Interactive content expression layer adaptive switching method based on artificial intelligence scene state perception
Patent Information
- Application Number
- CN202611426357.5
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-09-15
- Publication Date
- 2026-10-20
Smart Images
Figure FT_1 
Figure FT_2
Abstract
Description
Technical Field
[0001] This invention relates to the field of human-computer interaction content switching technology, specifically to an adaptive switching method for the interactive content expression layer based on artificial intelligence scene state perception. Background Technology
[0002] In artificial intelligence application scenarios, human-computer interaction systems typically rely on specific AI product interfaces, interaction parameters, prompt word templates, and resource configurations to present learning or task content. Existing systems mostly adopt a design pattern of coupled storage of content and expression methods, that is, a set of content corresponds to a fixed set of interface paths, parameter entries, prompt word formats, and task organization methods. When the application scenario changes dynamically, such as AI product interface upgrades leading to changes in interaction paths, parameter configuration adjustments causing the original settings to become invalid, or interaction format updates causing the original templates to be unable to be correctly recognized by the current system, existing systems often need to recreate the entire content scheme or carry out large-scale code-level updates, making it impossible to flexibly adjust its expression layer binding while maintaining the stability of the core content.
[0003] Meanwhile, as users continue to use the system, their understanding of the current content, their operational proficiency, and their willingness to explore will gradually change. This change does not mean that users need to change the learning or task content itself, but rather that the current content depth and task format may no longer be suitable for the user's latest state. Although existing adaptive systems can recommend new learning resources based on user profiles or historical behavior, their recommendation logic usually confuses changes in user state with content switching. They cannot distinguish whether external interaction conditions have changed or the user's internal state has evolved, nor can they accurately control the range of fields adjusted in the expression layer for different sources of change. This leads to frequent erroneous or excessive switching in actual operation, reducing system stability and user experience. At the same time, they cannot form a traceable and recalculated switching decision mechanism, affecting the system's reliability and continuous evolution capability in complex application scenarios.
[0004] Therefore, how to accurately distinguish the source of change in the dynamic process of artificial intelligence application scenarios and precisely control the adaptive switching of the interactive content expression layer accordingly, while ensuring the consistency and traceability of the switching process, has become an urgent problem to be solved in current human-computer interaction systems. Summary of the Invention
[0005] To address the shortcomings of existing technologies, this invention provides an adaptive switching method for interactive content expression layers based on artificial intelligence scene state perception, which can effectively solve the problems mentioned in the background technology.
[0006] To achieve the above objectives, the present invention provides the following technical solution: an adaptive switching method for interactive content expression layers based on artificial intelligence scene state perception, comprising: acquiring a stable content core of the target interactive content and its current expression layer binding, wherein the stable content core is stored separately from multiple expression layer bindings, and the expression layer bindings are used to characterize the content depth, task form, and execution parameters of the stable content core in the current artificial intelligence application scenario, and are associated with and stored with a historical state baseline vector generated when the expression layer binding takes effect; collecting observation events generated by the target object around the target interactive content in the artificial intelligence application scenario, and updating the scene adaptation state vector, cognitive state vector, and environmental dynamic vector of the target object respectively through the observation events within the target time window. The system generates a state vector and a current comprehensive state vector based on the updated state vector. It then detects the executability of the current expression layer binding to generate an execution condition change identifier, compares the current comprehensive state vector with the historical state baseline vector associated with the current expression layer binding to generate a state change identifier, and determines a switching trigger type based on the combination of the execution condition change identifier and the state change identifier. Finally, it matches a field update mask according to the switching trigger type, obtains a set of field changes for candidate expression layer bindings relative to the current expression layer binding, and verifies whether the set of field changes is completely covered by the field update mask. If so, it switches the current expression layer binding to the candidate expression layer binding while maintaining the stable content core unchanged.
[0007] As a further method, the scenario adaptation state vector includes interface operation reachability state, parameter pattern resolvability state, and resource pointer availability state. The update method for the interface operation reachability state is as follows: The system detects whether the execution paths of each action in the interface action sequence of the current artificial intelligence application scenario are complete and reachable, counts and calculates the ratio of the number of actions with complete and reachable execution paths to the total number of actions, and determines the reachability status value of the interface operation based on the ratio.
[0008] As a further method, when the interface operation reachability state value is lower than a preset executable threshold, the number of unexecutable actions in the interface action sequence bound to the current expression layer is detected, and when the number of unexecutable actions changes, the execution condition change identifier is generated.
[0009] As a further method, detecting the executability of the currently bound expression layer to generate an execution condition change flag also includes: The system detects whether the prompt word pattern bound to the current expression layer can be parsed by the current artificial intelligence model, and detects whether the resource pointer bound to the current expression layer is accessible; when the prompt word pattern is unparsable or the resource pointer is inaccessible, the system generates the execution condition change flag.
[0010] As a further method, the detection method for whether the prompt word pattern can be parsed by the current artificial intelligence model is as follows: The prompt word template in the current expression layer binding is input into the parser of the current artificial intelligence model to obtain the syntax tree structure returned by the parser. If there are unrecognized syntax nodes in the syntax tree structure, the prompt word pattern is determined to be unparseable.
[0011] As a further method, the current integrated state vector is compared with the historical state reference vector associated with the current expression layer to generate a state change identifier. The specific process is as follows: Calculate the vector distance between the current comprehensive state vector and the historical state baseline vector. When the vector distance exceeds a preset state change distance threshold, generate the state change identifier. The vector distance is calculated using weighted Euclidean distance, that is, the scene adaptation state vector, the cognitive state vector, and the environmental dynamic state vector are respectively matched with corresponding weight coefficients.
[0012] As a further method, the cognitive state vector includes task comprehension state, operation proficiency state, and exploration willingness state; when only the state change identifier is generated but the execution condition change identifier is not generated, the field update mask covers the content depth field and task form field corresponding to the task comprehension state, the operation proficiency state, and the exploration willingness state.
[0013] As a further method, when the execution condition change identifier and the state change identifier are generated simultaneously, the field update mask covers the execution parameter fields corresponding to the interface operation reachability state, the parameter pattern parseability state, and the resource pointer availability state, respectively. The field update mask also covers the content depth field and task form field corresponding to the task comprehension state, the operation proficiency state, and the exploration willingness state.
[0014] As a further method, the candidate expression layer bindings are obtained as follows: From multiple expression layer bindings associated with the core stable content, expression layer bindings that match the content depth field and the task form field with the task comprehension state, the operation proficiency state, and the exploration willingness state are selected, and the selected expression layer bindings are used as candidate expression layer bindings.
[0015] Compared with the prior art, the embodiments of the present invention have at least the following advantages or beneficial effects: (1) This invention provides an adaptive switching method for interactive content expression layers based on artificial intelligence scene state perception. It binds and stores the stable content core with multiple expression layers separately, so that when the expression mode or interaction mode changes, there is no need to rebuild the entire content scheme, which significantly reduces the data update cost of content maintenance and system upgrade. By binding and storing each expression layer in association with the historical state benchmark vector generated when it takes effect, each state comparison has a clear and fixed reference benchmark, avoiding decision bias caused by misuse of cross-version benchmarks. By collecting and observing events to update the scene adaptation state vector, cognitive state vector and environmental dynamic state vector respectively, the system can comprehensively perceive from multiple independent dimensions. By understanding the real-time interaction status between the target object and the AI application scenario, and generating execution condition change identifiers and state change identifiers to determine the switching trigger type, the system can accurately classify the switching requirements of the interactive content expression layer. Then, based on the switching trigger type, the system matches the corresponding field update mask, and only executes the switching when the set of field changes bound to the candidate expression layer is completely covered by the field update mask. This effectively prevents erroneous switching caused by the mismatch between the source of change and the rewritten field, while maintaining the core of stable content and ensuring the continuity and integrity of the core interaction logic. This significantly improves the accuracy, stability, and system reliability of the interactive content expression layer switching in AI application scenarios.
[0016] (2) This invention further configures the scene adaptation state vector into interface operation reachability state, parameter pattern resolvability state, and resource pointer availability state, and quantifies the interface operation reachability state value by the ratio of the number of actions with complete execution paths to the total number of actions. This enables the system to accurately measure the degree of support for interactive operations in the current artificial intelligence application scenario in a structured numerical manner. When the interface operation reachability state value is lower than the preset executable threshold, an execution condition change indicator is generated by detecting whether the number of non-interactive actions whose execution paths are blocked in the interface action sequence has changed. This enables the system to detect when interface elements are missing, paths are blocked, or... In scenarios such as unreachable interactions, the system automatically senses changes in execution conditions and promptly triggers a switch in the expression layer. Simultaneously, it uses the detection of whether the prompt word pattern can be parsed by the current AI model and whether resource pointers are accessible as supplementary criteria for determining changes in execution conditions. This further expands the system's perception coverage of various interaction condition changes. These mechanisms enable the cloud server to proactively identify execution condition failures caused by interface version updates, parameter configuration adjustments, or resource path changes when handling multiple concurrent interaction requests. Without manual intervention, it automatically switches the expression layer to a binding version adapted to the current conditions, significantly reducing system maintenance costs and the risk of interaction interruptions.
[0017] (3) This invention generates state change identifiers by calculating the vector distance between the current comprehensive state vector and the historical state baseline vector using weighted Euclidean distance, enabling the system to represent the comprehensive evolution degree of the multi-dimensional state of the target object in a continuous numerical space. The scene adaptation state vector, cognitive state vector, and environmental dynamic state vector are matched with corresponding weight coefficients to achieve differentiated responses to the importance of different types of state changes. Based on this, the cognitive state vector is further configured as task comprehension state, operation proficiency state, and exploration willingness state. When only state change identifiers are generated but execution condition change identifiers are not generated, the field update mask only covers the content depth field and task form field corresponding to the cognitive state. When the identifier is generated simultaneously, the field update mask simultaneously covers the execution parameter field, content depth field, and task format field. This mechanism enables the system to accurately control the scope of expression layer adjustments according to preset field-level constraints, while precisely distinguishing the source of change. This avoids incorrect rewriting of technical execution fields such as interface paths or prompt word templates due to changes in the internal state of the target object, and also avoids incorrect adjustment of semantic fields such as content depth or task format due to changes in external execution conditions. This ensures that the cloud server's access to system resources and modification of fields during expression layer switching are always kept within a reasonable and controllable range, effectively preventing system anomalies caused by unauthorized field modifications, and significantly improving the robustness and predictability of the system in scenarios with multiple sources of change. Attached Figure Description
[0018] The present invention will be further described with reference to the accompanying drawings, but the embodiments in the drawings do not constitute any limitation on the present invention. For those skilled in the art, other drawings can be obtained based on the following drawings without creative effort.
[0019] Figure 1 This is a schematic diagram of the method flow of the present invention.
[0020] Figure 2 This is a flowchart illustrating the process of generating the switching trigger type involved in this invention. Detailed Implementation
[0021] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.
[0022] Reference Figure 1As shown, this invention provides an adaptive switching method for interactive content expression layers based on artificial intelligence scene state awareness, applied to a cloud server. The method includes: acquiring a stable content core and a current expression layer binding for the target interactive content; the stable content core and multiple expression layer bindings are stored separately; the expression layer bindings characterize the content depth, task form, and execution parameters of the stable content core in the current artificial intelligence application scenario, and are associated with and stored in relation to a historical state baseline vector generated when the expression layer binding takes effect; collecting observation events generated by the target object around the target interactive content in the artificial intelligence application scenario; and updating the scene adaptation state vector, cognitive state vector, and environmental dynamic state of the target object respectively through the observation events within the target time window. The system generates a current comprehensive state vector based on the updated state vector; detects the executability of the current expression layer binding to generate an execution condition change identifier, and compares the current comprehensive state vector with the historical state baseline vector associated with the current expression layer binding to generate a state change identifier. It then determines a switching trigger type based on the combination of the execution condition change identifier and the state change identifier. Finally, it matches a field update mask according to the switching trigger type, obtains a set of field changes for candidate expression layer bindings relative to the current expression layer binding, and verifies whether the set of field changes is completely covered by the field update mask. If so, it switches the current expression layer binding to the candidate expression layer binding while maintaining the stable content core unchanged.
[0023] This embodiment is mainly applied to cloud servers. By leveraging the high-concurrency computing power and massive storage resources provided by the cloud server, real-time expression layer switching services are provided for multiple target objects. The cloud server establishes a communication connection with the user terminal of the target object (such as a personal computer, tablet computer, or virtual reality device), receives observation events generated by the target object during the interaction process, and completes core logic such as state update and switching decision in the cloud. After that, the new expression layer binding and its corresponding interactive interface, parameter configuration, and other content are sent to the user terminal for presentation.
[0024] The stable content core and the current expression layer binding are obtained by accessing the storage module of the cloud server; the target interactive content refers to a relatively stable learning task in the talent cultivation scenario, such as parameter adjustment and result verification in artificial intelligence image generation; the stable content core is the unchanging component in the target interactive content, including learning objectives, ability structure, task structure, and key judgment points, etc. For example, if the stable content core is an image generation task, it includes key judgment points such as describing the input intent, adjusting the generation parameters, and comparing output differences; the expression layer binding is the executable implementation of the stable content core in a specific artificial intelligence application scenario, which specifically includes content depth fields, task form fields, and execution parameter fields; in addition, the stable content core and multiple expression layer bindings are stored separately, specifically, the stable content core and expression layer bindings are displayed and stored separately in the system configuration database of the cloud server as two independent data objects.
[0025] The current AI application scenario refers to the specific AI product environment currently used by the target object, such as a specific version of an AI image generation tool or an AI dialogue writing tool; the content depth field is used to represent the knowledge level of the current interactive content, such as phenomenon comparison or parameter relationship analysis; the task form field is used to represent the current learning activity type, such as follow-up operation or constrained practice; the execution parameter field includes at least one of interface path, operation steps, parameter mode, prompt word mode, and resource pointer.
[0026] When an expression layer binding passes validation (such as field range validation and coverage validation) and is officially switched to the current expression layer binding by the system, the binding is considered effective. Based on the most recent observed event at the current moment, the scene adaptation state vector, cognitive state vector, and environmental dynamic state vector of the target object are recalculated or updated. These three state vectors are combined into a multi-dimensional vector, and this combined multi-dimensional vector is finally stored in the database as an attribute of the new expression layer binding, serving as its exclusive historical state benchmark vector. Therefore, the historical state benchmark vector is a vector composed of and saved from the scene adaptation state vector, cognitive state vector, and environmental dynamic state vector of the target object at the moment when the expression layer binding is effective. It is used as a comparison benchmark for subsequent judgment of whether the state of the target object has changed significantly. The associated storage means storing each expression layer binding and its corresponding historical state benchmark vector as a whole, ensuring that the state snapshot when the binding is effective can be accurately obtained during comparison.
[0027] The target audience refers to individuals participating in learning, experiencing, or practicing activities within an AI application scenario. The observed events are collected by monitoring the target audience's operational behaviors and system feedback during the interaction process. These events can originate from one or more of the following: scenario experience events, task operation events, test question-and-answer events, questionnaire feedback events, content selection events, abandonment events, and result comparison events. Each observed event is represented by its event vector, and its event score (VKQ) is recorded. i Event credibility weight (GXZM) i ) and the relevance weight to the core of stable content (LKG) i The target time window is a configurable time range parameter. The system summarizes the collected observation events at this time window granularity. The time window can be dynamically adjusted according to different learning scenarios and evaluation needs. The scenario adaptation state vector includes interface operation accessibility state, parameter pattern resolvability state, and resource pointer availability state. The cognitive state vector includes task comprehension state, operation proficiency state, and exploration willingness state. The environmental dynamic state vector includes scenario interaction state, deep learning state, and vertical innovation state. Each state value in the above three state vectors is calculated by weighted averaging based on the scores of all observation events within the target time window, combined with their event credibility weight, time decay weight, and relevance weight of stable content core. The specific state value calculation method is: XQZ(d,k)=Σ i=1 N [VKQ i (d, k) × GXZM i ×HBD i ×LKG i ] / Σ i=1 N [GXZM i ×HBD i ×LKG i ], where d represents the state dimension, k represents the specific state under that dimension, for example, when d is the cognitive state dimension, k can be the task comprehension, so (d, k) can be understood as (cognitive state dimension, task comprehension), i is the number of the observed event, i=1, 2, 3, ..., N, N is the total number of observed events; XQZ(d, k) is the current quantified value of the specific state item k under a specific state dimension d, which is a value between 0 and 1, used to reflect the performance level or tendency intensity of the target object in a certain specific state; VKQ i (d, k) represents the score of the i-th observation event on state (d, k), GXZM i HBD is the confidence weight for the i-th observation event. iLKG represents the time decay weight of the i-th observation event. i The relevance weight between the i-th observed event and the core of the stable content is defined as follows: The current comprehensive state vector is a multi-dimensional vector composed of the updated scene adaptation state vector, the cognitive state vector, and the environmental dynamic state vector. This is achieved by concatenating the three updated state vectors in a fixed order into a longer vector, which comprehensively reflects the overall state of the target object within the current time window.
[0028] The system first generates an execution condition change identifier and a state change identifier. The execution condition change identifier is generated by detecting the executability of the currently bound expression layer; it is generated when the interface action sequence is unexecutable, the prompt word pattern is unparseable, or the resource pointer is inaccessible. The state change identifier is generated by calculating the weighted Euclidean distance between the current comprehensive state vector and the historical state baseline vector; it is generated when this distance exceeds a preset state change distance threshold. Based on different combinations of the above two identifiers, the system determines three switching trigger types, referencing... Figure 2 The diagram shows the flowchart for generating the switching trigger type involved in this invention. When only the execution condition change identifier is generated, it is determined to be an execution condition change type; when only the status change identifier is generated, it is determined to be a talent status change type; when both identifiers are generated simultaneously, it is determined to be a composite change type.
[0029] A field update mask is a configuration object used to limit the scope of field changes. It is determined based on the switching trigger type. Specifically, for execution condition change types, the field update mask only covers the execution parameter field; for talent status change types, the field update mask covers the content depth field and the task form field; for composite change types, the field update mask covers all fields in both categories. The candidate expression layer bindings are selected from multiple expression layer bindings stored in association with the stable content core. The system iterates through all expression layer bindings associated with the stable content core, calculates the field differences between each binding and the current expression layer binding, and thus obtains the set of field changes. This set is represented as a list of changes in field names and values, for example, {contentdepth: value 1 → value 2, taskform: value A → value B}. The verification process checks whether each field in the set of field changes is included within the field range defined by the field update mask. If any field in the set of field changes is not within the field update mask, the verification fails, meaning the set of field changes is not fully covered. When verification fails, the system does not perform automatic switching but generates a manual confirmation request, sending the candidate expression layer binding and the reason for the verification failure to the administrator for manual intervention and confirmation. Regardless of whether it is fully covered, the entire verification and decision-making process is executed on the cloud server. Ultimately, both the generated switching command and the manual confirmation request are sent from the cloud server to the corresponding user terminal or management terminal.
[0030] Specifically, the scenario adaptation state vector includes interface operation reachability state, parameter pattern resolvability state, and resource pointer availability state. The update method for the interface operation reachability state is as follows: The system detects whether the execution paths of each action in the interface action sequence of the current artificial intelligence application scenario are complete and reachable, counts and calculates the ratio of the number of actions with complete and reachable execution paths to the total number of actions, and determines the reachability status value of the interface operation based on the ratio.
[0031] It should be noted that the interface operation reachability state indicates whether all operation steps required to complete the predetermined task in the interactive interface of the current artificial intelligence application scenario can be executed smoothly; the parameter pattern resolvability state indicates whether the parameter configuration defined in the expression layer binding is compatible with the parameter interface of the current application scenario; and the resource pointer availability state indicates whether the external resources (such as model files, datasets, template files, etc.) referenced in the expression layer binding can be accessed and loaded normally. Detecting whether the execution path of each action in the interface action sequence is complete and reachable is achieved by traversing the interface action sequence defined in the current expression layer binding. The interface action sequence refers to a set of interactive actions that the target object needs to perform on the AI application interface to complete a learning task. For example, in an AI image generation task, the action sequence may include operations such as opening a parameter panel, selecting a model version, inputting prompts, and clicking the generate button. The execution path refers to the interface element hierarchy and navigation steps required to complete a specific action, starting from the current interface state. For example, completing the action of inputting prompts requires traversing the path of locating the input box, clicking to activate, and inputting text. Complete reachability means that in the current application interface, every interface element (such as a button, input box, or drop-down menu) on the execution path is visible and interactive, without being obscured, unusable, or missing. The system traverses each action in the interface action sequence, records the number of actions whose execution paths are completely reachable, and then divides this number by the total number of actions. The resulting ratio is the interface operation reachability status value, which is between 0 and 1; the closer to 1, the smoother the interface operation. The process of counting the number of fully reachable actions and calculating the ratio described above is the update of the accessibility status of the interface operation.
[0032] Furthermore, when the reachability status value of the interface operation is lower than a preset executable threshold, the number of unexecutable actions in the interface action sequence bound to the current expression layer is detected, and when the number of unexecutable actions changes, the execution condition change identifier is generated.
[0033] It should be noted that the preset executable threshold is a value between 0 and 1, used to determine whether the accessibility of the interface operation is within an acceptable range. This threshold is pre-stored in the system configuration database of the cloud server and can be configured and adjusted by the administrator according to the characteristics of different artificial intelligence application scenarios. Detecting the number of unexecutable actions refers to traversing the interface action sequence, performing accessibility checks on the execution path of each action, and counting the number of actions whose execution paths are unreachable. Unexecutability means that a certain step of the action cannot be completed because the interface element is invisible, unclickable, or the function is unavailable. A change refers to an increase or decrease in the number of unexecutable actions within the current observation time window compared to the number of unexecutable actions in the previous time window.
[0034] Specifically, detecting the executability of the currently bound expression layer to generate an execution condition change flag also includes: The system detects whether the prompt word pattern bound to the current expression layer can be parsed by the current artificial intelligence model, and detects whether the resource pointer bound to the current expression layer is accessible; when the prompt word pattern is unparsable or the resource pointer is inaccessible, the system generates the execution condition change flag.
[0035] It should be further explained that "resource pointer inaccessible" means that the external storage location (such as file server address, database table name, model storage path, etc.) pointed to by the resource pointer in the expression layer binding cannot be successfully read or loaded under the current network environment and permission configuration. When the reachability status value of the interface operation is detected to be lower than the preset executable threshold, or the prompt word pattern is detected to be unresolvable, or the resource pointer is detected to be inaccessible, the system will automatically generate the execution condition change flag to indicate that the external execution conditions on which the current expression layer binding depends have changed or are no longer satisfied.
[0036] Furthermore, the detection method for whether the prompt word pattern can be parsed by the current artificial intelligence model is as follows: The prompt word template in the current expression layer binding is input into the parser of the current artificial intelligence model to obtain the syntax tree structure returned by the parser. If there are unrecognized syntax nodes in the syntax tree structure, the prompt word pattern is determined to be unparseable.
[0037] It should also be explained that the prompt word pattern refers to the structured specification of prompt words defined in the expression layer binding for interaction with the artificial intelligence model. It defines the format, placeholders, variables, and invocation methods of the prompt words. The current artificial intelligence model is a large language model or multimodal model used to process interactive requests from target object inputs and generate corresponding content. It obtains the prompt word text input by the user as model input, segments and encodes the prompt word text, and then inputs it into the model's neural network structure for computation. Based on its pre-trained parameters, the neural network structure calculates and generates corresponding text, images, or other modal content as model output. The prompt word template is a specific string containing fixed text and variable placeholders, for example: Please generate an image based on the following description: {description}, style {style}. The parser is a component in the artificial intelligence model used to convert the original prompt word text into a structured representation that can be understood by the model's internal computational units. It is responsible for lexical analysis, syntactic analysis, and semantic analysis of the prompt words, generating the corresponding syntax tree structure. Determining whether there are unrecognized syntax nodes in the syntax tree structure is done by comparing the syntax tree nodes generated by the parser with the standard syntax node set supported by the artificial intelligence model. If a node is found to be not in the standard set, the node is marked as unrecognized, and at this time, the prompt word pattern is determined to be unparseable.
[0038] Specifically, the current integrated state vector is compared with the historical state baseline vector bound to the current expression layer to generate a state change identifier. The specific process is as follows: Calculate the vector distance between the current comprehensive state vector and the historical state baseline vector. When the vector distance exceeds a preset state change distance threshold, generate the state change identifier. The vector distance is calculated using weighted Euclidean distance, that is, the scene adaptation state vector, the cognitive state vector, and the environmental dynamic state vector are respectively matched with corresponding weight coefficients.
[0039] In the above, the preset state change distance threshold is a non-negative value used to measure whether the current comprehensive state of the target object has changed significantly relative to the historical baseline state. This threshold is pre-stored in the system configuration database of the cloud server. The specific process of generating the state change identifier is as follows: First, the weighted Euclidean distance between the current comprehensive state vector and the historical baseline vector is calculated, and then the distance is compared with the preset state change distance threshold. When the distance is greater than the threshold, it is determined that the state of the target object has changed significantly, and the system automatically generates the state change identifier. The vector distance is calculated based on the weighted Euclidean distance algorithm, which assigns different weight coefficients to the scene-adapted state vector, the cognitive state vector, and the environmental dynamic state vector to reflect the differentiated impact of different state dimensions on the sensitivity of overall state changes. Specifically, the system compares the state values of the two vectors (current comprehensive state vector and historical baseline vector) at the same position one by one. For each state dimension, the system calculates the difference between the current value and the historical baseline value, and then squares this difference to amplify the significance of the difference. After obtaining the squared difference for each dimension, the system multiplies the different state dimensions by a preset value. The weighting coefficients are set first. The significance of these weighting coefficients is that the system can assign different influences to different state dimensions based on their importance to the overall judgment. For example, if it is believed that changes in cognitive state reflect the substantial progress of the target object more than changes in environmental state, then the dimensions related to cognitive state will be given higher weights. After completing the weighted square calculation of all dimensions, the system adds all the results to obtain a sum. Finally, the square root of this sum is taken. This final value is the weighted Euclidean distance between the two vectors. It comprehensively reflects the overall degree of change of the target object's current state in all dimensions compared to the historical baseline state. The larger the value, the more significant the state change.
[0040] Furthermore, the cognitive state vector includes task comprehension state, operation proficiency state, and exploration willingness state; when only the state change identifier is generated but the execution condition change identifier is not generated, the field update mask covers the content depth field and task form field corresponding to the task comprehension state, the operation proficiency state, and the exploration willingness state.
[0041] The task comprehension state characterizes the target object's mastery of the current learning task objectives and core knowledge points; the operation proficiency state characterizes the target object's familiarity with the operation process and interface interaction of the current artificial intelligence application tool; the exploration willingness state characterizes the target object's tendency to actively try new parameters, new methods, or new application scenarios beyond completing basic tasks. The condition under which the execution condition change indicator is not generated is: the interface action sequence bound to the current expression layer is fully reachable, the prompt word pattern can be parsed, and the resource pointer is accessible, meaning the current expression layer is still valid on the external execution conditions. The overriding method of the field update mask refers to including the content depth field and the task form field in the set of fields that can be changed; the content depth field defines the knowledge depth level of the current learning content, such as: basic concept understanding, parameter relationship analysis, or cross-scenario innovative application; the task form field defines the specific form of the current learning activity, such as: guided operation, problem-based exploration, or open-ended practice. The system uses the values of the task comprehension status and the operation proficiency status to jointly determine the value of the content depth field (e.g., when both comprehension and proficiency are high, the content depth is increased to parameter relationship analysis), and uses the values of the operation proficiency status and the exploration willingness status to jointly determine the value of the task format field (e.g., when proficiency is high and exploration willingness is strong, the task format is switched to open practice).
[0042] Specifically, when the execution condition change identifier and the state change identifier are generated simultaneously, the field update mask covers the execution parameter fields corresponding to the interface operation reachability state, the parameter pattern parseability state, and the resource pointer availability state, respectively. The field update mask also covers the content depth field and task form field corresponding to the task comprehension state, the operation proficiency state, and the exploration willingness state.
[0043] It should be explained that the way the field update mask covers the execution parameter field is by including specific execution parameters such as interface path, operation steps, parameter mode, prompt word mode, and resource pointer into a set of fields that can be changed. The execution parameter field is a set of fields in the expression layer binding used to control the details of interaction with the artificial intelligence application environment. It directly determines the specific way the target object interacts with the current artificial intelligence product, such as the jump logic of the operation path, the layout of the parameter input box, and the format specifications of the prompt words. When the composite change type is triggered, the field update mask covers the execution parameter field, the content depth field, and the task form field simultaneously. The logic of this coverage process is similar to that of the talent status change type mentioned above. The difference is that the field update mask corresponding to the composite change type includes the execution parameter field as well as the content depth and task form fields. That is, it allows the candidate expression layer binding to change the execution parameters and content / task form at the same time to adapt to the situation where the execution conditions and talent status change simultaneously.
[0044] Furthermore, the method for obtaining the candidate expression layer binding is as follows: From multiple expression layer bindings associated with the core stable content, expression layer bindings that match the content depth field and the task form field with the task comprehension state, the operation proficiency state, and the exploration willingness state are selected, and the selected expression layer bindings are used as candidate expression layer bindings.
[0045] In this embodiment, "matching" specifically refers to the values of the content depth field and task format field bound to the candidate expression layer being compatible with the target content level and target task type determined based on the task comprehension state, operation proficiency state, and exploration willingness state. Specifically, the system determines the expected content depth value and expected task format value corresponding to the current state based on the specific values of each state dimension in the cognitive state vector, using a preset state-field mapping rule. Then, it searches among multiple expression layer bindings for the binding whose content depth field and task format field are consistent with or closest to the expected values as the candidate expression layer binding.
[0046] The above description is merely an example and illustration of the structure of the present invention. Those skilled in the art can make various modifications or additions to the specific embodiments described, or use similar methods to replace them, as long as they do not deviate from the structure of the invention or exceed the scope defined in the claims, all of which should fall within the protection scope of the present invention.
Claims
1. An adaptive switching method for interactive content expression layer based on artificial intelligence scene state perception, applied to a cloud server, characterized in that: include: The stable content core of the target interactive content is obtained and bound to the current expression layer. The stable content core is stored separately from multiple expression layer bindings. The expression layer bindings are used to characterize the content depth, task form and execution parameters of the stable content core in the current artificial intelligence application scenario, and are associated with the historical state benchmark vector generated when the expression layer bindings take effect. Collect observation events generated by the target object around the target interaction content in the artificial intelligence application scenario, update the scene adaptation state vector, cognitive state vector and environmental dynamic state vector of the target object respectively through the observation events within the target time window, and generate the current comprehensive state vector based on the updated state vector; The executability of the current expression layer binding is detected to generate an execution condition change identifier, and the current comprehensive state vector is compared with the historical state baseline vector associated with the current expression layer binding to generate a state change identifier. The switching trigger type is determined based on the combination of the execution condition change identifier and the state change identifier. Based on the field update mask of the switching trigger type, obtain the set of field changes of the candidate expression layer binding relative to the current expression layer binding, and verify whether the set of field changes is completely covered by the field update mask. If so, switch the current expression layer binding to the candidate expression layer binding, and keep the core of the stable content unchanged.
2. The adaptive switching method for interactive content expression layer based on artificial intelligence scene state awareness according to claim 1, characterized in that: The scenario adaptation state vector includes interface operation reachability state, parameter pattern resolvability state, and resource pointer availability state. The update method for the interface operation reachability state is as follows: The system detects whether the execution paths of each action in the interface action sequence of the current artificial intelligence application scenario are complete and reachable, counts and calculates the ratio of the number of actions with complete and reachable execution paths to the total number of actions, and determines the reachability status value of the interface operation based on the ratio.
3. The adaptive switching method for interactive content expression layer based on artificial intelligence scene state awareness according to claim 2, characterized in that: When the reachability status value of the interface operation is lower than the preset executable threshold, the number of unexecutable actions in the interface action sequence bound to the current expression layer is detected, and when the number of unexecutable actions changes, the execution condition change flag is generated.
4. The adaptive switching method for interactive content expression layer based on artificial intelligence scene state awareness according to claim 3, characterized in that: Detecting the executability of the currently bound expression layer to generate an execution condition change flag also includes: Detect whether the prompt word pattern bound to the current expression layer can be parsed by the current artificial intelligence model, and detect whether the resource pointer bound to the current expression layer is accessible; The execution condition change flag is generated when the prompt word pattern cannot be parsed or the resource pointer is inaccessible.
5. The adaptive switching method for interactive content expression layer based on artificial intelligence scene state awareness according to claim 4, characterized in that: The method for detecting whether the prompt word pattern can be parsed by the current artificial intelligence model is as follows: The prompt word template in the current expression layer binding is input into the parser of the current artificial intelligence model to obtain the syntax tree structure returned by the parser. If there are unrecognized syntax nodes in the syntax tree structure, the prompt word pattern is determined to be unparseable.
6. The adaptive switching method for interactive content expression layer based on artificial intelligence scene state awareness according to claim 1, characterized in that: The current integrated state vector is compared with the historical state baseline vector associated with the current expression layer to generate a state change identifier. The specific process is as follows: Calculate the vector distance between the current integrated state vector and the historical state baseline vector, and generate the state change identifier when the vector distance exceeds a preset state change distance threshold. The vector distance is calculated using weighted Euclidean distance, meaning that the scene adaptation state vector, the cognitive state vector, and the environmental dynamic state vector are each matched with corresponding weight coefficients.
7. The adaptive switching method for interactive content expression layer based on artificial intelligence scene state awareness according to claim 6, characterized in that: The cognitive state vector includes task comprehension state, operational proficiency state, and exploration willingness state; When only the state change identifier is generated but the execution condition change identifier is not generated, the field update mask covers the content depth field and task form field corresponding to the task comprehension state, the operation proficiency state, and the exploration willingness state.
8. The adaptive switching method for interactive content expression layer based on artificial intelligence scene state awareness according to claim 7, characterized in that: When the execution condition change identifier and the state change identifier are generated simultaneously, the field update mask covers the execution parameter fields corresponding to the interface operation reachability state, parameter pattern resolvability state, and resource pointer availability state, respectively. The field update mask also covers the content depth field and task form field corresponding to the task comprehension state, the operation proficiency state, and the exploration willingness state.
9. The adaptive switching method for interactive content expression layer based on artificial intelligence scene state awareness according to claim 8, characterized in that: The method for obtaining the candidate expression layer binding is as follows: From multiple expression layer bindings associated with the core stable content, expression layer bindings that match the content depth field and the task form field with the task comprehension state, the operation proficiency state, and the exploration willingness state are selected, and the selected expression layer bindings are used as candidate expression layer bindings.