Method and system for generating virtual character interactive content based on behavior analysis

Abstract

The present invention discloses a method and system for generating virtual character interactive content based on behavior analysis, and belongs to the field of artificial intelligence technology. The present invention first realizes real-time capture and understanding of user behavior by efficiently acquiring and converting user interaction behavior logs into data sets that can be processed by AI models. Secondly, multimodal behavior semantic mining technology is used to deeply explore the potential needs behind user behavior, so that the response of virtual characters is more accurate and personalized. Further, the behavioral response features are decoded based on the mined potential demand features, and a rich and diverse virtual character response behavior data set is generated, which greatly enriches the interactive content. Finally, these response contents are seamlessly converted back to the visual interface state, so that users can instantly feel the intelligent feedback of the virtual character. In this way, not only the timeliness of user interaction is improved and delays are reduced, but also the fluency of interaction and personalized experience are greatly enhanced.

Description

Technical Field

[0001] The present invention belongs to the field of artificial intelligence technology, and in particular relates to a method and system for generating virtual character interactive content based on behavior analysis. Background Art

[0002] In traditional games and virtual interactive environments, the interaction between users and virtual characters is often limited to preset scripts and fixed response modes, resulting in a monotonous and lack of personalization in the interactive experience. At the same time, due to the complexity and diversity of user interactive behaviors, how to accurately capture and understand these behaviors in real time and generate appropriate virtual character responses has become a technical problem that needs to be solved urgently. In addition, existing interactive systems often have a large delay between processing user behaviors and generating responses, which seriously affects the smoothness of the interaction and the user's immersion. Therefore, there is an urgent need for a technical solution that can efficiently capture user interactive behaviors, deeply mine behavioral semantics, and instantly generate personalized virtual character responses to improve the experience and efficiency of user interaction. Summary of the invention

[0003] The present invention provides a method and system for generating virtual character interactive content based on behavior analysis, which can solve or partially solve the technical problems involved in the above-mentioned background technology.

[0004] An embodiment of the present invention provides a method for generating virtual character interaction content based on behavior analysis, which is applied to an electronic device. The method includes:

[0005] Obtain the user interaction behavior log to be interactively responded to, transform the user interaction behavior log from the visual interface state to the AI ​​model matching state, and obtain the corresponding interaction behavior data set to be mined;

[0006] Performing multimodal behavior semantic mining on a target interactive behavior data subset in the interactive behavior data set to be mined to obtain a target multimodal behavior semantic vector, and obtaining potential expected demand features corresponding to the target multimodal behavior semantic vector;

[0007] Decoding the target interactive behavior data subset with behavioral response characteristics according to the potential expected demand characteristics, and obtaining a virtual character response behavior data set through personalized behavioral response characteristics obtained through the behavioral response feature decoding;

[0008] The virtual character response behavior data set is transformed from an AI model matching state to a visual interface state to obtain virtual character interaction content corresponding to the user interaction behavior log to be interactively responded to.

[0009] In one implementation, the step of transforming the user interaction behavior log from a visual interface state to an AI model matching state to obtain a corresponding interaction behavior data set to be mined includes:

[0010] Obtaining an interface scanning core, and scanning the interface scanning core in the user interaction behavior log according to a set scanning period;

[0011] For each interface area scanned by the interface scanning core, determining a data format conversion label at the corresponding interface area through the interface scanning core and the behavior log set contained in the interface scanning core in the user interaction behavior log;

[0012] Based on the data format conversion labels at each interface area scanned by the interface scanning core, the interactive behavior dataset to be mined is transformed.

[0013] In one implementation, the step of performing multimodal behavior semantic mining on a target interactive behavior data subset in the interactive behavior data set to be mined to obtain a target multimodal behavior semantic vector includes:

[0014] Obtaining user behavior thermal characteristics of each interactive behavior event in the target VR interactive environment in the interactive behavior data set to be mined;

[0015] Through the various behavior semantic mining branches in the pre-configured behavior semantic mining model, multimodal behavior semantic mining is performed on the user behavior thermal characteristics of each interactive behavior event in the target VR interactive environment to obtain the target multimodal behavior semantic vector of the target interactive behavior data subset in each behavior semantic mining branch.

[0016] In one implementation, the configuration of the behavior semantic mining model includes:

[0017] For any one of the predefined multiple behavior operation modes, a corresponding behavior semantic mining branch is determined according to a behavior operation preference feature corresponding to the behavior operation mode, wherein the behavior operation preference feature includes at least one of an action tendency feature, an emotion tendency feature or an expression tendency feature;

[0018] Based on the behavior semantic mining branches corresponding to each behavior operation mode, a behavior semantic mining model is configured.

[0019] In one implementation, the multimodal behavior semantic mining is performed on the user behavior thermal characteristics of each interactive behavior event in the target VR interactive environment through each behavior semantic mining branch in the pre-configured behavior semantic mining model to obtain the target multimodal behavior semantic vector of the target interactive behavior data subset in each behavior semantic mining branch, including:

[0020] For each behavior semantic mining branch, a target interactive behavior event in the interactive behavior event relationship tree corresponding to the current behavior semantic mining branch is retrieved from the interactive behavior events contained in the target VR interactive environment; the current behavior semantic mining branch is any behavior semantic mining branch;

[0021] Retrieve event element information corresponding to the target interactive behavior event in the current behavior semantic mining branch, and perform event element recognition on the user behavior thermal characteristics of the target interactive behavior event based on the event element information corresponding to the target interactive behavior event;

[0022] Based on the event element recognition results corresponding to the target interactive behavior event, the target multimodal behavior semantic vector of the target interactive behavior data subset in the current behavior semantic mining branch is determined.

[0023] In one implementation, there are multiple target multimodal behavior semantic vectors, and obtaining potential expected demand features corresponding to the target multimodal behavior semantic vectors includes:

[0024] A plurality of target multimodal behavior semantic vectors are input into a debugged interactive response prediction algorithm, and the plurality of target multimodal behavior semantic vectors are processed respectively by the debugged interactive response prediction algorithm to generate potential expected demand features corresponding to each target multimodal behavior semantic vector.

[0025] In one implementation, the debugging method of the interactive response prediction algorithm includes:

[0026] Obtaining an algorithm learning example set, wherein each algorithm learning example in the algorithm learning example set includes an interactive behavior log adversarial example and an interactive behavior log authentication example;

[0027] Transform the interactive behavior log confrontation sample from the visualization interface state to the AI ​​model matching state to obtain the interactive behavior confrontation data set to be mined, and transform the interactive behavior log authentication sample from the visualization interface state to the AI ​​model matching state to obtain the interactive behavior authentication data set to be mined;

[0028] Perform multimodal behavior semantic mining on the target interactive behavior data subset in the interactive behavior adversarial dataset to be mined to obtain a multimodal behavior perturbation semantic vector, and perform multimodal behavior semantic mining on the target interactive behavior data subset in the interactive behavior authentication dataset to be mined to obtain a multimodal behavior authentication semantic vector;

[0029] Determining a first potential expected demand feature through the multimodal behavior disturbance semantic vector and the multimodal behavior authentication semantic vector;

[0030] Determine the second potential expected demand feature through the multimodal behavior disturbance semantic vector and the interactive response prediction algorithm to be debugged;

[0031] Determining an algorithm learning error through the first potential expected demand feature and the second potential expected demand feature;

[0032] The algorithm parameters of the interactive response prediction algorithm to be debugged are improved by using the algorithm learning error until the algorithm training meets the standard requirements, thereby obtaining a debugged interactive response prediction algorithm.

[0033] In one implementation, the multimodal behavior disturbance semantic vector is obtained by mining each behavior semantic mining branch in the behavior semantic mining model, and the multimodal behavior authentication semantic vector is obtained by mining each behavior semantic mining branch in the behavior semantic mining model;

[0034] The determining of the first potential expected demand feature by using the multimodal behavior disturbance semantic vector and the multimodal behavior authentication semantic vector includes:

[0035] For any behavior semantic mining branch, combine the multimodal behavior perturbation semantic vector and the multimodal behavior authentication semantic vector mined by the behavior semantic mining branch to determine the semantic comparison result corresponding to the behavior semantic mining branch;

[0036] Based on the semantic comparison results corresponding to each behavior semantic mining branch, the first potential expected demand feature is determined.

[0037] In one implementation, the step of decoding the behavior response characteristics of the target interaction behavior data subset according to the potential expected demand characteristics includes:

[0038] Determine the potential expected demand features corresponding to each interactive behavior event in the target VR interactive environment in the interactive behavior data set to be mined by using the potential expected demand features corresponding to the target multimodal behavior semantic vector;

[0039] Based on the potential expected demand characteristics corresponding to each interactive behavior event, the user behavior thermal characteristics of each interactive behavior event in the target VR interactive environment in the interactive behavior data set to be mined are updated to obtain the personalized behavior response characteristics obtained by decoding the behavior response characteristics.

[0040] In one implementation, the method of using the potential expected demand features corresponding to the target multimodal behavior semantic vector to determine the potential expected demand features corresponding to each interactive behavior event in the target VR interactive environment in the interactive behavior data set to be mined includes:

[0041] For each interactive behavior event in the target VR interactive environment in the interactive behavior data set to be mined, determining the target behavior operation mode of the current interactive behavior event; the current interactive behavior event is any interactive behavior event in the target VR interactive environment;

[0042] Acquire the behavior operation preference characteristics of the target behavior operation mode and the behavior operation preference characteristics of the past behavior operation modes of the target behavior operation mode; the behavior operation preference characteristics include action tendency characteristics and emotion tendency characteristics;

[0043] When the current interactive behavior event is between the action tendency feature and the emotion tendency feature of the target behavior operation mode, determining the potential expected demand feature of the current interactive behavior event through the behavior operation preference feature of the past behavior operation mode and the potential expected demand feature of the target multimodal behavior semantic vector corresponding to the past behavior operation mode;

[0044] When the current interactive behavior event matches the target behavior operation mode and the emotional tendency characteristics, the potential expected demand characteristics of the current interactive behavior event are determined based on the potential expected demand characteristics of the target multimodal behavior semantic vector corresponding to the target behavior operation mode.

[0045] In one implementation, obtaining a virtual character response behavior dataset by decoding the personalized behavior response feature includes:

[0046] Acquire user behavior thermal characteristics of the remaining interactive behavior events except the target VR interactive environment from the interactive behavior data set to be mined;

[0047] Based on the user behavior thermal characteristics of the remaining interactive behavior events except the target VR interactive environment and the personalized behavior response characteristics decoded from the behavior response characteristics, a virtual character response behavior dataset is integrated.

[0048] An embodiment of the present invention provides an electronic device, comprising at least one processor and a memory; the memory stores computer-executable instructions; the at least one processor executes the computer-executable instructions stored in the memory, so that the at least one processor executes the above method.

[0049] An embodiment of the present invention provides a readable storage medium, on which a program or instruction is stored. When the program or instruction is executed by a processor, the steps of the above method are implemented.

[0050] In an embodiment of the present invention, firstly, by efficiently acquiring and converting the user interaction behavior log into a data set that can be processed by the AI ​​model, real-time capture and understanding of user behavior is achieved. Secondly, multimodal behavior semantic mining technology is used to deeply explore the potential needs behind user behavior, so that the response of the virtual character is more accurate and personalized. Further, the behavioral response characteristics are decoded based on the mined potential demand characteristics, and a rich and diverse virtual character response behavior data set is generated, which greatly enriches the interactive content. Finally, these response contents are seamlessly converted back to the visual interface state, so that the user can instantly feel the intelligent feedback of the virtual character. In this way, not only the timeliness of user interaction is improved and delays are reduced, but also the smoothness of interaction and personalized experience are greatly enhanced. BRIEF DESCRIPTION OF THE DRAWINGS

[0051] Figure 1 The present invention provides a flowchart of a method for generating virtual character interactive content based on behavior analysis according to an embodiment of the present invention.

[0052] Figure 2 The present invention is a schematic diagram of the structure of an electronic device provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0053] Figure 1 A method for generating virtual character interactive content based on behavior analysis is shown, which is applied to an electronic device. The method includes the following steps 110 to 140.

[0054] Step 110: Obtain the user interaction behavior log to be interactively responded to, transform the user interaction behavior log from a visual interface state to an AI model matching state, and obtain a corresponding interaction behavior data set to be mined.

[0055] Step 120: Perform multimodal behavior semantic mining on the target interactive behavior data subset in the interactive behavior data set to be mined to obtain a target multimodal behavior semantic vector, and obtain potential expected demand features corresponding to the target multimodal behavior semantic vector.

[0056] Step 130: Decode the behavior response characteristics of the target interaction behavior data subset according to the potential expected demand characteristics, and obtain a virtual character response behavior data set through the personalized behavior response characteristics obtained by decoding the behavior response characteristics.

[0057] Step 140: transform the virtual character response behavior data set from the AI ​​model matching state to the visual interface state, and obtain the virtual character interaction content corresponding to the user interaction behavior log to be interactively responded.

[0058] In conjunction with the above steps 110 to 140, in an exemplary electronic game application scenario, there is a role-playing game (RPG) called "XYXY", the game world is vast and complex, full of various NPCs (non-player controlled characters) and rich and diverse interactive elements. In this scenario, the electronic device (such as a game server or a personal computer) is the executor of the technical solution, responsible for processing the interactive behavior between the user (i.e., the game player) and the NPC (virtual character) to enhance the realism and immersion of the game experience. The following is a detailed introduction on how to achieve this goal through specific steps.

[0059] In the "XYXY" game, every operation of the player, such as clicking, moving, dialogue selection, etc., will be recorded as a user interaction behavior log. The user interaction behavior log first exists in a visual interface state, that is, the interface elements and their changes that the player sees and operates on the screen. For subsequent analysis and processing, electronic devices need to convert these log data from the visual interface state to a matching state that the AI ​​model can understand and process. The conversion process involves converting the interactive behaviors on the interface (such as clicking buttons, talking to NPCs, etc.) into a structured data format, including timestamps, interaction types, interactive objects and other information, to form an interactive behavior data set to be mined.

[0060] Once the structured interactive behavior datasets are obtained, the next task is to conduct in-depth analysis on a specific subset of these datasets - the target interactive behavior data subset. The target interactive behavior data subset can contain all the interaction records between the player and a specific NPC, or the player's behavior records when performing a specific task. Multimodal behavior semantic mining refers to the use of advanced machine learning algorithms, such as deep learning models, to identify and understand the complex patterns and meanings behind these behaviors. For example, by analyzing the text content of the player's conversation with the NPC, the context in which the conversation occurred (such as location, time), and the player's behavior sequence (such as what was done first and what was done later), the model can extract the multimodal behavior semantic vector hidden in these behaviors. The multimodal behavior semantic vector not only reflects the player's surface behavior, but also reveals the player's potential expected demand characteristics. For example, if the player frequently asks a certain NPC questions about task clues, this may indicate that the player's current main need is to obtain task guidance. By identifying such potential needs, electronic devices can respond to players' behaviors more intelligently and provide more personalized and considerate services.

[0061] After identifying the potential expected demand characteristics of the player, the electronic device needs to further convert these characteristics into specific virtual character response behaviors. This process involves decoding the behavioral response characteristics, that is, using the AI ​​model to generate personalized response strategies based on the player's demand characteristics. For example, if the player's potential demand is to seek task guidance, the NPC's response may include providing detailed task instructions, marking the location of the task location on the map, or giving key tips for completing the task. Through the decoding process, the electronic device is able to generate a virtual character response behavior dataset that contains a variety of possible responses to the player's specific behavior. These responses are not only based on the player's immediate behavior, but also take into account the player's historical behavior patterns, the current game progress, and the NPC's personality settings and background stories to ensure that the response is consistent with the game logic and can bring players a real and interesting interactive experience.

[0062] Finally, the electronic device needs to convert the virtual character response behavior dataset from the matching state of the AI ​​model back to the visual interface state so that the player can see and experience the NPC's response in the game interface. The conversion process includes rendering the AI-generated response content (such as text, actions, expressions, etc.) to the game interface to ensure that they are seamlessly integrated with other elements of the game (such as the environment, character animations). For example, when the NPC provides task instructions to the player, the relevant dialogue text will appear in the dialog box, and the NPC may make corresponding gestures or facial expressions to enhance the realism of the communication.

[0063] In this way, the "XYXY" game can provide players with a more intelligent and personalized NPC interaction experience. Players no longer simply exchange information with NPCs, but can feel that NPCs have their own thoughts and emotions, and can respond appropriately according to the player's behavior and needs. This not only enhances the immersion and fun of the game, but also makes the game world more vivid and real.

[0064] It is worth mentioning that in the "XYXY" game, every interactive behavior of the user is recorded in real time and efficiently as a user interactive behavior log. The user interactive behavior log is quickly converted from the visual interface state to the matching state that the AI ​​model can understand and process. This rapid conversion process ensures the immediacy of the data, laying a solid foundation for subsequent analysis and processing, thereby greatly improving the timeliness of the response and reducing latency. Once the structured interactive behavior data set is obtained, the electronic device immediately performs multimodal behavioral semantic mining on the target interactive behavior data subset. Using advanced machine learning algorithms, such as deep learning models, the electronic device can identify and understand the complex patterns and meanings behind these behaviors in real time. This real-time processing capability ensures that the player's potential expected demand characteristics can be quickly identified, further improving the response speed and reducing latency. After identifying the player's potential expected demand characteristics, the electronic device immediately converts these characteristics into specific virtual character response behaviors. Through the real-time decoding process, the electronic device can quickly generate a virtual character response behavior data set containing personalized response strategies. This instant generation capability ensures that the NPC's response can keep up with the player's behavior, further improving the timeliness and fluency of the interaction. Finally, the electronic device quickly converts the virtual character response behavior data set from the matching state of the AI ​​model back to the visual interface state. This fast conversion process ensures that the NPC's response content can be instantly presented on the game interface, allowing players to immediately experience the NPC's feedback. This fast visual presentation not only enhances the immersion and fun of the game, but also makes the interaction process smoother and more efficient. It can be seen that through the execution process of steps 110 to 140, combined with the real-time processing capabilities of the neural network algorithm, the electronic device can greatly improve the response timeliness and reduce latency. This allows players to enjoy a smoother and more immediate gaming experience when interacting with NPCs.

[0065] In the embodiment of the present invention, step 110 is the core invention point of the embodiment of the present invention, and its technical solution mainly revolves around how to efficiently obtain and process the user interaction behavior logs so as to provide a basis for subsequent AI model analysis and application. Specifically, this step first involves capturing and recording the interaction behavior logs generated by users during the game. These logs cover all operations of users on the visual interface, such as clicks, slides, inputs, etc. They exist in the original visual interface state, that is, the interface elements and their changes that players see and operate on the screen. In order to convert these log data into a form that the AI ​​model can understand and process, the system needs to perform a state transformation process. The transformation first maps the interaction behavior on the visual interface into a series of structured events or actions, each of which contains detailed information, such as the timestamp of occurrence, the type of interaction, the object of interaction, and the specific content of the interaction. With this information, the system can build a comprehensive and structured interactive behavior data set, which provides a solid foundation for subsequent multimodal behavior semantic mining and potential demand feature extraction. In the process of state transformation, the system also needs to clean and preprocess the log data to remove noise data, fill in missing values, and ensure that the format and type of the data match the input requirements of the AI ​​model. This step is crucial to improving the accuracy and efficiency of subsequent analysis. It can be seen that step 110 obtains an interactive behavior dataset to be mined by capturing and recording the user's interactive behavior logs on the visual interface and converting these logs from the original visual state to a structured state that can be matched by the AI ​​model. This step not only provides high-quality data input for subsequent AI model analysis, but also lays a solid foundation for achieving a more intelligent and personalized user interaction experience.

[0066] Furthermore, steps 120 to 130 focus on deeply mining the semantic information behind the user's interactive behavior and generating personalized virtual character responses accordingly.

[0067] First, in step 120, the electronic device further processes the interactive behavior data set to be mined obtained in the previous step. In particular, the electronic device will focus on those data subsets related to the target interactive behavior, which may represent the user's behavior patterns in specific scenarios, specific tasks, or specific NPC interactions. In order to deeply understand the meaning behind these behaviors, the electronic device will use multimodal behavior semantic mining technology. This technology combines a variety of machine learning algorithms, especially deep learning models, to identify and analyze complex semantic information in behavioral patterns. Through this process, the electronic device can extract the target multimodal behavior semantic vector from the target interactive behavior data subset. These vectors not only reflect the user's surface behavior, but also reveal the user's potential expected demand characteristics, such as the user's current goals, emotional state, or preferences.

[0068] Secondly, step 130 is performed on the basis of step 120. Once the electronic device obtains the user's potential expected demand characteristics, it will use this information to decode the behavioral response characteristics of the target interactive behavior data subset. This process involves converting the user's potential needs into specific, personalized virtual character response strategies. The electronic device will generate a virtual character response behavior dataset containing various possible responses based on the user's characteristics, historical behavior patterns, current game progress, and the NPC's personality settings and background stories. This dataset not only contains text responses, but may also contain actions, expressions, or other types of interactive elements to ensure the diversity and richness of the responses.

[0069] It can be seen that the technical solution of steps 120 to 130 greatly improves the interactive experience of the game and the user's sense of immersion by deeply mining the semantic information behind the user's interactive behavior and generating personalized virtual character responses accordingly.

[0070] Furthermore, step 140, as one of the core inventive points of the embodiment of the present invention, is dedicated to converting the virtual character response behavior data set generated by the AI ​​model into interactive content that the player can directly experience in the game interface.

[0071] During the execution of step 140, the electronic device's primary task is to convert the virtual character response behavior dataset from its current AI model matching state. After being processed by the AI ​​model, the dataset already contains personalized response strategies for user interaction behaviors, but these responses still exist in the form of data structures within the model. In order for players to see and feel these responses in the game interface, the electronic device needs to render them into visual interface elements.

[0072] In order to achieve this conversion, the electronic device will use a series of technical means. First, it will parse each record in the virtual character response behavior data set and extract the interactive elements such as text, actions, and expressions. Then, the electronic device will render these elements one by one to the corresponding positions of the game interface according to the game's interface design and rendering rules. For example, if the response contains a dialogue text, the electronic device will display it in the dialog box with appropriate background and animation effects. If the response contains a character's action, the electronic device will call the corresponding animation resources to let the character make the corresponding action in the game.

[0073] During the rendering process, the electronic device will also consider a series of factors to ensure the realism and smoothness of the interactive content. For example, it will adjust the display size and position of the interactive content according to the player's perspective and distance. It will also add appropriate light and shadow processing to the interactive content based on the lighting and shadow effects of the game. In addition, the electronic device will also update and manage the interactive content in real time to ensure that they can be seamlessly connected with the player's other game behaviors.

[0074] In summary, the embodiments of the present invention firstly realize the real-time capture and understanding of user behavior by efficiently acquiring and converting the user interaction behavior log into a data set that can be processed by the AI ​​model. Secondly, the multimodal behavior semantic mining technology is used to deeply explore the potential needs behind the user behavior, so that the response of the virtual character is more accurate and personalized. The behavioral response characteristics are further decoded based on the mined potential demand characteristics, and a rich and diverse virtual character response behavior data set is generated, which greatly enriches the interactive content. Finally, these response contents are seamlessly converted back to the visual interface state, so that the user can instantly feel the intelligent feedback of the virtual character. In this way, not only the timeliness of user interaction is improved and delays are reduced, but also the smoothness of interaction and personalized experience are greatly enhanced.

[0075] In an optional embodiment, the user interaction behavior log is transformed from a visual interface state to an AI model matching state to obtain a corresponding interaction behavior data set to be mined, including: obtaining an interface scanning core, and scanning the interface scanning core in the user interaction behavior log according to a set scanning cycle; for each interface area scanned by the interface scanning core, determining a data format conversion label at the corresponding interface area through the interface scanning core and the behavior log set contained in the interface scanning core in the user interaction behavior log; based on the data format conversion labels at each interface area scanned by the interface scanning core, transforming to obtain the interaction behavior data set to be mined.

[0076] In this embodiment, the focus is on effectively transforming the user interaction behavior log from the visual interface state to the AI ​​model matching state, and then obtaining the corresponding interactive behavior data set to be mined. This process involves multiple key links, aiming to ensure the accurate conversion and efficient use of data, laying a solid foundation for subsequent AI model training and behavior analysis.

[0077] First, you need to obtain a core component - the interface scanning core. The interface scanning core plays a vital role. It is equivalent to an intelligent detector that can shuttle through the user interaction behavior log, identify and extract valuable information. Set the scanning cycle according to actual needs. The cycle can be flexibly adjusted according to the log generation speed, the processing power of the electronic device, and the real-time requirements of the required data. The interface scanning core will perform regular scans in the user interaction behavior log according to the set cycle to ensure that no behavior log that may contain important interaction information is missed.

[0078] During the scanning process of the interface scanning core, it will conduct a detailed analysis of each scanned interface area. For each scanned interface area, the interface scanning core will combine the behavior log set contained in the area in the user interaction behavior log to conduct in-depth information mining and analysis. The key to this step is to determine the data format conversion label at the corresponding interface area. The data format conversion label is a kind of metadata that describes how to convert the original behavior log data into a format suitable for AI model processing. These labels may include the type of data (such as text, numbers, timestamps, etc.), the structure of the data (such as lists, dictionaries, etc.), and the conversion rules of the data (such as deduplication, normalization, etc.).

[0079] After the data format conversion labels are determined, the user interaction behavior logs in each interface area scanned by the interface scanner can be format converted and data cleaned based on these labels. This process is an important part of data preprocessing, which ensures the consistency and accuracy of the data so that the converted data set can better meet the needs of the AI ​​model. Specifically, according to the instructions of the data format conversion labels, the original behavior log data will be converted and cleaned as necessary, such as converting text data into numerical data, removing duplicate or invalid records, and filling missing values.

[0080] After completing the above steps, we will get a preprocessed, unified format, and high-quality interactive behavior dataset to be mined. This dataset not only retains the key information in the original user interactive behavior log, but also achieves data structuring and normalization through the guidance of data format conversion labels, greatly improving the availability and value of the data. This interactive behavior dataset to be mined will be used as the input of the AI ​​model for training and optimizing the model to achieve in-depth analysis and accurate prediction of user interactive behavior.

[0081] In this way, firstly, through the periodic scanning of the interface scanning core, the user interaction behavior log can be captured and processed in real time to ensure the timeliness and integrity of the data. Secondly, the introduction of data format conversion tags makes the data preprocessing process more efficient and accurate, avoiding tedious manual data cleaning and format adjustment. Finally, through this series of processing flows, a high-quality interactive behavior dataset to be mined is obtained, which provides strong data support for subsequent AI model training and behavior analysis.

[0082] It can be seen that this embodiment achieves the goal of effectively transforming the user interaction behavior log from the visual interface state to the AI ​​model matching state through a series of steps such as obtaining the interface scanning core, setting the scanning cycle for log scanning, determining the data format conversion label, and performing data preprocessing and format conversion based on these labels. In this process, the intelligent scanning capability of the interface scanning core and the guiding role of the data format conversion label are fully utilized to ensure the accurate conversion and efficient use of the data. The final interactive behavior data set to be mined is not only high-quality and structured, but also provides a solid foundation and strong support for subsequent AI model training and behavior analysis.

[0083] In an alternative embodiment, the performing multimodal behavior semantic mining on a target interactive behavior data subset in the interactive behavior data set to be mined to obtain a target multimodal behavior semantic vector includes: obtaining user behavior thermal features of each interactive behavior event in the interactive behavior data set to be mined in a target VR interactive environment; performing multimodal behavior semantic mining on the user behavior thermal features of each interactive behavior event in the target VR interactive environment through each behavioral semantic mining branch in a pre-configured behavioral semantic mining model, and obtaining a target multimodal behavior semantic vector of the target interactive behavior data subset in each behavioral semantic mining branch.

[0084] It can be understood that this embodiment will deeply explore how to perform multimodal behavior semantic mining on interactive behavior events specific to the target VR interactive environment in the interactive behavior data set to be mined, and extract the target multimodal behavior semantic vector from it. This process involves a detailed analysis of user behavior characteristics and a clever application of the behavior semantic mining model, aiming to reveal the deep semantic information hidden behind the complex interactive behavior.

[0085] First, it is necessary to obtain the user behavior thermal features of the interactive behavior events that occur in the target VR interactive environment in the interactive behavior dataset to be mined. User behavior thermal features are a quantitative indicator that reflects the intensity and activity of users' behavior when performing various operations in the VR interactive environment. These features may include the user's movement frequency in a specific area, the number of operations on a specific object, the duration of gaze retention, etc. By extracting these thermal features, it is possible to preliminarily outline the user's behavior profile in the VR interactive environment, providing basic data for subsequent multimodal behavior semantic mining.

[0086] Next, we will use the pre-configured behavior semantic mining model to conduct in-depth analysis and mining of these user behavior thermal characteristics. The behavior semantic mining model is a complex machine learning model that contains multiple behavior semantic mining branches, each of which focuses on mining a specific type of behavior semantic information. These branches may include spatial behavior semantic mining branches, temporal behavior semantic mining branches, object interaction behavior semantic mining branches, etc., which are used to mine users' spatial movement patterns, temporal behavior patterns, and interaction methods with virtual objects in the VR interactive environment.

[0087] In the process of multimodal behavior semantic mining, the user behavior thermal features of the interactive behavior events to be mined are input into the various branches of the behavior semantic mining model. Each branch will conduct in-depth analysis and processing of these thermal features according to its specific mining goals and algorithms, thereby extracting the behavioral semantic information hidden in them. For example, the spatial behavior semantic mining branch may analyze the user's movement path and stop points in the VR space, extracting the user's spatial exploration mode and preferred area; the temporal behavior semantic mining branch may focus on the user's behavior changes in different time periods, revealing the user's behavioral rhythm and habits.

[0088] After multimodal behavior semantic mining, the target multimodal behavior semantic vectors of the target interactive behavior data subset in each behavior semantic mining branch will be obtained. These vectors are numerical representations of the mining results. They contain rich behavioral semantic information and can accurately describe the various interactive behaviors of users in the VR interactive environment. For example, a spatial behavior semantic vector may contain information on dimensions such as the user's movement speed, direction change, and residence time in the VR space; while a temporal behavior semantic vector may reflect the user's activity level and behavior type distribution in different time periods of the day.

[0089] These target multimodal behavior semantic vectors are extremely valuable. They not only provide a new perspective for in-depth understanding of user interaction behavior, but also provide strong data support for subsequent AI model training, user behavior prediction, personalized response and other applications. For example, in VR game development, these vectors can be used to optimize game design and create a gaming experience that is more in line with user behavior habits and preferences; in VR education applications, personalized learning plans can be customized based on students' learning behavior semantic vectors to improve teaching effectiveness.

[0090] On the one hand, the structure and parameters of the behavior semantic mining model can be adjusted to adapt to different VR interaction environments and application scenarios; on the other hand, new behavior semantic mining branches can be added as needed to mine more types of behavior semantic information. This flexibility and scalability enables this technical solution to be widely used in various VR interaction behavior analysis tasks and continue to bring new discoveries and insights.

[0091] In this way, by obtaining the thermal characteristics of user behavior in the target VR interactive environment in the interactive behavior data set to be mined, and using the pre-configured behavioral semantic mining model to perform multimodal behavioral semantic mining, we finally obtained the target multimodal behavior semantic vector of the target interactive behavior data subset in each behavioral semantic mining branch. In this process, we not only deeply mined the deep semantic information behind the user interactive behavior, but also provided strong data support for a variety of subsequent applications.

[0092] In the next step, the configuration method of the behavior semantic mining model includes: for any behavior operation mode among the pre-defined multiple behavior operation modes, determining the corresponding behavior semantic mining branch according to the behavior operation preference characteristics corresponding to the behavior operation mode, the behavior operation preference characteristics include: at least one of action tendency characteristics, emotion tendency characteristics or expression tendency characteristics; based on the behavior semantic mining branches corresponding to each behavior operation mode, configuring the behavior semantic mining model.

[0093] In another possible embodiment, the method performs multimodal behavior semantic mining on the user behavior thermal characteristics of each interactive behavior event in the target VR interactive environment through each behavior semantic mining branch in a pre-configured behavior semantic mining model to obtain a target multimodal behavior semantic vector of the target interactive behavior data subset in each behavior semantic mining branch, including: for each behavior semantic mining branch, retrieving the target interactive behavior event in the interactive behavior event relationship tree corresponding to the current behavior semantic mining branch from the interactive behavior events contained in the target VR interactive environment; the current behavior semantic mining branch is any behavior semantic mining branch; retrieving event element information corresponding to the target interactive behavior event in the current behavior semantic mining branch, and based on the event element information corresponding to the target interactive behavior event, performing event element recognition on the user behavior thermal characteristics of the target interactive behavior event; and determining the target multimodal behavior semantic vector of the target interactive behavior data subset in the current behavior semantic mining branch based on the event element recognition result corresponding to the target interactive behavior event.

[0094] The above-mentioned embodiments involve the configuration method of the behavior semantic mining model, and how to use the model to perform multimodal behavior semantic mining on interactive behavior events in the target VR interactive environment to obtain the target multimodal behavior semantic vector of the target interactive behavior data subset in each behavior semantic mining branch.

[0095] First, regarding the configuration of the behavior semantic mining model, this is a delicate and strategic process. For multiple pre-defined behavior operation modes, they are analyzed and configured one by one. For each behavior operation mode, the corresponding behavior semantic mining branch will be determined based on its unique behavior operation preference characteristics. These behavior operation preference characteristics may include at least one of action tendency characteristics, emotion tendency characteristics, or expression tendency characteristics, which together constitute the comprehensive preference representation of users when performing specific behavior operations.

[0096] Specifically, if a certain behavior operation mode is mainly related to the user's hand movements, then when configuring the corresponding behavior semantic mining branch, special attention will be paid to the user's action tendency characteristics, such as the hand movement trajectory, speed, strength, etc. If a certain behavior operation mode is more related to the user's emotional state, then when configuring it, attention will be paid to extracting the user's emotional tendency characteristics, such as judging the user's emotional state through clues such as facial expressions and voice intonation. Similarly, if the behavior operation mode is closely related to the user's expression changes, then attention will be paid to the user's expression tendency characteristics, such as facial muscle movements, changes in eyes, etc.

[0097] Based on this configuration strategy, a dedicated behavior semantic mining branch can be configured for each behavior operation mode. These branches are responsible for mining and extracting semantic information related to the user's specific behavior operation. Finally, integrating all these branches together constitutes a complete behavior semantic mining model.

[0098] Then, we will see how to use the configured behavior semantic mining model to conduct multimodal behavior semantic mining on interactive behavior events in the target VR interactive environment. This process is also full of challenges and opportunities.

[0099] For each behavior semantic mining branch in the model, the corresponding interactive behavior events will be retrieved from the target VR interactive environment. These events form an interactive behavior event relationship tree, which clearly depicts the user's behavior trajectory and interaction mode in the VR environment. In this relationship tree, those interactive behavior events that are closely related to the target behavior semantic mining branch can be found, which are the target interactive behavior events to be mined.

[0100] Next, in the current behavior semantic mining branch, these target interactive behavior events will be deeply analyzed. The event element information corresponding to these events will be retrieved. This information is the basic unit of interactive behavior events, which may include user behavior actions, behavior objects, behavior time, behavior space, etc. Based on this event element information, the user behavior thermal characteristics of the target interactive behavior events can be accurately identified and analyzed.

[0101] Finally, based on the event element recognition results corresponding to the target interactive behavior event, the target multimodal behavior semantic vector of the target interactive behavior data subset in the current behavior semantic mining branch can be determined. This vector is a high-dimensional numerical representation that contains rich behavioral semantic information and can accurately describe the user's interactive behavior characteristics and preferences in a specific behavior operation mode.

[0102] In this way, by finely configuring the behavior semantic mining model and skillfully using the model to conduct multimodal behavior semantic mining of interactive behavior events in the target VR interactive environment, we finally obtained the target multimodal behavior semantic vector rich in behavior semantic information. This process not only demonstrates the profound background and innovative ability in the field of behavior analysis, but also provides strong data support and algorithm guarantee for subsequent AI model training, user behavior prediction, personalized response and other applications.

[0103] In other possible embodiments, there are multiple target multimodal behavior semantic vectors, and obtaining potential expected demand features corresponding to the target multimodal behavior semantic vectors includes: inputting the multiple target multimodal behavior semantic vectors into a debugged interaction response prediction algorithm, and processing the multiple target multimodal behavior semantic vectors respectively through the debugged interaction response prediction algorithm to generate potential expected demand features corresponding to each target multimodal behavior semantic vector.

[0104] In detail, the debugging method of the interactive response prediction algorithm includes: obtaining an algorithm learning example set, each algorithm learning example in the algorithm learning example set includes an interactive behavior log confrontation sample and an interactive behavior log authentication sample; transforming the interactive behavior log confrontation sample from a visual interface state to an AI model matching state to obtain an interactive behavior confrontation data set to be mined, and transforming the interactive behavior log authentication sample from a visual interface state to an AI model matching state to obtain an interactive behavior authentication data set to be mined; performing multimodal behavior semantic mining on a target interactive behavior data subset in the interactive behavior confrontation data set to be mined to obtain a multimodal behavior perturbation semantic vector, and Multimodal behavior semantic mining is performed on the target interactive behavior data subset in the behavior authentication data set to obtain a multimodal behavior authentication semantic vector; the first potential expected demand feature is determined through the multimodal behavior perturbation semantic vector and the multimodal behavior authentication semantic vector; the second potential expected demand feature is determined through the multimodal behavior perturbation semantic vector and the interactive response prediction algorithm to be debugged; the algorithm learning error is determined through the first potential expected demand feature and the second potential expected demand feature; the algorithm parameters of the interactive response prediction algorithm to be debugged are improved through the algorithm learning error until the algorithm training meets the standard requirements, and the debugged interactive response prediction algorithm is obtained.

[0105] It can be understood that the realization of obtaining potential expected demand features corresponding to multiple target multimodal behavior semantic vectors depends on a debugged interactive response prediction algorithm that can cleverly process these multimodal behavior semantic vectors and generate corresponding potential expected demand features.

[0106] First of all, it should be clear that these target multimodal behavior semantic vectors do not exist in isolation. They are mined from various interactive behavior events of users in the VR interactive environment. Each vector contains user-specific behavioral semantic information. The goal is to predict the user's potential expected needs based on this information, so as to provide strong support for subsequent personalized responses, interaction optimization and other applications.

[0107] To achieve this goal, multiple target multimodal behavior semantic vectors are input into the debugged interactive response prediction algorithm. The algorithm is like a sophisticated decoder, which can deeply analyze and process each input behavior semantic vector and extract feature information related to the user's potential expected needs. These feature information may include user preferences, intentions, habits, etc., which together constitute a comprehensive representation of the user's potential expected needs.

[0108] Furthermore, the debugging of the above-mentioned interactive response prediction algorithm involves a complex process of multiple steps and delicate operations.

[0109] First, you need to obtain an algorithm learning example set. This example set contains a large number of algorithm learning examples, each of which consists of adversarial examples containing interactive behavior logs and authenticated examples of interactive behavior logs. These examples are valuable data resources for algorithm debugging, and they can help train an accurate and efficient interactive response prediction algorithm.

[0110] Next, these interactive behavior log adversarial samples and interactive behavior log authentication samples are transformed from the visual interface state to the AI ​​model matching state. This is a data preprocessing process, which enables these original data samples to be converted into a format that the AI ​​model can understand and process. Through such processing, the interactive behavior adversarial dataset to be mined and the interactive behavior authentication dataset to be mined are obtained.

[0111] Then, multimodal behavior semantic mining is performed on the target interaction behavior data subset in the interaction behavior adversarial dataset to be mined, and a multimodal behavior perturbation semantic vector is obtained. Similarly, multimodal behavior semantic mining is performed on the target interaction behavior data subset in the interaction behavior authentication dataset to be mined, and a multimodal behavior authentication semantic vector is obtained. These two processes are key steps in obtaining user behavior semantic information, which provide rich user behavior characteristics and preference information.

[0112] With these multimodal behavior perturbation semantic vectors and multimodal behavior authentication semantic vectors, we can start debugging the interactive response prediction algorithm. First, the first potential expected demand feature is determined through these two vector sets. This feature is obtained based on the user's real behavior data, and it reflects the user's potential expected demand under specific interactive behaviors.

[0113] Then, only through the multimodal behavior perturbation semantic vector, the interactive response prediction algorithm to be debugged is used to determine the second potential expected demand feature. This feature is obtained based on the algorithm prediction, and it represents the algorithm's prediction result of the user's potential expected demand.

[0114] Next, by comparing these two potential expected demand characteristics, the algorithm learning error is determined. This error reflects the difference between the algorithm prediction result and the actual result, which is an important basis for algorithm improvement.

[0115] Finally, the algorithm parameters of the interactive response prediction algorithm to be debugged are improved through the algorithm learning error. This is an iterative optimization process, which continuously adjusts the algorithm parameters and configuration until the algorithm learning error is reduced to meet the algorithm training requirements. Finally, a debugged interactive response prediction algorithm is obtained, which can accurately predict the user's potential expected demand characteristics.

[0116] In this way, through the steps of fine data preprocessing, multimodal behavior semantic mining, interactive response prediction algorithm debugging, etc., the potential expected demand characteristics corresponding to multiple target multimodal behavior semantic vectors were successfully obtained. This process not only demonstrates the profound background and innovative ability in the field of behavior analysis, but also provides strong data support and algorithm guarantee for subsequent personalized response, interactive optimization and other applications.

[0117] Under some other possible design ideas, the multimodal behavior perturbation semantic vector is obtained by mining each behavior semantic mining branch in the behavior semantic mining model, and the multimodal behavior authentication semantic vector is obtained by mining each behavior semantic mining branch in the behavior semantic mining model; the first potential expected demand feature is determined by the multimodal behavior perturbation semantic vector and the multimodal behavior authentication semantic vector, including: for any behavior semantic mining branch, combining the multimodal behavior perturbation semantic vector and the multimodal behavior authentication semantic vector mined by the targeted behavior semantic mining branch, determining the semantic comparison result corresponding to the targeted behavior semantic mining branch; based on the semantic comparison results corresponding to each behavior semantic mining branch, determining the first potential expected demand feature.

[0118] The design idea further involves how to mine multimodal behavior disturbance semantic vectors and multimodal behavior authentication semantic vectors through various behavior semantic mining branches in the behavior semantic mining model in the role-playing game (RPG) "XYXY", and determine the player's first potential expected demand characteristics based on these two types of vectors, thereby optimizing the interaction experience between the player and NPC (non-player controlled character).

[0119] First of all, it should be made clear that the multimodal behavior perturbation semantic vector and the multimodal behavior authentication semantic vector are both mined through the various behavior semantic mining branches in the behavior semantic mining model. These branches are like a pair of sharp eyes, which can deeply analyze the various behaviors of players in the game and extract vectors containing rich semantic information. These vectors not only reflect the behavioral characteristics of players, but also reveal the potential needs and intentions of players. In the "XYXY" game, every click, move, dialogue choice and other operations of the player will be captured by these behavior semantic mining branches and converted into multimodal behavior perturbation semantic vectors and multimodal behavior authentication semantic vectors.

[0120] Next, let's look at how to use these two types of vectors to determine the first potential expected demand characteristics of players. This is a complex and challenging process, but there is an effective method to deal with it. For any behavior semantic mining branch, the multimodal behavior perturbation semantic vector and multimodal behavior authentication semantic vector mined from the branch will be combined for in-depth analysis and comparison.

[0121] For example, in the "XYXY" game, there is a behavioral semantic mining branch that is specifically responsible for mining the behavioral semantics of players talking to NPCs. When players talk to NPCs, this branch mines multimodal behavioral perturbation semantic vectors and multimodal behavioral authentication semantic vectors. By comparing these two vectors, we can find the behavioral characteristics and potential needs of players during conversations. If players often ask questions about task clues during conversations, then we can infer that the player's current main need is to obtain task guidance.

[0122] Of course, the work does not stop here. Because the behaviors of players are diverse, it is necessary to consider the potential needs of players under all behavioral modes. Therefore, a comprehensive analysis and judgment will be made based on the semantic comparison results corresponding to each behavioral semantic mining branch. In the specific implementation, various challenges and difficulties will be encountered. For example, how to ensure the accuracy and reliability of multimodal behavior perturbation semantic vectors and multimodal behavior authentication semantic vectors; how to conduct effective semantic comparisons to ensure the accuracy and effectiveness of the comparison results; how to integrate the semantic comparison results of each behavioral semantic mining branch to determine the accurate first potential expected demand characteristics; these issues all require in-depth thinking and research.

[0123] In order to solve these problems, a series of effective measures were taken. First, the behavior semantic mining model was designed and debugged to ensure that it can accurately mine multimodal behavior perturbation semantic vectors and multimodal behavior authentication semantic vectors. Advanced machine learning algorithms and deep learning technologies were used, and the model was extensively trained and optimized to ensure that it can accurately identify various behaviors of players and extract semantic information from them.

[0124] Secondly, an advanced semantic comparison algorithm is used to effectively compare and analyze the multimodal behavior disturbance semantic vector and the multimodal behavior authentication semantic vector. Algorithms such as vector space model and cosine similarity are used to calculate the similarity and difference between the two vectors. Through such comparison, the behavioral characteristics and potential needs of players under specific behavior patterns can be accurately revealed.

[0125] Finally, the semantic comparison results of each behavior semantic mining branch are comprehensively considered, and the accurate first potential expected demand characteristics are determined through reasonable weight allocation and fusion strategies. Different weights are assigned to different behavior semantic mining branches according to factors such as the frequency and importance of the player's behavior in the game. Then, the semantic comparison results of each branch are weighted and fused to obtain the final first potential expected demand characteristics.

[0126] In this way, through detailed behavior semantic mining, effective semantic comparison and comprehensive analysis and judgment, the first potential expected demand characteristics of players were successfully determined. This process not only demonstrated the profound background and innovative ability in the field of behavior analysis, but also provided strong data support and algorithm guarantee for the game's personalized interactive response, interactive optimization and other applications.

[0127] Such a design can first help game developers to have a deeper understanding of players' behavioral characteristics and potential needs, thereby providing them with a more personalized and interesting gaming experience. For example, by mining the behavioral semantics of players' conversations with NPCs, the game can more intelligently interact and respond to task clues or plot developments that are suitable for players. Secondly, it can help game developers optimize the interactive design of the game. By understanding the potential needs of players under specific behavioral patterns, developers can adjust the game's interactive methods to make them more in line with players' expectations and habits. This will greatly improve the game's ease of use and player satisfaction. For example, if players often express a need for task guidance during conversations, game developers can consider adding more task prompts and guidance functions to NPC conversations.

[0128] Under some other optional design ideas, the target interactive behavior data subset is decoded with behavioral response features based on the potential expected demand features, including: using the potential expected demand features corresponding to the target multimodal behavior semantic vector to determine the potential expected demand features corresponding to each interactive behavior event in the interactive behavior data set to be mined and in the target VR interactive environment; based on the potential expected demand features corresponding to each interactive behavior event, updating the user behavior thermal features of each interactive behavior event in the interactive behavior data set to be mined and in the target VR interactive environment, to obtain personalized behavioral response features obtained by decoding the behavioral response features.

[0129] It can be understood that after identifying the potential expected demand characteristics of the player, the electronic device needs to further convert these characteristics into specific virtual character response behaviors. This process involves decoding the behavioral response characteristics, that is, using the AI ​​model to generate a personalized response strategy based on the player's demand characteristics. In order to achieve this goal, the electronic device will first use the potential expected demand characteristics corresponding to the target multimodal behavior semantic vector to determine the potential expected demand characteristics corresponding to each interactive behavior event in the target VR interactive environment in the interactive behavior data set to be mined. The VR interactive environment here refers to the virtual reality environment within the game, which provides players with a more immersive gaming experience. By determining the potential expected demand characteristics corresponding to each interactive behavior event, the electronic device can better understand the player's behavior and needs in a specific situation.

[0130] Next, based on the potential expected demand characteristics corresponding to each interactive behavior event, the electronic device will update the user behavior thermal characteristics of each interactive behavior event in the target VR interactive environment in the interactive behavior dataset to be mined. User behavior thermal characteristics refer to the hot spots and active areas generated when players interact in the game, which reflects the player's behavior patterns and preferences. By updating these thermal characteristics, the electronic device can more accurately capture the personalized behavior characteristics of the player and provide a basis for the subsequent response strategy generation.

[0131] Through this decoding process, the electronic device can generate a virtual character response behavior data set, which contains various possible responses to the player's specific behavior. These responses are not only based on the player's immediate behavior, but also take into account the player's historical behavior patterns, the current game progress, and the NPC's personality settings and background stories, ensuring that the response is consistent with the game logic and can bring players a real and interesting interactive experience. For example, if the player's potential need is to seek task guidance, the NPC's response may include providing detailed task instructions, marking the location of the task location on the map, or giving key tips for completing the task.

[0132] Finally, the electronic device needs to convert the virtual character response behavior dataset from the matching state of the AI ​​model back to the visual interface state so that the player can see and experience the NPC's response in the game interface. The conversion process includes rendering the AI-generated response content (such as text, actions, expressions, etc.) to the game interface to ensure that they are seamlessly integrated with other elements of the game (such as the environment, character animations). When the NPC provides task instructions to the player, the relevant dialogue text will appear in the dialog box, and the NPC may make corresponding gestures or facial expressions to enhance the realism of the communication.

[0133] In this way, by combining the real-time processing capabilities of neural network algorithms, electronic devices can greatly improve response timeliness and reduce latency in the "XYXY" game. This allows players to enjoy a smoother and more immediate gaming experience when interacting with NPCs. From the recording and conversion of user interaction behavior logs, to the mining and understanding of multimodal behavior semantics, to the decoding and presentation of behavior response features, each link is closely connected and together constitutes this intelligent and personalized NPC interaction experience.

[0134] In the following preferred embodiment, the use of the potential expected demand features corresponding to the target multimodal behavior semantic vector to determine the potential expected demand features corresponding to each interactive behavior event in the target VR interactive environment in the interactive behavior data set to be mined includes: for each interactive behavior event in the target VR interactive environment in the interactive behavior data set to be mined, determining the target behavior operation mode in which the current interactive behavior event is located; the current interactive behavior event is any interactive behavior event in the target VR interactive environment; obtaining the behavior operation preference features of the target behavior operation mode, and the behavior operation preference features of the past behavior operation modes of the target behavior operation mode; the behavior operation preference features include action tendency features and emotional tendency features; when the current interactive behavior event is between the action tendency features and emotional tendency features of the target behavior operation mode, determining the potential expected demand features of the current interactive behavior event through the behavior operation preference features of the past behavior operation mode and the potential expected demand features of the target multimodal behavior semantic vector corresponding to the past behavior operation mode; when the current interactive behavior event matches the target behavior operation mode and the emotional tendency features, determining the potential expected demand features of the current interactive behavior event based on the potential expected demand features of the target multimodal behavior semantic vector corresponding to the target behavior operation mode.

[0135] Furthermore, after identifying the potential expected demand characteristics of the player, the electronic device needs to further convert these characteristics into specific virtual character response behaviors. This process involves decoding the behavioral response characteristics, that is, using the AI ​​model to generate personalized response strategies based on the player's demand characteristics. For example, if the player's potential demand is to seek task guidance, the NPC's response may include providing detailed task instructions, marking the location of the task location on the map, or giving key tips for completing the task. Through the decoding process, the electronic device is able to generate a virtual character response behavior dataset that contains various possible responses to the player's specific behavior. These responses are not only based on the player's immediate behavior, but also take into account the player's historical behavior patterns, the current game progress, and the NPC's personality settings and background stories, ensuring that the response is consistent with the game logic and can bring players a real and interesting interactive experience.

[0136] It is worth mentioning that in the "XYXY" game, the real-time processing capabilities of electronic devices have been fully utilized. Once a structured interactive behavior data set is obtained, the electronic device immediately performs multimodal behavior semantic mining on the target interactive behavior data subset. Using advanced machine learning algorithms, such as deep learning models, electronic devices can identify and understand the complex patterns and meanings behind these behaviors in real time. This real-time processing capability ensures that the player's potential expected demand characteristics can be quickly identified, further improving the response speed and reducing latency. After identifying the player's potential expected demand characteristics, the electronic device also immediately converts these characteristics into specific virtual character response behaviors. Through the real-time decoding process, the electronic device can quickly generate a virtual character response behavior data set containing personalized response strategies. This instant generation capability ensures that the NPC's response can keep up with the player's behavior, further improving the timeliness and fluency of the interaction.

[0137] In addition, when determining the response behavior of the virtual character, the electronic device also takes into account the target VR interaction environment in which the player is located. For each interactive behavior event in the target VR interaction environment in the interactive behavior data set to be mined, the electronic device first determines the target behavior operation mode in which the current interactive behavior event is located. Then, the behavioral operation preference characteristics of the target behavior operation mode and the behavioral operation preference characteristics of past behavioral operation modes are obtained. These behavioral operation preference characteristics include action tendency characteristics and emotional tendency characteristics. By comparing the degree of match between the action tendency characteristics and emotional tendency characteristics of the current interactive behavior event and the target behavior operation mode, the electronic device can more accurately determine the player's potential expected demand characteristics.

[0138] Specifically, if the current interactive behavior event is between the action tendency characteristics and emotional tendency characteristics of the target behavior operation mode, the electronic device will use the behavior operation preference characteristics of the past behavior operation mode and the potential expected demand characteristics of the target multimodal behavior semantic vector corresponding to the past behavior operation mode to determine the potential expected demand characteristics of the current interactive behavior event. This analysis method based on past behavior patterns can help electronic devices better understand the player's behavioral habits and expected needs, thereby generating virtual character responses that are more in line with the player's expectations.

[0139] If the emotional tendency characteristics of the current interactive behavior event and the target behavior operation mode match, the electronic device will directly determine the potential expected demand characteristics of the current interactive behavior event based on the potential expected demand characteristics of the target multimodal behavior semantic vector corresponding to the target behavior operation mode. This direct matching method can further improve the accuracy and efficiency of the response.

[0140] In an optional embodiment, the obtaining of a virtual character response behavior dataset through the personalized behavior response features obtained by decoding the behavior response features includes: obtaining user behavior thermal features of the remaining interactive behavior events except the target VR interactive environment from the interactive behavior dataset to be mined; integrating the user behavior thermal features of the remaining interactive behavior events except the target VR interactive environment and the personalized behavior response features obtained by decoding the behavior response features to obtain the virtual character response behavior dataset.

[0141] Based on this embodiment, by deeply analyzing and integrating diverse interactive behavior data, a rich and personalized virtual character response behavior dataset is constructed. This dataset not only includes the behavioral response characteristics of users in the target VR interactive environment, but also comprehensively considers their behavioral thermal characteristics in other interactive environments, thereby ensuring that the response behavior of the virtual character is closer to reality, comprehensive and layered.

[0142] First, it is necessary to extract information from the interactive behavior dataset to be mined. This dataset is a huge information base that records the behavioral data of users in various VR interactive environments. The goal is to filter out all other interactive behavior events except the target VR interactive environment. This is because we want to build a comprehensive virtual character response model, which needs to take into account the behavioral differences and commonalities of users in different environments. Through screening, the user behavior thermal features of these remaining interactive behavior events are obtained. User behavior thermal features are a quantitative representation method that can reveal the behavior density and preferences of users in a specific interactive environment. For example, in some interactive environments, users may be more inclined to explore, while in other environments they may be more inclined to socialize. These thermal features provide a macro view of user behavior and are an indispensable part of building a virtual character response behavior dataset.

[0143] Secondly, these user behavior thermal features are integrated with the personalized behavior response features obtained by decoding the behavior response features. The personalized behavior response features are obtained by decoding the user's behavior data in the target VR interactive environment through a deep learning model. This process involves complex algorithms and a lot of calculations, but it can reveal the user's unique behavior patterns and preferences in the target environment. For example, some users may be more inclined to take aggressive actions, while others may be more defensive or cooperative. Combining these personalized behavior response features with the user behavior thermal features mentioned above, a comprehensive dataset can be obtained that takes into account both the commonalities and individualities of users.

[0144] The process of integrating the two datasets is a delicate and complex task. It is necessary to ensure that the two features are properly weighted and represented in the dataset so that the avatar's response behavior can reflect both the user's general behavior patterns and their unique personality. To achieve this goal, advanced machine learning algorithms are used, which can intelligently adjust the weights of the features based on the characteristics of the data and the context.

[0145] Then, through this integration process, a comprehensive and hierarchical virtual character response behavior dataset was obtained. This dataset not only contains the thermal characteristics of users' behaviors in various interactive environments, but also incorporates their personalized behavioral response characteristics in the target VR interactive environment. This means that when this dataset is used to train the response behavior model of virtual characters, the virtual characters will be able to more accurately simulate and respond to the behavior of users in various situations.

[0146] For example, in a multiplayer online VR game, the virtual character needs to respond based on the player's behavior. If the player shows a strong desire to explore in the previous interactive environment, the virtual character may show more curiosity and exploration behavior when encountering new game elements. At the same time, if the player tends to adopt a cooperative strategy in the target VR interactive environment, the virtual character may be more inclined to cooperate and share when interacting with other characters.

[0147] In addition, the virtual character response behavior dataset is highly flexible and scalable. As user behavior continues to change and the VR interactive environment continues to update, new behavior features and interaction events can be continuously added to the dataset. This allows the virtual character's response behavior to keep pace with the times and always keep in sync with the user's real behavior.

[0148] In this way, by combining the thermal characteristics of users' behaviors in different VR interactive environments and their personalized behavioral response characteristics in the target environment, a comprehensive, rich and personalized virtual character response behavior dataset is constructed. This dataset provides a solid foundation for the behavior simulation and response of virtual characters, allowing virtual characters to more realistically reflect and respond to users' behaviors and preferences. In future research, we will further explore how to optimize this dataset to improve the accuracy and realism of virtual character response behaviors and provide users with a more immersive and personalized VR experience.

[0149] In some independent embodiments, the method also includes: data association between the user interaction behavior log to be interactively responded to and the virtual character interaction content corresponding to the user interaction behavior log to be interactively responded to obtain an interactive response information group; and securely encrypting the interactive response information group; wherein the secure encryption processing includes information anonymization processing, access permission authentication processing, and information disaster recovery processing.

[0150] It can be understood that the embodiment of the present invention not only focuses on the processing and response of the user interaction behavior log and the virtual character interaction content, but also pays special attention to the data security in the process. Therefore, the method also includes data association of the user interaction behavior log to be interactively responded and the virtual character interaction content corresponding to these logs to form an interactive response information group. More importantly, the interactive response information group is fully securely encrypted to ensure the privacy, integrity and availability of the data.

[0151] First of all, it should be clear that user interaction behavior logs and virtual character interaction content are extremely valuable data. They not only contain sensitive information such as user behavior habits and preferences, but also reflect the interaction logic and response mechanism of virtual characters. Therefore, once these data are leaked or used maliciously, they may infringe on user privacy or damage the stability of the virtual environment. Therefore, it is crucial to encrypt the data securely.

[0152] When performing secure encryption processing, the first consideration is information anonymization. The purpose of this step is to desensitize sensitive data in the interactive response information group to prevent direct identification of specific users or virtual characters through these data. Advanced anonymization technologies such as hash functions and pseudonym replacement are used to ensure that user privacy is protected to the maximum extent while retaining the value of the data.

[0153] Secondly, access rights authentication is also an indispensable part of security encryption. Strict access rights are set for the interactive response information group, and only authenticated and authorized users or systems can access this data. This mechanism effectively prevents unauthorized access and data leakage, providing a strong guarantee for data security.

[0154] In addition to information anonymization and access rights authentication, special attention is also paid to information disaster recovery. In a virtual interactive environment, data loss or damage may have a serious impact on user experience and system stability. Therefore, multiple backup and fault tolerance mechanisms are adopted to ensure that data can be quickly restored in the event of data loss or damage, ensuring the continuity and stability of the system.

[0155] Specifically, information disaster recovery includes regular data backup, off-site backup, and real-time synchronization. Through these measures, data can be quickly restored from backups when data is lost or damaged, ensuring the integrity and availability of user interaction behavior logs and virtual character interaction content.

[0156] In addition, advanced encryption technologies such as AES and RSA are used to encrypt the storage and transmission of interactive response information groups. These encryption technologies can effectively prevent data from being stolen or tampered with during storage and transmission, further improving data security.

[0157] In this way, we not only achieve effective association and processing of user interaction behavior logs and virtual character interaction content, but also pay special attention to data security in the process. Through security encryption measures such as information anonymization, access rights authentication, and information disaster recovery, we ensure the privacy, integrity, and availability of data, providing users with a safe, stable, and reliable virtual interaction environment.

[0158] In some other independent embodiments, in response to the security encryption processing being the information anonymization processing, the security encryption processing of the interactive response information group includes: obtaining the interactive request information flow and the interactive response information flow in the interactive response information group; based on the information flow correlation between the interactive request information flow and the interactive response information flow in the interactive response information group, performing associative anonymous matching on the interactive request information flow and the interactive response information flow in the interactive response information group to obtain an anonymous result of the associated information flow; determining the interactive response information flow with abnormal associative anonymous matching as a sensitive interactive response information flow, and performing associative anonymous matching on the interactive response information flow according to the associated information flow. The information flow commonality weight between the interactive response information flow in the anonymous result and the sensitive interactive response information flow is used to determine the desensitizing feature that matches the sensitive interactive response information flow; the desensitizing feature that matches the sensitive interactive response information flow is associated and anonymously matched with the sensitive interactive response information flow to obtain a desensitizing feature matching result; based on the desensitizing feature matching result and the associated information flow anonymous result, the information cluster to be securely encrypted in the interactive response information group and the anonymous desensitizing guidance vector corresponding to the information cluster to be securely encrypted are determined; the information cluster to be securely encrypted is securely encrypted using the anonymous desensitizing guidance vector.

[0159] In the next step, the interactive request information flow and the interactive response information flow in the interactive response information group are obtained, including: obtaining at least two interactive request texts and at least two interactive response texts in the interactive response information group; obtaining the interactive request text commonality value and the interactive request text difference between the at least two interactive request texts, and obtaining the interactive response text commonality value and the interactive response text difference between the at least two interactive response texts; combining the at least two interactive request texts according to the interactive request text commonality value and the interactive request text difference to obtain the interactive request information flow in the interactive response information group; one interactive request information flow includes at least one interactive request text; combining the at least two interactive response texts according to the interactive response text commonality value and the interactive response text difference to obtain the interactive response information flow in the interactive response information group; one interactive response information flow includes at least one interactive response text.

[0160] In the next step, based on the information flow correlation between the interaction request information flow and the interaction response information flow in the interaction response information group, the interaction request information flow and the interaction response information flow in the interaction response information group are associated and anonymously matched to obtain an anonymous result of the associated information flow, including: determining the interaction response information flow in the interaction response information group as a scene interaction response information flow, and determining the interaction request information flow in the interaction response information group as a scene interaction request information flow; the interaction response text in the scene interaction response information flow is extracted from the target crawler text for the interaction response information group; obtaining the interaction request text in the target crawler text; determining the text commonality value between the interaction request text in the target crawler text and the interaction request text in the scene interaction request information flow as the information flow correlation between the scene interaction response information flow and the scene interaction request information flow; when the information flow correlation is greater than or equal to the correlation threshold, the scene interaction response information flow and the scene interaction request information flow are associated and anonymously matched to obtain the anonymous result of the associated information flow.

[0161] Based on the above three embodiments involving information anonymization processing, in order to ensure the security of user interaction behavior logs and their corresponding virtual character interaction content after association, a series of sophisticated and efficient anonymous encryption measures are taken.

[0162] First, the interactive request information flow and the interactive response information flow are extracted from the interactive response information group. This step is the basis for subsequent anonymous processing, which requires accurate identification and separation of these two types of information flows. In order to achieve this goal, at least two interactive request texts and at least two interactive response texts in the interactive response information group are obtained. Then, the commonality values ​​and differences between these texts are calculated, and these values ​​reflect the similarities and differences between the texts. Based on these values, the interactive request texts are combined into an interactive request information flow, and similarly, the interactive response texts are combined into an interactive response information flow. In this way, the information flow required for subsequent processing is obtained.

[0163] Next, according to the information flow correlation between the interaction request information flow and the interaction response information flow, they are associated and anonymously matched. The purpose of this step is to find those interaction response information flows that may leak user privacy or sensitive information and anonymize them. In order to achieve this goal, the interaction response information flow is determined as the scene interaction response information flow, and the interaction request information flow is determined as the scene interaction request information flow. The interaction response text in the scene interaction response information flow is extracted from the target crawler text for the interaction response information group, which ensures that the processing is based on actual and dynamic interaction data. Then, the text commonality value between the interaction request text in the target crawler text and the interaction request text in the scene interaction request information flow is calculated, which reflects the degree of association between the two. When the degree of association is greater than or equal to the set correlation threshold, the scene interaction response information flow and the scene interaction request information flow are considered to be related, and they are associated and anonymously matched.

[0164] In the process of associated anonymous matching, the anonymous results of associated information flows are obtained. This result reveals which interactive response information flows are sensitive and need to be desensitized. These sensitive interactive response information flows are identified as sensitive interactive response information flows, and based on the information flow commonality weights between the interactive response information flows in the anonymous results of associated information flows and the sensitive interactive response information flows, the desensitized features that match the sensitive interactive response information flows are determined. These desensitized features are the key to anonymization, and they can ensure the availability and integrity of data while protecting user privacy.

[0165] After obtaining the desensitized features, the next step is to associate and anonymously match the desensitized features that match the sensitive interaction response information flow with the sensitive interaction response information flow. The purpose of this step is to find the best anonymity method to protect user privacy to the greatest extent. Through this process, the desensitized feature matching results are obtained, which guide how to anonymize the sensitive interaction response information flow.

[0166] Finally, based on the desensitized feature matching results and the anonymous results of the associated information flow, the information clusters to be securely encrypted in the interactive response information group and the anonymous desensitized guidance vectors corresponding to these information clusters to be securely encrypted are determined. These guidance vectors are the key to secure encryption processing, and they ensure that the encryption processing is accurate and effective. By performing secure encryption processing on the information clusters to be securely encrypted using anonymous desensitized guidance vectors, an interactive response information group is finally obtained that protects user privacy while maintaining data integrity and availability.

[0167] In this way, through a series of steps such as fine information flow extraction, associated anonymous matching, desensitized feature determination and secure encryption processing, efficient and secure encryption processing of interactive response information groups is achieved. This processing not only ensures the protection of user privacy, but also maintains the integrity and availability of data, providing a strong guarantee for the security of the virtual interactive environment and user experience.

[0168] In some other independent embodiments, in response to the security encryption processing being the access right authentication processing, the security encryption processing of the interactive response information group includes: generating an access right authentication relationship topology based on a security assessment report of the interactive response information group, wherein the security assessment report is used to indicate an access right authentication operation, and the access right authentication relationship topology includes multiple access right authentication items and multiple live feature recognition items that are cross-related in the access right authentication operation; adding the first template access application information and the second template access application information of the interactive response information group to at least one of the access right authentication items, and adjusting the noise disturbance tolerance coefficient of at least one of the live feature recognition items that matches the access right authentication item; and updating the access right policy of the interactive response information group based on the noise disturbance tolerance coefficient.

[0169] The above embodiments pay special attention to the secure encryption processing of the interactive response information group, especially the core link of access rights authentication processing. In order to ensure the security of the user interaction behavior log and its corresponding virtual character interaction content after association, a series of innovative and efficient technical solutions are adopted.

[0170] First, based on the security assessment report of the interactive response information group, a permission authentication relationship topology is generated. The security assessment report is an important basis for access right authentication operations. It comprehensively analyzes the security status of the interactive response information group and points out the key points that require access right authentication. The permission authentication relationship topology is a complex network structure, which includes multiple cross-related access right authentication items and live feature recognition items in the access right authentication operation. These items and the relationships between them constitute the basic framework for access right authentication.

[0171] Next, the first template access application information and the second template access application information of the interactive response information group are added to at least one access right authentication item. The purpose of this step is to associate the specific access application information with the authentication item for subsequent authentication processing. At the same time, the noise disturbance tolerance coefficient of at least one live feature recognition item matched with the access right authentication item is also adjusted. This coefficient is an important parameter for live feature recognition, which determines the degree of noise and disturbance that can be tolerated during the recognition process. By adjusting this coefficient, it is possible to better adapt to different interactive response information groups and access environments, and improve the accuracy and robustness of recognition.

[0172] After completing the above steps, the access permission policy for the interactive response information group is updated based on the noise disturbance tolerance coefficient. This step is the core of the entire access permission authentication process, which ensures that the access permission policy can keep pace with the times and adapt to the ever-changing interactive response information group and access environment. By updating the access permission policy, user privacy and data security can be better protected to prevent unauthorized access and data leakage.

[0173] Specifically, access rights authentication processing includes the following key steps:

[0174] (1) Security assessment and topology generation: First, conduct a comprehensive security assessment on the interactive response information group and generate a detailed security assessment report. Then, based on the report, build a permission authentication relationship topology to clarify the relationship and dependency between each access permission authentication project and the live feature recognition project;

[0175] (2) Adding access application information and adjusting coefficients: Add the first template access application information and the second template access application information in the interactive response information group to the appropriate access permission authentication items. At the same time, according to actual needs, adjust the noise disturbance tolerance coefficient of the living feature recognition items that match these authentication items to ensure the accuracy and stability of recognition;

[0176] (3) Access permission policy update: After completing the above steps, the access permission policy of the interactive response information group is updated based on the noise disturbance tolerance coefficient and other relevant factors. This step ensures that the access permission policy can adapt to the ever-changing interactive environment and security requirements.

[0177] Through this series of designed steps, efficient and secure access rights authentication processing for interactive response information groups is achieved. This processing not only ensures user privacy and data security, but also improves the availability and stability of the system, providing a strong guarantee for the security of the virtual interactive environment and user experience.

[0178] In practical applications, this embodiment can be widely used in various scenarios that need to protect user privacy and data security, such as online games, virtual social networking, distance education, etc. By adopting technical solutions, these scenarios can run more safely and stably, providing users with better experience and services.

[0179] It can be seen that through the innovative access rights authentication processing mechanism, efficient and secure protection of the interactive response information group is achieved. This technical solution not only has broad application prospects and practical value, but also provides new ideas and methods for the security and user experience of the virtual interactive environment.

[0180] In some other independent embodiments, in response to the secure encryption processing being the information disaster recovery processing, the secure encryption processing of the interactive response information group includes: obtaining information group structured vector samples and information disaster recovery pattern annotations of respective multiple interactive response information group samples; determining, from the basic vector coordinate system, information disaster recovery pattern vectors of respective disaster recovery rules of respective information disaster recovery pattern annotations according to the mapping distribution of each of the information group structured vector samples in the basic vector coordinate system; for each of the interactive response information group samples, fusing the information group structured vector sample of the interactive response information group sample, and the difference vector between the information group structured vector sample of the interactive response information group sample and at least a part of the information disaster recovery pattern vector, to obtain a structured disaster recovery output vector sample of the interactive response information group sample; combining the information group structured vector sample containing each of the information group structured vector samples with the information disaster recovery pattern vector sample ... The structured disaster recovery output vector sample and the training set of each information disaster recovery mode annotation are used for network training to obtain an information disaster recovery decision network for information disaster recovery decision; the current information group structured vector of the interactive response information group is obtained; the current information group structured vector and the difference vector between the current information group structured vector and at least a part of each information disaster recovery mode vector are integrated to obtain the current structured disaster recovery output vector of the interactive response information group; the current structured disaster recovery output vector is obtained to be the same as the information disaster recovery mode vector used in the process of obtaining the structured disaster recovery output vector sample; based on the information disaster recovery decision network, the current structured disaster recovery output vector is used to perform information disaster recovery decision to obtain the information disaster recovery decision view of the interactive response information group; and the interactive response information group is securely encrypted based on the information disaster recovery decision view.

[0181] An advanced and efficient technical solution is adopted for the secure encryption processing of interactive response information groups, especially the information disaster recovery processing part. This solution not only ensures the security of interactive response information groups during storage and transmission, but also greatly improves the efficiency and accuracy of information disaster recovery, thereby ensuring the recoverability and integrity of data in the face of various catastrophic events.

[0182] First, we build a training foundation by acquiring multiple interactive response information group samples. These samples not only contain the structured vectors of their respective information groups, but also come with annotations of information disaster recovery modes. This step is the prerequisite for building an information disaster recovery decision network. It ensures that there is sufficient data support in the subsequent training process, so that the trained model can accurately identify and apply different information disaster recovery modes.

[0183] Secondly, based on the mapping distribution of these information group structured vector samples in the basic vector coordinate system, the information disaster recovery pattern vector of the disaster recovery rule corresponding to each information disaster recovery pattern annotation is determined from the coordinate system. This process is the key to model learning, which enables the model to understand and distinguish different disaster recovery modes, laying the foundation for the subsequent generation of structured disaster recovery output vectors.

[0184] Next, for each interactive response information group sample, by fusing its information group structured vector sample and the difference vector between the vector sample and at least a part of each information disaster recovery mode vector, the structured disaster recovery output vector sample of the sample is obtained. This step is the core of the model training data preparation. It generates input data for training the information disaster recovery decision network, ensuring that the network can learn how to convert the interactive response information group into a structured disaster recovery output.

[0185] After completing the generation of structured disaster recovery output vector samples, a training set was constructed by combining these vector samples and the corresponding information disaster recovery mode annotations. The training set was used for network training, and finally an information disaster recovery decision network for information disaster recovery decision-making was obtained. This network is the core of the technical solution. It can output the corresponding information disaster recovery decision viewpoint based on the input interactive response information group and guide the subsequent security encryption processing.

[0186] When the actual interactive response information group needs to be securely encrypted, the current information group structured vector of the information group is first obtained. Then, by fusing the current information group structured vector and the difference vector between the vector and at least a part of each information disaster recovery mode vector, the current structured disaster recovery output vector of the interactive response information group is obtained. This process is similar to the process of generating a sample of structured disaster recovery output vectors, ensuring that the model can consistently process actual data and training data.

[0187] It is worth noting that in the process of obtaining the current structured disaster recovery output vector, the information disaster recovery pattern vector used is the same as the vector used when generating the structured disaster recovery output vector sample. This design ensures the accuracy and consistency of the model when processing actual data.

[0188] Finally, based on the information disaster recovery decision network, the information disaster recovery decision of the current structured disaster recovery output vector is made, and the information disaster recovery decision view of the interactive response information group is obtained. This view guides the subsequent security encryption processing and ensures that the interactive response information group can be effectively protected and restored in the face of various catastrophic events.

[0189] In this way, by building an information disaster recovery decision network, efficient and accurate information disaster recovery processing of interactive response information groups is achieved. It not only improves data security, but also ensures the recoverability and integrity of data in the face of catastrophic events. The implementation of this solution provides new ideas and methods for the secure encryption processing of interactive response information groups, which is of great significance for ensuring data security.

[0190] For example, in order to transform the user interaction behavior log from the visual interface state to the AI ​​model matching state, a method based on the interface scanning core and the data format conversion tag can be used. The following will illustrate how to implement this transformation process through a specific algorithm example and parameter example.

[0191] 1. Algorithm Example

[0192] 1.1. Initialization:

[0193] Set the scanning period T (for example, once per second).

[0194] Create an empty interactive behavior dataset D to be mined.

[0195] 1.2. Get the interface scan kernel:

[0196] Define an interface scanning core SCAN_CORE, which contains a set of rules for identifying and extracting interface elements.

[0197] 1.3. Periodic scanning:

[0198] Every time T, use SCAN_CORE to scan the user interaction behavior log.

[0199] 1.4. Determine the data format conversion tag:

[0200] For each interface area scanned by SCAN_CORE, analyze the behavior log set it contains.

[0201] According to the content and structure of the behavior log set, one or more data format conversion tags FORMAT_TAGS are assigned to the interface area.

[0202] 1.5. Data preprocessing and format conversion:

[0203] According to FORMAT_TAGS, the behavior logs in the scanned interface area are preprocessed and format converted.

[0204] Add the converted data to the interactive behavior dataset D to be mined.

[0205] 1.6 Output:

[0206] Return the interactive behavior dataset D to be mined.

[0207] 2. Parameter examples

[0208] Scanning cycle T: 1000 milliseconds (i.e., one scan per second).

[0209] Interface scanning core SCAN_CORE:

[0210] Contains rules: identifying click events, identifying text input events, identifying sliding events, etc.

[0211] Data format conversion tags FORMAT_TAGS:

[0212] Click event:

[0213] {"type":"click","timestamp":" <timestamp>","x":"<x_coord> ","y":"<y_coord> "}

[0214] Text input events:

[0215] {"type":"text_input","timestamp":" <timestamp>","text":"<input_text> "}

[0216] Slide event:

[0217] {"type":"swipe","timestamp":" <timestamp>","start_x":"<start_x> ","start_y":"<start_y> ","end_x":"<end_x> ","end_y":"<end_y> "}

[0218] 3. Implementation process

[0219] Initialize the scanning period T to 1000 milliseconds and create an empty interactive behavior dataset D to be mined.

[0220] Define the interface scanning core SCAN_CORE, which contains rules for identifying click events, text input events, and sliding events.

[0221] Every 1000 milliseconds, use SCAN_CORE to scan the user interaction behavior log.

[0222] For each scanned interface area, the behavior log set it contains is analyzed, and corresponding data format conversion tags FORMAT_TAGS are assigned according to the content.

[0223] According to FORMAT_TAGS, the behavior logs are preprocessed and format converted, and then the converted data is added to the interactive behavior dataset D to be mined.

[0224] Returns the interactive behavior dataset D to be mined for subsequent AI model training and behavior analysis.

[0225] Through this method, the user interaction behavior log can be effectively transformed from the visual interface state to the AI ​​model matching state, obtaining a structured, high-quality interaction behavior dataset to be mined.

[0226] In another example, to transform the user interaction behavior log from the visual interface state to the AI ​​model matching state and obtain the corresponding interaction behavior data set to be mined, the process can be implemented using the Python language.

[0227] class InterfaceScannerCore:

[0228] def __init__(self, scan_period):

[0229] self.scan_period = scan_period

[0230] self.data_format_conversion_tags = {}

[0231] def scan_user_interaction_logs(self, user_interaction_logs):

[0232] #Simulate the scanning process. In actual applications, it needs to be replaced with the real interface scanning logic.

[0233] scanned_data = {}

[0234] for log in user_interaction_logs:

[0235] interface_region = log['interface_region']

[0236] behavior_data = log['behavior_data']

[0237] # Assume that each interface area has a preset data format conversion label

[0238] if interface_region not in self.data_format_conversion_tags:

[0239] # In actual applications, these tags may be obtained from configuration files or databases

[0240] self.data_format_conversion_tags[interface_region] = {

[0241] 'data_type': 'structured',

[0242] 'structure': 'dict',

[0243] 'conversion_rules': ['remove_duplicates', 'normalize']

[0244] }

[0245] # Convert data according to data format conversion label

[0246] converted_data = self.convert_data(behavior_data,self.data_format_conversion_tags[interface_region])

[0247] # Store the converted data in the dictionary of the corresponding interface area

[0248] if interface_region not in scanned_data:

[0249] scanned_data[interface_region] = []

[0250] scanned_data[interface_region].append(converted_data)

[0251] return scanned_data

[0252] def convert_data(self, data, conversion_tags):

[0253] # Transform data according to the transformation rules

[0254] converted_data = data # Initially, assume that data does not need to be converted

[0255] if 'conversion_rules' in conversion_tags:

[0256] for rule in conversion_tags['conversion_rules']:

[0257] if rule == 'remove_duplicates':

[0258] # Remove duplicates (example logic)

[0259] if isinstance(converted_data, list):

[0260] converted_data = list(set(converted_data))

[0261] elif rule == 'normalize':

[0262] # Normalize data (example logic)

[0263] if isinstance(converted_data, dict) and 'value' in converted_data:

[0264] converted_data['value'] = converted_data['value'] / 100.0

[0265] return converted_data

[0266] # Example usage

[0267] user_interaction_logs =

[0268] {'interface_region': 'game_menu', 'behavior_data': {'action': 'click', 'timestamp': 1672531200}},

[0269] {'interface_region': 'inventory', 'behavior_data': {'action': 'open', 'timestamp': 1672531201, 'items': ['sword','shield']}},

[0270] #... Other log entries

[0272] scanner_core = InterfaceScannerCore(scan_period=10) # Scan every 10 seconds

[0273] scanned_data = scanner_core.scan_user_interaction_logs(user_interaction_logs)

[0274] print(scanned_data)

[0275] ​In this example, the InterfaceScannerCore class simulates the functionality of the interface scanning core. It has a scan_user_interaction_logs method that accepts user interaction behavior logs as input and returns the converted interaction behavior dataset to be mined. The data conversion process is performed according to the data format conversion label of each interface area.

[0276] Further, Figure 2 FIG. 2 is a schematic diagram of the structure of an electronic device 200 provided in an embodiment of the present invention. Figure 2 The electronic device 200 shown includes a processor 210. The processor 210 can call and run a computer program from a memory to implement the method in the embodiment of the present invention.

[0277] Alternatively, if Figure 2 As shown, the electronic device 200 may further include a memory 230. The processor 210 may call and run a computer program from the memory 230 to implement the method in the embodiment of the present invention.

[0278] The memory 230 may be a separate device independent of the processor 210 , or may be integrated into the processor 210 .

[0279] Alternatively, if Figure 2 As shown, the electronic device 200 may further include a transceiver 220, and the processor 210 may control the transceiver 220 to interact with other devices, specifically, to send information or data to other devices, or to receive information or data sent by other devices.

[0280] Optionally, the electronic device 200 may implement the corresponding processes corresponding to the storage engine or components in the storage engine (such as a processing module) or a device deployed with a storage engine in each method of the embodiments of the present invention, which will not be described in detail here for the sake of brevity.

[0281] It should be understood that the processor of the embodiment of the present invention may be an integrated circuit chip with signal processing capability.

[0282] The embodiments of the present invention are described above in conjunction with the accompanying drawings, but the present invention is not limited to the above-mentioned specific implementation methods. The above-mentioned specific implementation methods are merely illustrative and not restrictive. Under the guidance of the present invention, ordinary technicians in this field can also make many forms without departing from the purpose of the present invention and the scope of protection of the present invention, all of which are within the protection of the present invention.< / timestamp> < / timestamp> < / timestamp>

Claims

1. A method for generating virtual character interactive content based on behavior analysis, characterized in that: Applied to electronic equipment, the method comprises: Obtain the user interaction behavior log to be interactively responded to, transform the user interaction behavior log from the visual interface state to the AI ​​model matching state, and obtain the corresponding interaction behavior data set to be mined; Performing multimodal behavior semantic mining on the target interactive behavior data subset in the interactive behavior data set to be mined to obtain a target multimodal behavior semantic vector, and obtaining potential expected demand features corresponding to the target multimodal behavior semantic vector; wherein multimodal behavior semantic mining refers to using machine learning algorithms to identify and understand the complex patterns and meanings of the target interactive behavior data subset; the target multimodal behavior semantic vector includes analyzing the text content of the dialogue between the player and the NPC, the context in which the dialogue occurs, and the player's behavior sequence; the potential expected demand features include the user's current goal, emotional state or preference; Decoding the target interactive behavior data subset with behavioral response characteristics according to the potential expected demand characteristics, and obtaining a virtual character response behavior data set through personalized behavioral response characteristics obtained through the behavioral response feature decoding; Transforming the virtual character response behavior data set from an AI model matching state to a visual interface state, and obtaining virtual character interaction content corresponding to the user interaction behavior log to be interactively responded to; The method of performing multimodal behavior semantic mining on a target interactive behavior data subset in the interactive behavior data set to be mined to obtain a target multimodal behavior semantic vector includes: obtaining user behavior thermal characteristics of each interactive behavior event in the target VR interactive environment in the interactive behavior data set to be mined; performing multimodal behavior semantic mining on the user behavior thermal characteristics of each interactive behavior event in the target VR interactive environment through each behavioral semantic mining branch in a pre-configured behavioral semantic mining model to obtain a target multimodal behavior semantic vector of the target interactive behavior data subset in each behavioral semantic mining branch; The configuration method of the behavior semantic mining model includes: for any behavior operation mode among a plurality of predefined behavior operation modes, determining a corresponding behavior semantic mining branch according to a behavior operation preference feature corresponding to the behavior operation mode, wherein the behavior operation preference feature includes at least one of an action tendency feature, an emotion tendency feature or an expression tendency feature; and configuring a behavior semantic mining model based on the behavior semantic mining branches corresponding to each behavior operation mode; The method performs multimodal behavior semantic mining on the user behavior thermal characteristics of each interactive behavior event in the target VR interactive environment through each behavior semantic mining branch in a pre-configured behavior semantic mining model, and obtains a target multimodal behavior semantic vector of the target interactive behavior data subset in each behavior semantic mining branch, including: for each behavior semantic mining branch, retrieving the target interactive behavior event in the interactive behavior event relationship tree corresponding to the current behavior semantic mining branch from the interactive behavior events contained in the target VR interactive environment; the current behavior semantic mining branch is any behavior semantic mining branch; retrieving event element information corresponding to the target interactive behavior event in the current behavior semantic mining branch, and based on the event element information corresponding to the target interactive behavior event, performing event element recognition on the user behavior thermal characteristics of the target interactive behavior event; and determining the target multimodal behavior semantic vector of the target interactive behavior data subset in the current behavior semantic mining branch based on the event element recognition result corresponding to the target interactive behavior event.

2. The method according to claim 1, characterized in that The step of transforming the user interaction behavior log from a visual interface state to an AI model matching state to obtain a corresponding interaction behavior data set to be mined includes: Obtaining an interface scanning core, and scanning the interface scanning core in the user interaction behavior log according to a set scanning period; For each interface area scanned by the interface scanning core, determining a data format conversion label at the corresponding interface area through the interface scanning core and the behavior log set contained in the interface scanning core in the user interaction behavior log; Based on the data format conversion labels at each interface area scanned by the interface scanning core, the interactive behavior dataset to be mined is transformed.

3. The method according to claim 1, characterized in that There are multiple target multimodal behavior semantic vectors, and obtaining potential expected demand features corresponding to the target multimodal behavior semantic vectors includes: A plurality of target multimodal behavior semantic vectors are input into a debugged interactive response prediction algorithm, and the plurality of target multimodal behavior semantic vectors are processed respectively by the debugged interactive response prediction algorithm to generate potential expected demand features corresponding to each target multimodal behavior semantic vector.

4. The method according to claim 3, characterized in that The debugging method of the interactive response prediction algorithm includes: Obtaining an algorithm learning example set, wherein each algorithm learning example in the algorithm learning example set includes an interactive behavior log adversarial example and an interactive behavior log authentication example; Transform the interactive behavior log confrontation sample from the visualization interface state to the AI ​​model matching state to obtain the interactive behavior confrontation data set to be mined, and transform the interactive behavior log authentication sample from the visualization interface state to the AI ​​model matching state to obtain the interactive behavior authentication data set to be mined; Perform multimodal behavior semantic mining on the target interactive behavior data subset in the interactive behavior adversarial dataset to be mined to obtain a multimodal behavior perturbation semantic vector, and perform multimodal behavior semantic mining on the target interactive behavior data subset in the interactive behavior authentication dataset to be mined to obtain a multimodal behavior authentication semantic vector; Determining a first potential expected demand feature through the multimodal behavior disturbance semantic vector and the multimodal behavior authentication semantic vector; Determine the second potential expected demand feature through the multimodal behavior disturbance semantic vector and the interactive response prediction algorithm to be debugged; Determining an algorithm learning error through the first potential expected demand feature and the second potential expected demand feature; The algorithm parameters of the interactive response prediction algorithm to be debugged are improved by using the algorithm learning error until the algorithm training meets the standard requirements, thereby obtaining a debugged interactive response prediction algorithm; The multimodal behavior disturbance semantic vector is obtained by mining each behavior semantic mining branch in the behavior semantic mining model, and the multimodal behavior authentication semantic vector is obtained by mining each behavior semantic mining branch in the behavior semantic mining model; The determining of the first potential expected demand feature by using the multimodal behavior disturbance semantic vector and the multimodal behavior authentication semantic vector includes: For any behavior semantic mining branch, determine the semantic comparison result corresponding to the behavior semantic mining branch in combination with the multimodal behavior perturbation semantic vector and the multimodal behavior authentication semantic vector mined by the behavior semantic mining branch; Based on the semantic comparison results corresponding to each behavior semantic mining branch, the first potential expected demand feature is determined.

5. The method according to claim 1, characterized in that The step of decoding the behavior response characteristics of the target interaction behavior data subset according to the potential expected demand characteristics includes: Determine the potential expected demand features corresponding to each interactive behavior event in the target VR interactive environment in the interactive behavior data set to be mined by using the potential expected demand features corresponding to the target multimodal behavior semantic vector; Based on the potential expected demand characteristics corresponding to each interactive behavior event, the user behavior thermal characteristics of each interactive behavior event in the target VR interactive environment in the interactive behavior data set to be mined are updated to obtain the personalized behavior response characteristics obtained by decoding the behavior response characteristics; The method of using the potential expected demand features corresponding to the target multimodal behavior semantic vector to determine the potential expected demand features corresponding to each interactive behavior event in the target VR interactive environment in the interactive behavior data set to be mined includes: For each interactive behavior event in the target VR interactive environment in the interactive behavior data set to be mined, determining the target behavior operation mode of the current interactive behavior event; the current interactive behavior event is any interactive behavior event in the target VR interactive environment; Acquire the behavior operation preference characteristics of the target behavior operation mode and the behavior operation preference characteristics of the past behavior operation modes of the target behavior operation mode; the behavior operation preference characteristics include action tendency characteristics and emotion tendency characteristics; When the current interactive behavior event is between the action tendency feature and the emotion tendency feature of the target behavior operation mode, determining the potential expected demand feature of the current interactive behavior event through the behavior operation preference feature of the past behavior operation mode and the potential expected demand feature of the target multimodal behavior semantic vector corresponding to the past behavior operation mode; When the current interactive behavior event matches the target behavior operation mode and the emotional tendency characteristics, the potential expected demand characteristics of the current interactive behavior event are determined based on the potential expected demand characteristics of the target multimodal behavior semantic vector corresponding to the target behavior operation mode.

6. The method according to any one of claims 1 to 5, characterized in that: The step of obtaining a virtual character response behavior data set by decoding the personalized behavior response characteristics comprises: Acquire user behavior thermal characteristics of the remaining interactive behavior events except the target VR interactive environment from the interactive behavior data set to be mined; Based on the user behavior thermal characteristics of the remaining interactive behavior events except the target VR interactive environment and the personalized behavior response characteristics decoded from the behavior response characteristics, a virtual character response behavior dataset is integrated.

7. An electronic device, characterized in that: The method comprises at least one processor and a memory; the memory stores computer-executable instructions; the at least one processor executes the computer-executable instructions stored in the memory, so that the at least one processor executes the method according to any one of claims 1 to 6.

Images