and the initial dynamic content of the game, and combining these with the content evolution factors to perform evolutionary processing on the initial dynamic content of the game, generates game evolution dynamic content, including: The initial dynamic content of the game is loaded into the game runtime environment, the interaction imprint collection program is started, and the character control operations of the game test user when interacting with the initial dynamic content of the game are recorded to form a continuous sequence of character operation coordinates as the character operation trajectory. Record the sequence of scene position coordinates when the game test user moves in the game scene to form a continuous sequence of scene dwell coordinates, which serves as the scene dwell trajectory; Record the sequence of task operation steps performed by game test users to form a continuous sequence of task execution steps, which serves as the task execution trajectory. Integrate the character's operation trajectory, the scene's dwell trajectory, and the task's execution trajectory to form a content interaction imprint; Extract the operation frequency distribution of the character operation trajectory in the content interaction imprint, and adjust the interval of character actions and the number of character dialogues in the character behavior flow according to the statistical characteristics of the operation frequency distribution itself to obtain the evolved character behavior flow. Extract the dwell time distribution of the scene dwell trajectory in the content interaction imprint, and adjust the interval of scene terrain transformation and the duration of scene weather transformation in the scene change flow according to the statistical characteristics of the dwell time distribution itself to obtain the evolution scene change flow; Extract the execution step distribution of the task execution trajectory in the content interaction imprint, and adjust the order of task condition triggering and the steps of task stage transition in the task triggering flow according to the statistical characteristics of the execution step distribution itself to obtain the evolved task triggering flow. The evolutionary character behavior flow, the evolutionary scene change flow, and the evolutionary task trigger flow are integrated to generate dynamic game evolution content.

5. The method for generating dynamic content for game development based on artificial intelligence according to claim 4, characterized in that, The recording of the game test user's character control operations during interaction with the game's initial dynamic content forms a continuous sequence of character operation coordinates, serving as the character operation trajectory, including: A character operation acquisition node is set up in the game runtime environment, and the character operation acquisition node runs synchronously with the character behavior flow in the initial dynamic content of the game. When a game test user performs character control operations, the character operation acquisition node records the trigger time and corresponding character operation type for each operation; Based on the trigger time, each character operation type is arranged in chronological order to form a character operation time sequence; Each character operation type is assigned a corresponding operation coordinate value, which is related to the character's movement direction and movement range in the game scene; Replace each type of character operation in the character operation time sequence with the corresponding operation coordinate value to form a coordinate sequence containing timestamps and operation coordinate values; The coordinate sequence is made continuous by filling in the blank coordinate values ​​between adjacent timestamps, so that the coordinate sequence remains continuous in the time dimension. The continuous coordinate sequence is marked as the character operation trajectory, which includes continuous timestamps and corresponding operation coordinate values.

6. The method for generating dynamic content for game development based on artificial intelligence according to claim 4, characterized in that, The sequence of scene position coordinates recorded when the test user moves within the game scene forms a continuous sequence of scene dwell coordinates, which serves as the scene dwell trajectory, including: The game scene is divided into multiple scene location areas, and each scene location area corresponds to a unique area coordinate identifier; When the game test user moves in the game scene, the current position of the user-controlled character is tracked in real time to determine the scene location area to which the current position belongs and the corresponding area coordinates. Record the entry time of the user-controlled character into each scene location area and the exit time of the scene location area; Calculate the difference between the departure time and the entry time for each scene location area, and use this as the dwell time for that scene location area; Arrange the area coordinates, entry time, exit time, and duration of stay for each scene location area in chronological order of entry time to form a scene location record sequence. The scene location recording sequence is subjected to time continuity verification. The scene location recording sequence that passes the verification is converted into a sequence containing continuous area coordinates and corresponding dwell time, forming a scene dwell trajectory.

7. The method for generating dynamic content for game development based on artificial intelligence according to claim 4, characterized in that, The sequence of task operation steps recorded when the game test user performs a task forms a continuous sequence of task execution steps, which serves as the task execution trajectory, including: Analyze the task trigger flow in the initial dynamic content of the game to determine all task stages and the corresponding task operation requirements for each task stage. Assign a unique stage identifier to each task stage, and assign a unique operation identifier to each task operation requirement. When game test users perform tasks, the system monitors their actions in real time to determine whether each action meets the task requirements of the current task stage. If the operation meets the task operation requirements, then record the operation identifier, operation time point and task stage identifier corresponding to the operation. Arrange all eligible operation identifiers, operation time points, and task stage identifiers in chronological order of operation time points to form a task operation record sequence. Check whether the task operation record sequence covers all task stage identifiers. If there are any uncovered task stage identifiers, supplement the default operation record corresponding to the corresponding task stage identifier. The supplemented task operation record sequence is converted into a continuous task execution step sequence. Each step in the task execution step sequence includes an operation identifier, an operation time point, and a task stage identifier, forming a task execution trajectory.

8. A game development dynamic content generation system based on artificial intelligence, characterized in that, The AI-based game development dynamic content generation system includes a processor and a memory, the memory and the processor being connected. The memory is used to store programs, instructions or code, and the processor is used to execute the programs, instructions or code in the memory to implement the AI-based game development dynamic content generation method according to any one of claims 1-7.

Images