Personalized model construction method and system applied to intelligent interaction

Abstract

The application provides a personalized model construction method and system applied to intelligent interaction, relates to the technical field of intelligent interaction, and first generates a scene intention linkage basic logic based on the linkage relationship between scene elements (including interactive environment elements, user operation elements and interactive target elements) of an intelligent interaction scene and user interaction demands (including information acquisition demands, function use demands, opinion expression demands and emotional communication demands); then constructs a feedback collection nested link, collects related feedback information to generate a scene feedback linkage set; then constructs a dynamic adaptation iteration link with the scene feedback linkage set as a driving source, generates a dynamic adaptation optimization link; then constructs a collaborative operation process of a personalized model architecture and the dynamic adaptation optimization link, generates a complete personalized model architecture; and finally outputs a scene adaptation type intelligent interaction response instruction through the complete personalized model architecture, so as to realize personalized adaptation of intelligent interaction.

Description

Technical Field

[0001] This application relates to the field of intelligent interaction technology, and more specifically, to a method and system for constructing personalized models for intelligent interaction. Background Technology

[0002] With the continuous development of intelligent interaction technology, intelligent interaction systems are widely used in smart home scenarios. However, existing intelligent interaction systems often have some significant shortcomings in handling user interactions. On the one hand, traditional intelligent interaction systems tend to have a simplistic approach to intelligent interaction scenarios, typically focusing only on some explicit scene elements, such as simple interactive environment information, while neglecting the complex impact of user operation elements and interaction target elements on user interaction needs. This makes it difficult to comprehensively and accurately grasp the intrinsic relationship between scene elements and user interaction needs. On the other hand, in terms of obtaining user feedback, existing technologies are mostly limited to collecting direct feedback information. They lack effective means of collecting indirect feedback information and information on the adaptation and correlation between scene elements and intelligent interaction responses, resulting in an inability to fully and deeply understand users' true feelings and needs regarding intelligent interaction. Furthermore, existing intelligent interaction systems lack dynamic adaptation and optimization mechanisms, making it difficult to adjust interaction strategies in a timely manner according to real-time changes in scene elements and user interaction needs. This prevents the realization of truly personalized intelligent interaction, and the user experience needs further improvement. Summary of the Invention

[0003] In view of this, the purpose of this application is to provide a method and system for constructing personalized models for intelligent interaction.

[0004] According to a first aspect of this application, a method for constructing a personalized model for intelligent interaction is provided, the method comprising: This process extracts scene elements and user interaction requests from intelligent interaction scenarios, analyzes the bidirectional linkage between scene elements and user interaction requests, and generates basic logic for scene intent linkage. The scene elements include interaction environment elements, user operation elements, and interaction target elements. The user interaction requests include information acquisition requests, function usage requests, opinion expression requests, and emotional communication requests. The bidirectional linkage is the triggering correspondence between scene elements and user interaction requests, and the adaptation requirement relationship between user interaction requests and scene elements. The basic logic for scene intent linkage is an executable logic set containing scene element combination rules, user interaction request matching rules, and intelligent interaction response invocation rules, used to match corresponding user interaction requests based on scene element combinations and generate adapted intelligent interaction responses. A nested feedback collection chain is constructed, embedding the basic logic of scene intent linkage into the core execution node of the nested feedback collection chain. Through the layered collection channels of the nested feedback collection chain, direct feedback information and indirect feedback information of users to intelligent interaction responses, as well as the adaptation and association information between scene elements and intelligent interaction responses, are collected to generate a scene feedback linkage set. The nested feedback collection chain is a layered data collection chain containing a direct feedback collection layer, an indirect feedback collection layer, and an adaptation status collection layer. Each layer is equipped with an independent collection channel and cross-layer data association rules. The scene feedback linkage set is a structured data set containing user feedback content, scene element adaptation status, feedback background information, and feedback type identifier. Using the aforementioned scene feedback linkage set as the driving source, a dynamic adaptation iteration link is constructed. Through multiple rounds of closed-loop iteration of the dynamic adaptation iteration link, the scene element combination rules and user interaction request matching rules in the basic logic of scene intent linkage are adjusted to generate a dynamic adaptation optimization link. The dynamic adaptation iteration link is a closed-loop iteration link that includes an adjustment instruction generation module, a logic modification execution module, an optimization effect analysis module, and a feedback loop input module. The dynamic adaptation optimization link is an executable iterative optimization logic set that includes the iteratively optimized scene intent linkage logic, iterative adjustment rules, and a feedback driving mechanism. A collaborative operation process is constructed between the personalized model architecture and the dynamic adaptation and optimization link. The adjustment logic of the dynamic adaptation and optimization link is embedded into the corresponding hierarchical structure of the personalized model architecture to generate a complete personalized model architecture. The complete personalized model architecture is a hierarchical model architecture that includes a linkage adaptation layer, a feedback processing layer, and an iterative optimization layer. Preset data transmission rules, triggering conditions, and interaction methods are set between each layer. The linkage adaptation layer is used to generate intelligent interactive responses, the feedback processing layer is used to process feedback data to generate adjustment driving data, and the iterative optimization layer is used to update the core logic of the model. Through the complete architecture of the personalized model, real-time scene elements and user interaction requests are received, and scene-adaptive intelligent interaction response instructions are output. The scene-adaptive intelligent interaction response instructions dynamically adjust the response content, response rhythm and interaction path of the intelligent interaction according to the real-time scene elements and user interaction requests, so as to complete the personalized adaptation of the intelligent interaction.

[0005] According to a second aspect of this application, a personalized model building system for intelligent interaction is provided. The personalized model building system for intelligent interaction includes a machine-readable storage medium and a processor. The machine-readable storage medium stores machine-executable instructions. When the processor executes the machine-executable instructions, the personalized model building system for intelligent interaction implements the aforementioned personalized model building method for intelligent interaction.

[0006] Based on any of the above aspects, the technical effect of this application is as follows: By generating the basic logic of scene intent linkage based on the interaction relationship between scene elements and user interaction requests in intelligent interaction scenarios, this paper depicts the triggering effect of scene elements on user interaction requests and the adaptation requirements of user interaction requests to scene elements. A nested feedback collection link is constructed, embedding the basic logic of scene intent linkage within it. This allows for the layered collection of direct and indirect user feedback information to intelligent interaction responses, as well as the adaptation correlation information between scene elements and intelligent interaction responses. A scene feedback linkage set is generated, and a dynamic adaptation iteration link is constructed using this set as the driving source. Through multiple rounds of iteration, the combination rules of scene elements and the matching logic of user interaction requests in the basic logic of scene intent linkage are adjusted, generating a dynamic adaptation optimization link. This enables the intelligent interaction system to continuously optimize and adjust itself based on real-time data. A collaborative operation process between a personalized model architecture and the dynamic adaptation optimization link is constructed. The adjustment logic of the dynamic adaptation optimization link is embedded into the hierarchical structure of the personalized model architecture, generating a complete personalized model architecture including a linkage adaptation layer, a feedback processing layer, and an iterative optimization layer. This achieves end-to-end personalized intelligent interaction from scene perception and feedback processing to dynamic optimization. Ultimately, the personalized model complete architecture outputs scenario-adaptive intelligent interactive response commands. Based on real-time scenario elements and user interaction needs, the response content, response rhythm, and interaction path of the intelligent interaction are dynamically adjusted, which can provide users with a highly personalized, accurate, and smooth intelligent interactive experience, significantly improving the performance of the intelligent interaction system and user satisfaction. Attached Figure Description

[0007] Figure 1 A flowchart illustrating the personalized model construction method for intelligent interaction provided in an embodiment of this application is shown. Figure 2 This paper illustrates a schematic diagram of the component structure of a personalized model building system for intelligent interaction provided in an embodiment of this application. Detailed Implementation

[0008] Figure 1 This application provides a flowchart illustrating a personalized model construction method and system for intelligent interaction, with detailed steps including: Step S110: Extract scene elements and user interaction requests from the intelligent interaction scenario, analyze the two-way linkage relationship between scene elements and user interaction requests, and generate the basic logic for scene intent linkage.

[0009] In this embodiment, all quantifiable raw data within the current and historical interaction periods can be collected from the system log database, sensor data bus, and user operation log files. These data are initially divided according to preset classification dimensions, forming raw data streams for interactive environment elements, user operation elements, and interactive target elements. A natural language understanding engine performs deep semantic analysis on the user's voice or text input to extract the raw data stream of user interaction requests. A bidirectional linkage analysis matrix is ​​constructed, where the row indices represent various combinations of scene elements, and the column indices represent various categories of user interaction requests. Statistical analysis methods are used to calculate the co-occurrence frequency and conditional probability of each cell in the matrix, thereby quantifying the triggering strength of scene elements for user interaction requests and the adaptation strength of user interaction requests to scene elements. The above analysis results are encoded into a set of executable logical rules. These rules collectively constitute the basic logic for scene intent linkage. This basic logic can receive real-time scene elements as input and output one or more predicted user interaction requests and a preliminary response strategy framework.

[0010] Step S111: Collect various environmental information related to user interaction behavior in intelligent interaction scenarios, filter and integrate them to form a set of interactive environment elements, bind each interactive environment element with the corresponding scene presentation details, and generate a set of interactive environment element detail descriptions.

[0011] Data is periodically retrieved from various data sources through an internally defined standard data access interface. This includes light intensity sequences in lux from light sensor nodes, temperature sequences in degrees Celsius and relative humidity sequences in percentages from temperature and humidity sensor nodes, noise intensity sequences in decibels from noise sensor nodes, time-coded values ​​in hours and minutes from a clock module, and textual descriptions of weather conditions from a network interface. These raw data streams from different sources with varying dimensions and update frequencies are aligned and encapsulated according to a unified timestamp to form an intermediate representation of the interactive environment elements. For example, the aligned data is encapsulated into a structure containing fields such as timestamp, light intensity array, temperature array, humidity array, noise array, time code, and weather description. Next, feature extraction and filtering are performed on the sequence data within this structure. The mean, variance, and trend slope of the light intensity array are calculated as the specific values ​​for the interactive environment elements. Similarly, the mean, variance, and trend slope of the temperature, humidity, and noise arrays are calculated to obtain the corresponding statistical parameters. The time encoding is converted into time period types, such as discrete categories like early morning, morning, noon, afternoon, evening, and late night. Weather descriptions are categorized into standard weather types like sunny, cloudy, rainy, and snowy. After the above processing, an interactive environment element set is generated, containing elements such as average light intensity, light intensity variance, light variation trend, average temperature, average noise, time period category, and weather category. Finally, each element in the interactive environment element set is bound to the original scene presentation details that generated that element. For example, for the average light intensity element, it is bound to the statistical feature description of the original light intensity sequence, such as "the main light in the living room is on, the natural light incident angle is the current solar altitude angle, and the illuminance distribution curve shows a bimodal characteristic." All elements are bound to their detailed descriptions to form an interactive environment element detailed description set, where each entry is stored in key-value pairs, with the key being the element identifier and the value being a combination of structured detailed description text and statistical feature values.

[0012] Step S112: Record various operation-related information generated by the user during the intelligent interaction process, filter and integrate them to form a set of user operation elements, associate each user operation element with the corresponding interactive action scenario, and generate a set of user operation element scenario associations.

[0013] A global event listener registered at the operating system kernel layer captures all low-level events triggered by the user and interacting with the smart home system. Captured events include: physical button press events on the touch panel, containing the button identifier and press duration; voice wake-up word detection events, containing the wake-up word audio feature vector and wake-up confidence score; voice command recognition completion events, containing the recognized text and domain classification results; interface switching events on the mobile application, containing the source and target interface identifiers; control operation events on the application, containing the control identifier and operation type such as click, swipe, long press; and complete rounds of dialogue between the user and the virtual assistant, containing the user's spoken text and the assistant's response text. The captured raw event stream is grouped according to user sessions, each session containing the complete interaction process from the start of a wake-up to the end of the dialogue or timeout. For each session, user operation elements are extracted. The extracted elements include: interaction initiation method (voice wake-up, physical button, application opening, etc.); operation command sequence (a time-ordered list of commands, each containing command type and parameters); stage response duration vector (recording the time interval between the end of the previous system response and the start of the user's next command); repeated operation identifier (marking whether the user repeated the same or similar command content within the same session or between adjacent sessions); and operation interruption point (recording at which stage the user interrupted the interaction process). Each user operation element is associated with the interaction action scenario in which it was generated. For example, a specific operation command sequence is associated with the interaction action scenario of "starting movie viewing mode," recording the specific command sequence in this scenario as: closing curtains, adjusting lights, turning on the TV and switching input sources. Through the above association, a user operation element scenario association set is generated, where each entry contains the specific value or category of the user operation element and the associated interaction action scenario identifier.

[0014] Step S113: Collect information related to various goals that users expect to achieve during the intelligent interaction process, filter and integrate them to form a set of interaction goal elements, bind each interaction goal element with the corresponding interaction purpose, and generate a set of interaction goal element purpose associations.

[0015] By analyzing user interaction content for intent and inferring context, information related to the user's desired goals is collected. For each user interaction, the core goal category is extracted from the intent analysis results output by the natural language understanding module. Goal categories are divided into four main categories: information acquisition, function control, opinion expression, and emotional interaction. For information acquisition goals, the specific information query object and query dimension are further analyzed; for example, if the query object is "weather forecast," the query dimension is "probability of precipitation in the next three days." For function control goals, the control object and control parameters are further analyzed; for example, if the control object is "living room air conditioner," the control parameter is "switching the working mode to cooling and setting the target temperature to a certain value." For opinion expression goals, the evaluation object and emotional tendency are further analyzed; for example, if the evaluation object is "air purifier operating noise," the emotional tendency is "negative." For emotional interaction goals, the emotional need type is further analyzed; for example, "seeking companionship," "sharing joy," and "expressing fatigue." All collected goal information is filtered and integrated, removing vague or unquantifiable goals to form a set of interaction goal elements, which includes: information acquisition goals, function control goals, opinion expression goals, and emotional interaction goals. Then, each interaction target element is bound to a specific interaction purpose. For example, for a function control target element, it is bound to the specific interaction purpose "set the bedroom air conditioner temperature to a certain value"; for an information acquisition target element, it is bound to the specific purpose "query the current playlist of the living room smart speaker". Through this process, a set of interaction target element purpose associations is generated, in which each entry precisely describes the specific thing that the user ultimately wants to achieve for the target element, including target category and purpose description fields.

[0016] Step S114: Determine the interaction relationships among the set of detailed descriptions of interactive environment elements, the set of scene associations of user operation elements, and the set of purpose associations of interactive target elements, and generate a total set of scene element associations.

[0017] After obtaining three sets—a detailed description set of interactive environment elements, a scenario-related set of user operation elements, and a purpose-related set of interactive target elements—we began analyzing the complex interactions between them. First, we analyzed how interactive environment elements influence user operation elements. By comparing user operation data under different environments, we identified whether the distribution of user operation elements changed significantly when the values ​​of certain environmental elements changed. For example, when the environmental noise level element was high, the proportion of voice wake-up decreased while the proportion of physical buttons increased in the distribution of interaction activation method elements. This relationship was recorded as a rule: if the environmental noise level element value is in the high noise range, the probability of the interaction activation method element being voice wake-up decreases below a certain threshold. Next, we analyzed the driving path of user operation elements on interactive target elements. By analyzing cases of successfully achieving a certain type of goal, we found a strong correlation between the order and combination of certain operation command sequences and the efficiency of goal achievement. For example, in functional control goals, the sequence of first querying the device status and then issuing control commands has a higher success rate than the sequence of directly issuing control commands. This relationship was recorded as a path rule: goal achievement depends on the standardized execution order of operation command sequences. Finally, the adaptation requirements of the interactive target elements to the interactive environment elements are analyzed. For example, when the target element is an emotional interaction type such as "entering sleep mode," the adaptation requirements for environmental elements are: the light intensity should be reduced to a low brightness range, and the noise level should be reduced to a quiet range. This requirement is recorded as an adaptation rule: after the target element type is determined, the expected values ​​of environmental elements should fall within a specific numerical range or category interval. All identified influence methods, driving paths, and adaptation requirements are summarized, deduplicated, and structured to form a comprehensive set of scene element associations. This set describes the dynamic, multi-directional association network between scene elements. Each association entry includes a source element identifier, a target element identifier, an association type, a quantitative value of association strength, and a description of the association rule.

[0018] Step S115: Collect various user interaction requests generated by users during the intelligent interaction process, describe the specific content of each user interaction request in detail, and generate a detailed set of user interaction requests.

[0019] Using natural language processing (NLP) technology, intent and sentiment analysis are performed on user voice or text input, and combined with the interaction context, four main categories of requests are identified. For information acquisition requests, the specific content is described as "the user requests attribute information, status information, or knowledge information about a specific entity or event." For function usage requests, the detailed description is "the user instructs to perform status queries, parameter adjustments, mode switching, or start / stop control on a specific device or service." For opinion expression requests, the detailed description is "the user expresses affirmation, criticism, suggestions, or complaints about the quality, effect, or feeling of a specific entity, service, or interaction process." For emotional communication requests, the detailed description is "the user's emotional state expressed at a specific moment, hoping to receive an emotional response, resonance, or care." All identified requests and their detailed descriptions are organized to form a detailed set of user interaction requests. Each entry includes a request identifier, request category, request subcategory, request content description field, and sentiment tendency tag.

[0020] Step S116: Based on the overall set of scene element associations and user interaction requests, construct scene request response rules and generate basic logic for scene intent linkage.

[0021] This step, based on the already generated set of associated scene elements and detailed set of user interaction requests, constructs the final basic logic for scene intent linkage. The construction process is explained in detail below through its sub-steps.

[0022] Step S1161: Analyze the correspondence between various scene element combinations in the overall scene element association set and each user interaction request in the detailed user interaction request set, determine the scene element combination form that can trigger each user interaction request, and generate a scene element request trigger set.

[0023] The process iterates through all possible combinations of scene elements in the overall set of scene element associations. For example, consider a combination where the environmental elements include a time of day (evening), low light intensity variance, and low noise level; the operational elements include a voice activation method, an empty command sequence, and an undefined target element. This combination is then matched against each request in the detailed set of user interaction requests. The analysis reveals that this combination is most frequently associated with emotional communication requests such as "seeking companionship" or "seeking security confirmation," thus identifying this scene element combination as a typical trigger for these requests. Similar analysis is performed on all possible element combinations, recording which specific user interaction requests(s) each combination can trigger. Finally, all identified correspondences are organized into a mapping table, i.e., the scene element request trigger set, where each entry is a binary tuple containing a scene element combination identifier and a list of triggerable user interaction requests.

[0024] Step S1162: Based on historical interaction data, generate interaction response logic rules corresponding to each scenario element demand trigger combination to form a scenario demand response rule set.

[0025] A large amount of historical interaction data is retrieved from the local database. For each combination in the scene element request trigger set generated in step S1161, the interaction response process executed when that combination appeared in history is filtered out, and effective response rules are extracted from it. This process is achieved through the following sub-steps.

[0026] For example, step S11621: Filter the interaction cases in the historical interaction data that correspond to the combination of scene element requests, extract the complete response process in each case, including the response start node, intermediate execution links, and termination completion node, and generate a set of historical response processes.

[0027] Taking a specific trigger combination as an example, the historical database is searched for all interaction cases occurring under similar combinations. For each case, its complete response flow is extracted. Response initiation nodes include events such as wake-up engine activation and user command reception confirmation. Intermediate execution stages include a series of atomic operations and their sequential relationships, such as intent parsing, status query, decision generation, and command issuance. Termination completion nodes include events such as device execution success feedback, user satisfaction exit, and automatic termination upon timeout. The flows of all cases are extracted to form a historical response flow set. Each flow record includes a flow identifier, the corresponding trigger combination identifier, the initiation node timestamp, the event sequence of each stage, and the termination node timestamp.

[0028] Step S11622: Split the operation instruction sequence in each historical response process, record the sequential relationship between instructions, the allocation of execution time, and the constraints of dependency conditions, and generate a split instruction sequence set.

[0029] For each extracted historical response process, it is further broken down into atomic-level operation instructions. For example, a process is broken down into an instruction sequence: wake-up engine listens, speech recognition module converts audio to text, natural language understanding module parses user intent, dialogue management module determines the action to be executed, device control module queries device status, device control module generates control parameters, device control module sends instructions to the target device, and speech synthesis module generates feedback speech. The sequential relationships between these instructions are recorded, such as whether an instruction can only be executed after another instruction has succeeded. The execution time allocation for each step is also recorded, such as the time from the end of one instruction to the end of another. Dependency constraints are recorded, such as whether the execution of an instruction depends on the target device being online and having normal communication. Through the above method, a detailed instruction sequence breakdown record is generated for each historical response process. All records are summarized to form an instruction sequence breakdown set, with each record containing a process identifier, instruction identifier, preceding instruction identifier, execution time, and dependency description.

[0030] Step S11623: Extract the response content components from each historical response process, including text information units, function call units, and data display units, record the combination order and association method of each unit, and generate a response content component set.

[0031] Analyze the content ultimately presented to the user for each historical response process. The response content consists of: text information units, i.e., the feedback text read by the voice assistant; function call units, i.e., the actual control commands sent to the device; and data display units, i.e., the charts, lists, or status indicators displayed on the application interface. Record the order in which these units are combined, for example, text information is read first, followed by function call. Record the association methods, for example, a parameter in the text information is obtained from the control parameters in the function call unit. Summarize the response content records for all cases to form a response content composition set. Each record includes a process identifier, unit type, unit content summary, unit sequence index, and associated parameter identifier.

[0032] Step S11624: Analyze the connection characteristics of interactive links in the historical response process, including link switching trigger conditions, exception handling rules, and user confirmation node settings, and generate a set of link connection characteristics.

[0033] Further analysis of the dynamic characteristics of the process is conducted. For example, the switch from the "intent resolution" stage to the "device control" stage is triggered when the intent resolution confidence level is higher than a preset threshold. If the confidence level is lower than this threshold, the process will enter the "user confirmation" stage, where the system will ask for confirmation. This reflects the setting of stage switching trigger conditions and user confirmation nodes. As another example, if the device control module detects that the target device is offline, it will trigger the exception handling rule, and the process will enter the "report exception" stage, informing the user that the device cannot connect, instead of continuing to send commands. Similar stage connection features from all historical response processes are extracted to form a stage connection feature set. Each record includes the source stage identifier, target stage identifier, trigger condition type, trigger condition parameters, exception handling type, and user confirmation node location.

[0034] Step S11625: Compare multiple historical response processes corresponding to the same scenario element demand trigger combination, extract the recurring instruction sequences, content composition, and link connection patterns, and generate a set of general response patterns.

[0035] A comparative analysis was conducted on the segmented data of all historical response processes belonging to the same trigger combination. Using pattern recognition algorithms, it was found that most successful cases contained a core instruction sequence, such as intent parsing followed by device status query, and then the generation of control instructions. It was also found that the response content of most cases included text confirmation information and actual control instructions. Furthermore, it was discovered that triggering a user confirmation step is a common and effective connection pattern when the user's intent is ambiguous. These repeatedly verified effective and universal process segments, content segments, and connection patterns were extracted and used as the cornerstone for constructing response rules, forming a set of universal response patterns. Each pattern includes a pattern identifier, applicable trigger combination type, core instruction sequence template, core content composition template, and core connection pattern description.

[0036] Step S11626: Add boundary conditions to the general response pattern, complete rule transformation and verification, and generate a set of scenario demand response rules.

[0037] This step builds upon the general response pattern by adding precise application boundaries and logical judgment conditions, transforming it into an executable rule.

[0038] Step S116261: For each pattern in the general response pattern set, supplement the boundary condition definition, including scene element change threshold, user operation timeout setting, and data missing handling scheme, and generate a pattern boundary condition set.

[0039] For the general response pattern "device status query after intent parsing and then control command generation," supplementary boundary conditions are provided. Scene element change threshold: If, after executing the device status query step, the absolute value of the difference between the current status value and the user's expected value is less than a preset adjustment threshold, it is considered that no adjustment is needed, and the process should terminate early or feedback should indicate that the status is close to the expected value. User operation timeout setting: In the user confirmation stage, if the user does not provide a clear response within the preset timeout period, the system should default to abandoning the operation and exiting the interaction. Data missing handling scheme: If necessary device status data cannot be obtained during execution, the system should skip the relevant query and generation steps and directly enter the anomaly reporting stage, informing the user that status information cannot be obtained. Similar boundary conditions are added to each general response pattern, generating a pattern boundary condition set. Each record includes a pattern identifier, boundary condition type, condition parameter threshold, timeout duration, and missing handling strategy.

[0040] Step S116262: Integrate the general response pattern set and the pattern boundary condition set, determine the applicable scenario element combination range for each pattern, and generate the response pattern adaptation range set.

[0041] The general response patterns are integrated with the boundary conditions generated in step S116261 to form complete response pattern prototypes. Then, the applicable scenario element combination range is determined for each prototype. For example, the pattern mentioned above, which includes status query and instruction generation, has the following applicable scenario element combination range: the interaction environment elements contain data dimensions related to device status, the user operation elements contain control-related keywords, and the interaction target elements are function control-related. However, for another general response pattern, such as the information query pattern, the applicable scenario element combination range is completely different; it requires the interaction environment elements to contain time or location information, the user operation elements to contain query-related keywords, and the interaction target elements to be information acquisition-related. Each response pattern is associated with its applicable scenario element combination range to generate a response pattern adaptation range set. Each record contains a pattern identifier, a list of applicable environment element types, a list of applicable operation element types, and a list of applicable target element types.

[0042] Step S116263: Transform each response mode, corresponding boundary conditions, and adaptation range into structured logic rules, determine the rule triggering conditions, execution steps, and output results, and generate an initial set of response logic rules.

[0043] The above information is transformed into structured logical rules that can be executed by a computer. For example, a function control pattern is transformed into a rule. The trigger condition for this rule is that the combination of scene elements currently input belongs to the "function control" range defined in the pattern's adaptation range set. The execution steps of the rule are: first, check boundary conditions, such as whether the target device is online; second, execute the core instruction sequence, including intent parsing and device status query; third, determine whether to generate control instructions or terminate early based on the query results and boundary conditions; fourth, generate control instructions; fifth, generate feedback information. The output of the rule is a structured data packet containing control instructions and feedback text. All patterns are transformed into similar rules to form an initial response logic rule set, where each rule contains a rule identifier, a trigger condition expression, a step sequence, and an output format definition.

[0044] Step S116264: Compare the execution logic of each rule in the initial response logic rule set, remove duplicate rule content, merge logically conflicting parts, and generate a rule deduplication and merging set.

[0045] A logical comparison is performed on all rules in the initial response logic rule set. For example, it is found that Rule A and Rule B have duplicate definitions in the timeout logic for handling user confirmation. These two rules are merged, and the timeout handling logic is extracted into a common sub-process, which is then uniformly called in both Rule A and Rule B. As another example, it is found that Rule C and Rule D have overlapping triggering conditions under certain scenario combinations, and their execution logic conflicts—one requires direct execution, while the other requires prior confirmation. Based on historical data, the rule with the higher success rate and better user satisfaction is selected to retain or modify its triggering conditions, thereby eliminating the conflict. After deduplication and merging, a concise and logically consistent deduplicated and merged rule set is generated.

[0046] Step S116265: Verify each rule in the rule deduplication and merging set. Test the completeness and consistency of rule execution by simulating the triggering process of scenario element demand trigger combination. Based on the test results, supplement and correct the rule details to generate a scenario demand response rule set.

[0047] A simulation testing environment was constructed. For each rule in the rule deduplication and merging set, a simulated scenario element request trigger combination that met its triggering conditions was constructed. Simulated input was injected into the rules, and their execution process was observed. Testing revealed that when the device status query result indicated the device was offline, although the rule could enter exception handling, the feedback information was not specific enough. Based on this test result, the details of the rules were supplemented, refining the exception handling feedback to include a specific description of the device identifier and the reason for offline status. Through repeated simulation testing and corrections, the execution integrity and logical coherence of all rules were guaranteed. Finally, all the verified rules constituted the scenario request response rule set.

[0048] Step S1163: Extract universally applicable response logic modules from the set of scenario request response rules and integrate them to form a set of general response logic modules.

[0049] Analyze each rule in the final generated set of scenario-based response rules. For example, rules controlling multiple different devices typically include a user intent parsing logic module and a device status query logic module. Extract the functional modules that appear repeatedly in multiple rules and have essentially the same implementation logic. For instance, abstract the user intent parsing module into a generic module that receives the user's raw voice or text input and outputs structured intent and parameters. Abstract the device status query module into another generic module that receives the device type and room identifier and returns the device's current status parameters. Standardize and encapsulate these generic modules, defining their standard input interfaces, output formats, and calling methods to form a set of generic response logic modules that can be invoked by different rules at any time.

[0050] Step S1164: For the scenario element request trigger combinations with special characteristics in the scenario request response rule set, supplement the exclusive response logic module and generate an exclusive response logic module set.

[0051] In contrast to step S1163, highly customized rules that cannot be covered by general modules are identified. For example, a scenario rule for "bedtime stories in a children's room" not only needs to invoke the player function but also needs to combine the child's age, previously listened-to story list, current time, and parental settings preferences to select and recommend a suitable story. This recommendation logic is very specific and cannot be implemented using a general device control module. Therefore, a dedicated response logic module is developed specifically for this particular scenario element request trigger combination. This module encapsulates complex recommendation algorithms and rules, specifically for handling bedtime story recommendation scenarios. All such modules customized for special scenarios are aggregated to form a set of dedicated response logic modules.

[0052] Step S1165: Associate the general response logic module set, the exclusive response logic module set, and the scene element request trigger set, determine the response logic module calling method corresponding to each scene element request trigger combination, and generate the basic logic for scene intent linkage.

[0053] Finally, the entire logic is assembled. This involves integrating the scene element request trigger set, the general response logic module set, the dedicated response logic module set, and the verified scene request response rule set. For each trigger combination in the scene element request trigger set, its corresponding response method is defined. This response method is not a simple rule call, but a modular call flow. For example, for a certain trigger combination, its response logic is defined as: calling a general module to parse the intent, then calling another general module to query the device status, and then, based on the parsed parameters and status, calling the corresponding rule in the scene request response rule set. This rule may also call other general modules or directly generate instructions. For another trigger combination, its response logic is defined as: after calling a general module to parse the intent, directly calling a dedicated module, which independently completes all subsequent processing. All trigger combinations and their corresponding module call sequences and rule execution logic are encapsulated into a complete set of executable logic that can be directly called by upper-layer applications. This set is the basic logic for scene intent linkage; it can intelligently match and call the most suitable processing logic based on the input scene element combination to generate an adapted intelligent interactive response.

[0054] Step S120: Construct a nested feedback collection link, embed the basic logic of scene intent linkage into the core execution node of the nested feedback collection link, and collect the user's direct feedback information, indirect feedback information, and adaptation association information between scene elements and intelligent interaction response through the hierarchical collection channels of the nested feedback collection link, respectively, to generate a scene feedback linkage set.

[0055] After generating the basic logic for scene intent linkage that enables intelligent responses, a data acquisition system for comprehensively evaluating response effectiveness is constructed—a nested feedback acquisition chain. First, a three-layer data acquisition architecture is designed, comprising a direct feedback acquisition layer, an indirect feedback acquisition layer, and an adaptation state acquisition layer. Each layer has an independent data acquisition channel and transmission path. Next, the execution node of the core scene intent linkage basic logic is used as the trigger core of the entire acquisition chain. Whenever the basic logic generates an intelligent interactive response and sends it to the user, this core execution node immediately notifies the three acquisition layers to initiate data acquisition. The direct feedback acquisition layer is responsible for capturing the user's explicit feedback to the response; the indirect feedback acquisition layer is responsible for monitoring subtle changes in the user's behavioral trajectory after receiving the response; and the adaptation state acquisition layer is responsible for analyzing the degree of matching between the response and the current scene elements. Finally, the data collected from these three layers is correlated and structured to generate a scene feedback linkage set containing user feedback content, scene element adaptation status, feedback background information, and feedback type identifiers. The following details its sub-steps.

[0056] Step S121: Construct a hierarchical structure of nested feedback acquisition links. This hierarchical structure has multiple levels, including a direct feedback acquisition layer, an indirect feedback acquisition layer, and an adaptation state acquisition layer. Each level is equipped with an independent acquisition channel and data transmission path. The data acquired by each level can be transmitted independently and are interconnected.

[0057] At the software level, three independent logical layers are defined. The first is the direct feedback acquisition layer, whose independent acquisition channels are the voice assistant's proactive inquiry interface and the user evaluation submission interface. When direct feedback is needed, this layer sends a question to the user through this channel and listens for the user's response; users can also proactively submit feedback through the application's evaluation function, and the data enters the system through this channel. The data transmission path of this layer is directly connected to the feedback data receiving queue. The second is the indirect feedback acquisition layer, whose independent acquisition channel is a listener on the system event bus, specifically listening to all user operation events, including button clicks, voice commands, application swipes, etc., but not explicit answers from users to system inquiries. The data of this layer is transmitted to another queue via a different path. The third is the adaptation state acquisition layer, whose independent acquisition channel is a periodically triggered snapshot capture service. Whenever an interaction response is completed, this service captures a snapshot of the current scene elements from the system state database. The data of this layer is transmitted to the third queue. These three queues are logically separate, ensuring the independence of the data at each layer. At the same time, each collected data item is marked with a globally unique interaction event identifier. This identifier is the only key for subsequent cross-layer data association, enabling data from different queues that belong to the same interaction event to be accurately associated.

[0058] Step S122: Embed the scene intent linkage basic logic into the core execution node of the feedback collection nested link. The core execution node can trigger the feedback collection operation at each level based on the intelligent interactive response generated by the scene intent linkage basic logic. The core execution node establishes a real-time data interaction channel with each collection level.

[0059] Within the execution engine of the scene intent linkage underlying logic, a response post-processing hook function is defined. This hook function is automatically invoked whenever the underlying logic completes the generation and distribution of a response instruction. This hook function is the core execution node. The primary task of the core execution node is to generate a unique interaction event identifier. Then, the core execution node broadcasts a collection start signal to the three collection layers through a pre-established real-time data interaction channel, such as a publish-subscribe pattern based on a message queue, and sends out the interaction event identifier along with brief information about the response, such as the response type and response content summary. For example, the core execution node sends a message to the direct feedback collection layer stating that the interaction event identifier is a certain value and requesting the start of collecting direct feedback for this response. Simultaneously, it also sends similar messages to the indirect feedback collection layer and the adaptation state collection layer, thereby accurately associating this response with all subsequent feedback data.

[0060] Step S123: Set up a feedback information collection unit in the direct feedback collection layer to collect the user's direct expression after receiving the intelligent interaction response in real time, including the user's expression of approval of the response content, suggestions for adjusting the response form, expression of questions about the interaction process, and operation instructions to terminate the interaction. Bind each direct feedback information to the corresponding intelligent interaction response to generate a direct feedback response association set.

[0061] The feedback information acquisition unit in the direct feedback acquisition layer begins listening to a designated channel after receiving the start signal from the core execution node. It monitors the microphone array; when the system broadcasts the response content, if the user says "Okay, thank you," this is collected as an expression of approval. If the user says "Too low, turn it up a bit," this is collected as an adjustment suggestion. If the user says "What did you say? I didn't hear you clearly," this is collected as a question. If the user directly says "Never mind, I won't adjust it," this is collected as an operation command to terminate the interaction. This collected voice content is converted into text and labeled with sentiment tags, such as positive, negative, and neutral. Then, this text, sentiment tags, and the interaction event identifier sent by the core execution node are packaged to form a direct feedback record. The data structure of this record is: event identifier, feedback type, feedback content, sentiment tag, and timestamp. All similar records are aggregated to form a direct feedback response association set, where each data point is precisely bound to an intelligent interaction response through an event identifier.

[0062] Step S124: Set up a behavior trajectory monitoring unit in the indirect feedback acquisition layer to monitor the changes in the user's behavior trajectory during the intelligent interaction process in real time, including the behavior of restarting the interaction after the interaction is interrupted, the operation of actively adjusting the interaction path, and the behavior of repeatedly expressing the same request. Associate each indirect feedback information with the corresponding scene elements to generate an indirect feedback scene association set.

[0063] The behavior trajectory monitoring unit in the indirect feedback acquisition layer, upon receiving the start signal from the core execution node, begins monitoring the user's behavior over a given period. This unit analyzes only what the user does, not what they say. For example, if the user reopens the relevant control panel via voice or application shortly after the system executes a response, this behavior is recorded as restarting the interaction after an interruption. If the user does not accept the system's recommended default parameters but manually drags the parameter adjustment slider to another value, this behavior is recorded as actively adjusting the interaction path. If the user repeats the same command immediately after the system completes a response, this behavior is recorded as repeatedly expressing the same request. These behavioral events, such as reopening the control panel, manually dragging the slider, and repeating voice commands, along with their timestamps, are recorded. Simultaneously, the interaction event identifier is obtained from messages sent by the core execution node, and a snapshot of the scene elements at the time of the behavior is retrieved from the system log, such as current environmental parameters, time, and user location, forming an indirect feedback record. The data structure of this record is: event identifier, behavior type, specific behavior description, and snapshot of the scene. All similar records are aggregated to form an indirect feedback scenario association set, in which each data point is associated with a specific combination of scenario elements.

[0064] Step S125: Set up an adaptation state analysis unit in the adaptation state acquisition layer to analyze the adaptation of intelligent interaction response with current scene elements in real time, including the consistency between response content and interaction environment elements, the synchronization between response rhythm and user operation elements, and the conformity between response process and interaction target elements, and generate scene element response adaptation set.

[0065] The adaptation state analysis unit in the adaptation state acquisition layer immediately performs a post-evaluation of the interaction response upon receiving the start signal from the core execution node. First, it analyzes the consistency between the response content and the elements of the interaction environment. For example, for a temperature adjustment response, it compares the target temperature after the response with the current actual room temperature and calculates the absolute value of the difference. If this difference is less than a preset comfortable temperature difference threshold, the content consistency is considered good; otherwise, it is considered poor. Second, it analyzes the synchronicity between the response rhythm and the user's operation elements. It records the delay from when the user issues the command to when the system begins to execute the response and compares this delay with the average acceptable delay in the user's historical operations. If the delay is significantly greater than the average acceptable delay, it is marked as poor rhythm synchronization. Finally, it analyzes the conformity between the response flow and the interaction target elements. It compares the actual executed response flow with the standard flow preset for that target in the scene intent linkage basic logic. For example, if the user's goal is to quickly complete an operation, the standard flow should be direct execution. However, if the actual executed flow involves additional queries, the flow conformity is low. The above analysis results are quantified to form a set of adaptation status indicators, which are then packaged into a single record: event identifier, content consistency indicator, rhythm synchronization indicator, process compliance indicator, and overall adaptation score. All similar records are aggregated to form a scenario element response adaptation set.

[0066] Step S126: Based on the associated data collected at each level, construct a feedback data association system and generate a scenario feedback linkage set.

[0067] This step associates and merges the datasets obtained from the three independent acquisition layers using globally unique interactive event identifiers to form a comprehensive feedback view. The following sections will explain each sub-step in detail.

[0068] Step S1261: Construct the association mapping relationship between the direct feedback response association set and the indirect feedback scenario association set, extract feedback information for the same intelligent interaction process from the direct feedback response association set and the indirect feedback scenario association set, and generate a comprehensive user feedback set.

[0069] Using the interaction event identifier as the key, a correlation query is performed between the direct feedback response association set and the indirect feedback scenario association set. For example, for a given interaction event identifier, one direct feedback is found in the direct feedback response association set, and multiple indirect feedbacks are found in the indirect feedback scenario association set, including instances where the user manually adjusted parameters a few seconds before the direct feedback occurred. This information from both sources is combined to form a more complete view of user feedback. It not only knows what the user said, but also what the user did before saying it. The above correlation mapping is established based on the time node of the interaction and the combination of scenario elements, i.e., the interaction event identifier. All feedback information from both the direct and indirect collection layers related to the same interaction process is merged into a single data structure, forming a comprehensive user feedback set.

[0070] Step S1262: Construct the association between the comprehensive set of user feedback and the set of scene element response adaptation, bind each user feedback information with the corresponding scene element adaptation status, and generate a feedback adaptation association set.

[0071] Using the interaction event identifier as the key, each record in the comprehensive user feedback set generated in step S1261 is associated with the record corresponding to the same event identifier in the scene element response adaptation set. The specific content of the user feedback is bound to the adaptation status indicators evaluated for this response. For example, a user feedback containing "adjustment suggestion" is bound to the adaptation status of this response, such as "content consistency indicator is low" or "rhythm synchronization indicator is medium." Through this binding, a feedback adaptation association set is generated, which clearly reflects the adaptation context in which the user feedback was generated, i.e., under what response effect the user gave the feedback.

[0072] Step S1263: Supplement the background information in the feedback adaptation association set, including the specific time when the feedback was generated, the corresponding intelligent interaction link, the user identifier of the user participating in the interaction, and the specific status data of the scene elements. Each feedback item contains context information.

[0073] Building upon the feedback adaptation association set, further supplementary contextual information is provided for each piece of feedback. The precise timestamp of the feedback's generation is extracted from system logs. The interaction flow record determines which interaction stage the feedback pertains to, such as whether it's for content playback or command execution. The unique identifier of the currently interacting user is obtained from user account information. Specific state data of various environmental and operational elements at the moment of feedback generation is extracted from scene element snapshots, such as specific values ​​for light intensity, temperature, and the user's recent operation sequence. These background information fields are added to each record in the feedback adaptation association set to ensure that each feedback item contains complete contextual information.

[0074] Step S1264: Classify and organize the feedback information in the feedback adaptation association set, and classify it into content optimization feedback, rhythm adjustment feedback, path modification feedback, and adaptation improvement feedback according to the feedback type, and generate a feedback classification set.

[0075] For each piece of feedback in the adaptation set supplemented with background information, it is automatically categorized based on its content. Feedback involving suggestions for modifying response text, data display, etc., is categorized as content optimization feedback. Feedback involving opinions on response speed, step intervals, etc., or indirectly expressing discomfort with the response rhythm, is categorized as rhythm adjustment feedback. Feedback involving cumbersome interaction steps, requests to skip certain steps, or indirectly demonstrating proactive modification of the interaction path, is categorized as path modification feedback. Feedback involving a mismatch between the response and the scenario, or a low score from the adaptation status analysis unit, is categorized as adaptation improvement feedback. Each piece of feedback is then tagged with a category, generating a set of feedback categories.

[0076] Step S1265: Integrate all feedback information in the feedback category set and the associated adaptation status information and background information to generate a scene feedback linkage set.

[0077] All data in the feedback category set, including the user feedback content for each feedback item, the related scene element adaptation status indicators, complete feedback background information, and feedback type identifiers, are ultimately integrated to form a structured scene feedback linkage set. Each data item in this set is a complete, self-contained feedback event record that comprehensively reflects the source, content, background, and system response status of a user feedback instance.

[0078] Step S130: Using the scene feedback linkage set as the driving source, construct a dynamic adaptation iteration link. Through multiple rounds of closed-loop iteration of the dynamic adaptation iteration link, adjust the scene element combination rules and user interaction demand matching rules in the scene intent linkage basic logic to generate a dynamic adaptation optimization link.

[0079] The system extracts the user feedback content and scene element adaptation status of each feedback item from the scene feedback linkage set. It analyzes the problem points in the basic logic of scene intent linkage pointed to by each feedback item, generating a logic adjustment direction set. This set contains the scene element combination rules and user interaction request matching logic that need adjustment. Based on the logic adjustment direction set, the iterative optimization direction of the dynamic adaptation iteration link is determined. This direction includes optimizing the accuracy of the correspondence between scene element combinations and user interaction requests, improving the adaptation status of intelligent interaction responses and scene elements, and improving the rationality of call to response logic modules, generating an iterative optimization direction set. The core architecture of the dynamic adaptation iteration link is constructed, including an adjustment instruction generation module, a logic modification execution module, an optimization effect analysis module, and a feedback loop input module. The logic adjustment direction set and the iterative optimization direction set are input into the adjustment instruction generation module to generate specific adjustment instructions for the scene element combination rules and user interaction request matching logic. These instructions include the specific content, method, and scope of adjustment, generating a logic adjustment instruction set. The logic adjustment instruction set is input into the logic modification execution module. Based on these instructions, the corresponding content in the basic logic of scene intent linkage is modified. This includes adjusting the composition of scene element combinations, optimizing the matching relationship between user interaction requests and scene element combinations, and improving the calling conditions of the response logic module, generating preliminary optimized scene intent linkage logic. Multiple rounds of iterative adjustments are completed based on the optimization effect verification, generating an iteration trajectory and a dynamically adapted optimization link. The following details each sub-step.

[0080] Step S131: Extract the user feedback content and scene element adaptation status corresponding to each feedback item from the scene feedback linkage set, analyze the problem points in the scene intent linkage basic logic pointed to by each feedback item, and generate a logic adjustment pointing set.

[0081] The process iterates through each feedback item in the scenario feedback linkage set. For each feedback item, it first parses the user feedback content field. If the feedback type is content optimization, the issue points to the rule module that generates the response content. Further analysis of the specific issues mentioned in the feedback content is then performed; for example, if the feedback indicates "incomplete information," the issue is refined to the missing data dimensions in the content generation rules. If the feedback type is rhythm adjustment, the issue points to the parameter setting rules that control the response rhythm, such as the interval duration parameter or response delay threshold. If the feedback type is path modification, the issue points to the rule module that determines the interaction flow, such as the stage switching conditions or exception handling paths. Then, the adaptation status of the scenario elements associated with the feedback item is considered. For example, if the content consistency index in the adaptation status is low, it confirms that the issue does indeed exist in the content generation rules. If the rhythm synchronization index is low, it confirms that the issue exists in the rhythm control rules. Record the problems identified in each feedback item, clarify the specific module or rule in the underlying logic to which the problem points, and form a set of logic adjustment targets. Each target item includes a description of the problem point, the identifier of the logic module to which it points, and the identifier of the specific rule or parameter to which it points.

[0082] Step S132: Based on the logical adjustment target set, determine the iterative optimization direction of the dynamic adaptation iteration link. The iterative optimization direction includes optimizing the accuracy of the correspondence between the combination of scene elements and user interaction requirements, improving the adaptation status of intelligent interaction response and scene elements, and improving the rationality of calling response logic modules, thereby generating an iterative optimization direction set.

[0083] Summarize all problem points in the logic adjustment target set generated in step S131. Perform cluster analysis on the problem points. If a large number of problem points indicate mismatch or missing matches between scene element combinations and user interaction requests, determine an iterative optimization direction as "optimizing the accuracy of the correspondence between scene element combinations and user interaction requests." If a large number of problem points indicate inconsistencies between response content and environment, or response rhythm and user habits, determine an iterative optimization direction as "improving the adaptability of intelligent interaction response and scene elements." If a large number of problem points indicate calling the wrong response module or an unreasonable module calling order, determine an iterative optimization direction as "improving the rationality of calling response logic modules." Assign a unique identifier to each determined iterative optimization direction to form an iterative optimization direction set, which contains several directions that need to be optimized in this iteration.

[0084] Step S133: Construct the core architecture of the dynamic adaptation iteration link. This core architecture includes an adjustment instruction generation module, a logic modification execution module, an optimization effect analysis module, and a feedback loop input module.

[0085] Four functional modules are defined at the software level, and data flow channels are established between them. The adjustment instruction generation module receives a set of logical adjustment directions and an iterative optimization direction set as inputs. Internally, it includes a rule conflict detection engine and an instruction synthesizer. Its output interface generates a set of logical adjustment instructions. The logic modification execution module receives the set of logical adjustment instructions as inputs. Internally, it includes a basic logic parser and a rule editor, capable of locating and modifying specific code segments or rule entries in the basic logic of scene intent linkage. Its output interface generates preliminary optimized scene intent linkage logic. The optimization effect analysis module receives the preliminary optimized scene intent linkage logic and the original or newly acquired scene feedback linkage set as inputs. Internally, it includes a simulation testing engine and an effect evaluator, capable of comparing the response effects before and after optimization. Its output interface generates optimization effect analysis results. The feedback loop input module receives the portion of the optimization effect analysis results that does not reach a preset threshold. Internally, it includes a feedback information formatter, capable of repackaging the substandard information into a format similar to logical adjustment directions. Its output interface sends the packaged supplementary adjustment directions back to the adjustment instruction generation module, forming a closed loop.

[0086] Step S134: Input the logical adjustment direction set and the iterative optimization direction set into the adjustment instruction generation module to generate specific adjustment instructions for matching the scene element combination rules and user interaction requirements. The adjustment instructions include the specific content, adjustment method, and adjustment scope of the adjustment, and generate a set of logical adjustment instructions.

[0087] The logical adjustment target set generated in step S131 and the iterative optimization direction set generated in step S132 are fed into the adjustment instruction generation module as input. The adjustment instruction generation module first determines the adjustment method for each problem point in the logical adjustment target set, based on the iterative optimization direction. For example, for a problem point pointing to "matching error," if the optimization direction is "improving the corresponding accuracy," the adjustment method might be "modifying the matching threshold" or "adjusting the matching weight." Next, the specific content of the adjustment is determined. For example, for modifying the matching threshold, it needs to be clarified whether to increase or decrease the threshold, and what the new candidate threshold value is. Then, the scope of the adjustment is determined—whether to modify a single rule, a group of rules, or the entire module. The adjustment instruction generation module encapsulates all determined adjustment actions into instructions in a unified format. Each adjustment instruction includes an instruction identifier, the identified problem point, the adjustment method code, adjustment content parameters, and an adjustment scope description. All instructions are summarized to form a logical adjustment instruction set.

[0088] Step S135: Input the set of logic adjustment instructions into the logic modification execution module, modify the corresponding content in the basic logic of scene intent linkage according to the adjustment instructions, including adjusting the composition of scene element combination, optimizing the matching relationship between user interaction requests and scene element combination, improving the calling conditions of the response logic module, and generating preliminary optimized scene intent linkage logic.

[0089] The set of logical adjustment instructions generated in step S134 is used as input and sent to the logical modification execution module. The logical modification execution module parses each adjustment instruction. Based on the adjustment scope description in the instruction, it locates the target code segment, rule base entry, or configuration file parameter in the scene intent linkage basic logic. Then, based on the adjustment method encoding and adjustment content parameters, it executes the specific modification operation. For example, if the instruction adjusts the matching threshold, the corresponding rule is found, and the threshold parameter is modified from its original value to a new value. If the instruction adjusts the module call condition, the precondition expression for the corresponding logic module is modified. If the instruction adjusts the composition of the element combination, the definition of the scene element combination is modified, adding or deleting certain element dimensions. After executing all modification instructions, a modified scene intent linkage basic logic is generated, i.e., the scene intent linkage logic is initially optimized.

[0090] Step S136: Based on the verification of the optimization effect, complete multiple rounds of iterative adjustments and generate iterative trajectories and dynamic adaptation optimization links.

[0091] The initially optimized logic is then evaluated again. If the results are not satisfactory, the next iteration is initiated. The following sections detail each of its sub-steps.

[0092] Step S1361: Input the preliminary optimized scene intent linkage logic into the optimization effect analysis module, combine it with the relevant feedback information in the scene feedback linkage set, analyze the improvement status of the preliminary optimized logic in meeting user interaction requirements and adapting to scene elements, and generate optimization effect analysis results.

[0093] The preliminary optimized scene intent linkage logic generated in step S135 is input into the optimization effect analysis module. The optimization effect analysis module retrieves feedback items related to the current optimization direction from the scene feedback linkage set. For example, if the current optimization direction is to improve the corresponding accuracy, feedback items whose feedback type is related to the matching problem are selected. The module internally starts a simulation test engine, using the original scene elements corresponding to these feedback items as input, and feeds them into the original logic and the preliminary optimized logic respectively, comparing their output responses. The performance improvement of the optimized logic on these feedback items is calculated. For example, for feedback pointing to a matching error, does the optimized logic output the correct matching result? Simultaneously, the module also calculates the distribution change of the adaptation status indicators of the preliminary optimized logic on the overall scene feedback linkage set, such as whether the average value of the content consistency indicator has improved. All these comparative data and calculation results are compiled into an optimization effect analysis result, which includes the quantitative improvement value for each optimization direction, whether the preset improvement threshold has been reached, and a specific analysis of directions that did not meet the target.

[0094] Step S1362: Based on the optimization effect analysis results, sequentially determine the improvement status of the preliminary optimization scenario intent linkage logic in each direction of the iterative optimization direction set. If there is a direction that has not reached the preset improvement threshold, the analysis results of the corresponding direction are fed back to the feedback loop input module, which generates a supplementary adjustment direction for that direction and re-inputs the adjustment instruction generation module.

[0095] The algorithm iterates through each direction in the set of iterative optimization directions, checking whether the improvement quantification value for that direction in the optimization effect analysis results generated in step S1361 has reached the preset improvement threshold. If all directions have met the threshold, the algorithm proceeds to step S1363. If one or more directions have not met the threshold, the analysis results for that direction, including the specific manifestations of the non-compliance and possible cause inferences, are sent to the feedback loop input module. The feedback loop input module reformats the above information, generating a new set of supplementary adjustment directions. These directions are more focused on the specific problems that have not been met compared to the initial logical adjustment directions. Then, this set of supplementary adjustment directions is sent back to the adjustment instruction generation module in step S134 as new input to begin a new round of iterative optimization.

[0096] Step S1363: Repeat the iteration process until the generated optimized scene intent linkage logic reaches the preset improvement threshold in all directions of the iteration optimization direction set, and generate the final optimized scene intent linkage logic.

[0097] The loop from steps S134 to S1362 continues. In each iteration, the adjustment instruction generation module generates new adjustment instructions based on the latest supplementary adjustment direction, the logic modification execution module generates new preliminary optimization logic, and the optimization effect analysis module evaluates the effect of the new logic. Iteration stops when, after a certain iteration, the optimization effect analysis results show that the improvement quantification value has reached or exceeded the preset improvement threshold for all iteration optimization directions. The preliminary optimization scenario intent linkage logic generated in this current iteration is the final optimization scenario intent linkage logic.

[0098] Step S1364: Record all adjustment information during the dynamic adaptation iteration process, including the feedback source of each adjustment, the content of the adjustment instruction, the modified logical part, and the optimization effect analysis results, and generate an iterative adjustment trajectory set.

[0099] Throughout the iteration process, a global logger is set up. For each iteration, the logger records the feedback sources that triggered the iteration, such as which specific scenario feedback sets the feedback came from. It also records all the contents of the logic adjustment instruction set generated in this iteration. Furthermore, it records which parts of the basic logic were actually modified by the logic modification execution module, including a comparison of the code or rules before and after the modification. Finally, it records all the analysis results data output by the optimization effect analysis module after this iteration. This information is organized according to the iteration rounds to form an iterative adjustment trajectory set. This iterative adjustment trajectory set records in detail every heartbeat of the dynamic adaptation iteration chain from start to finish.

[0100] Step S1365: Integrate the final optimized scenario intent linkage logic and iterative adjustment trajectory set, extract the rules of iterative adjustment and the action path of feedback information, and generate a dynamic adaptation optimization link.

[0101] The final optimized scenario intent linkage logic generated in step S1363 is integrated with the iterative adjustment trajectory set generated in step S1364. Effective adjustment patterns are analyzed from the iterative adjustment trajectory set; for example, it is found that adjusting a specific parameter yields the best results when a certain type of feedback occurs. These analysis results can be transformed into optimization suggestions. Simultaneously, the complete path of feedback information in the iterative chain is depicted, from a single feedback in the initial scenario feedback linkage set to its final use in generating the optimization logic. This complete process, integrating the final logic, adjustment trajectories, and optimization rules, is encapsulated into a reusable, dynamically adaptable optimization chain. This chain not only includes the optimized results but also the methodology and process records for how to optimize based on feedback.

[0102] Step S140: Construct a collaborative operation process between the personalized model architecture and the dynamic adaptation optimization link, embed the adjustment logic of the dynamic adaptation optimization link into the corresponding hierarchical structure of the personalized model architecture, and generate a complete personalized model architecture.

[0103] This paper extracts the final optimized scene intent linkage logic from the dynamic adaptation and optimization chain, using it as the core logic of the linkage adaptation layer in the personalized model architecture. The linkage adaptation layer receives real-time scene elements and user interaction requests, generating corresponding preliminary intelligent interaction response instructions. It also extracts the scene feedback linkage set processing logic from the dynamic adaptation and optimization chain, using it as the core logic of the feedback processing layer in the personalized model architecture. This layer receives user feedback information and scene element adaptation status information, generating logic adjustment driving data. Finally, it extracts the iterative adjustment logic from the dynamic adaptation and optimization chain, using it as the core logic of the iterative optimization layer in the personalized model architecture. This layer receives the logic adjustment driving data generated by the feedback processing layer and optimizes the core logic of the linkage adaptation layer. The paper constructs a hierarchical connection process between the linkage adaptation layer, feedback processing layer, and iterative optimization layer, determining the data transmission format, triggering conditions, and interaction methods between each layer. This includes the linkage adaptation layer transmitting preliminary intelligent interaction response instructions and corresponding scene element information to the feedback processing layer; the feedback processing layer transmitting logic adjustment driving data to the iterative optimization layer; and the iterative optimization layer transmitting logic optimization instructions to the linkage adaptation layer, generating a set of hierarchical connection rules. Add a data storage layer to the personalized model architecture. This layer stores data related to scene elements, user interaction requests, user feedback, iterative trajectory data of the dynamic adaptation and optimization process, and operational status data for each layer, generating data storage management rules. Then, add additional functional layers to the model architecture, complete architecture integration and testing, and generate the complete personalized model architecture. The following sections detail each sub-step.

[0104] Step S141: Extract the final optimized scene intent linkage logic in the dynamic adaptation optimization link and use it as the core logic of the linkage adaptation layer of the personalized model architecture. The linkage adaptation layer is used to receive real-time scene elements and user interaction requests, and generate corresponding intelligent interaction response preliminary instructions.

[0105] From the dynamic adaptation and optimization link generated in step S136, the final optimized scene intent linkage logic obtained in step S1363 is extracted. This logic is deployed as a whole to a new layer in the personalized model architecture, named the linkage adaptation layer. The input interface of this layer is defined, which can receive real-time scene element data and user interaction request data. The data format needs to match the input requirements of the final optimized scene intent linkage logic. The output interface of this layer is defined, which outputs the preliminary intelligent interactive response command generated by the final optimized scene intent linkage logic based on the input data. This preliminary command is a structured data packet containing a response content template identifier, a list of parameters to be filled, expected response rhythm parameters, and an interaction path framework.

[0106] Step S142: Extract the scene feedback linkage set processing logic in the dynamic adaptation optimization link and use it as the core logic of the feedback processing layer of the personalized model architecture. The feedback processing layer is used to receive user feedback information and scene element adaptation status information and generate logic adjustment driving data.

[0107] From the entire process of generating and processing the scene feedback linkage set described in steps S120 to S126, the core processing logic is extracted: how to construct structured feedback data that can be used for iteration from the original feedback and adaptation state analysis. This logic is deployed to the second layer of the personalized model architecture, named the feedback processing layer. The input interface of this layer is defined, which receives user feedback information from external sources, including direct and indirect feedback, as well as scene element adaptation state information from the adaptation state monitoring module. The output interface of this layer is defined, which outputs the processed and correlated logic adjustment driving data. Its data format is similar to the logic adjustment target set generated in step S131, clearly indicating potential problems in the current basic logic.

[0108] Step S143: Extract the iterative adjustment logic from the dynamic adaptation optimization link and use it as the core logic of the iterative optimization layer of the personalized model architecture. The iterative optimization layer is used to receive the logic adjustment driving data generated by the feedback processing layer and optimize and adjust the core logic of the linkage adaptation layer.

[0109] From the entire dynamic adaptation iteration chain described in steps S130 to S136, the core iterative adjustment logic is extracted, namely, how to generate adjustment instructions based on feedback-driven data, execute modifications, evaluate effects, and iterate cyclically. This logic is deployed to the third layer of the personalized model architecture, named the Iterative Optimization Layer. The input interface of this layer is defined, which receives logic adjustment-driven data output from the feedback processing layer. The output interface of this layer is defined, which outputs optimization adjustment instructions for the core logic of the linkage adaptation layer, with a data format similar to the logic adjustment instruction set generated in step S134. The Iterative Optimization Layer internally includes the adjustment instruction generation module, logic modification execution module, optimization effect analysis module, and feedback loop input module defined in step S133, but at this point, these modules operate on the core logic of the linkage adaptation layer.

[0110] Step S144: Construct the hierarchical connection process between the linkage adaptation layer, feedback processing layer, and iterative optimization layer, determine the data transmission format, triggering conditions, and interaction methods between each layer, including the linkage adaptation layer transmitting preliminary intelligent interaction response instructions and corresponding scene element information to the feedback processing layer, the feedback processing layer transmitting logical adjustment driving data to the iterative optimization layer, the iterative optimization layer transmitting logical optimization instructions to the linkage adaptation layer, and generating a set of hierarchical connection rules.

[0111] Define the collaborative working method among the three core layers. After generating an initial intelligent interactive response command, the linkage adaptation layer, in addition to outputting the command to downstream execution units, must also package a copy of the command and the input scene element data used to generate the command into a data packet and actively push it to the feedback processing layer via the internal bus. The feedback processing layer continuously monitors the internal bus; once it receives the data packet from the linkage adaptation layer, it begins preparing to receive subsequent feedback information. When the feedback processing layer accumulates sufficient data and generates logic adjustment driving data, it sends the driving data to the iterative optimization layer via the internal bus according to preset trigger conditions, such as reaching a certain data volume or timed triggering. After receiving the driving data, the iterative optimization layer initiates its internal iterative optimization process and, upon completion, sends the generated logic optimization command to the linkage adaptation layer via the internal bus. The linkage adaptation layer monitors the internal bus; once it receives the optimization command from the iterative optimization layer, it immediately adjusts its own core logic according to the command content. The above data transmission format, trigger conditions, and interaction methods are explicitly recorded to form a set of hierarchical connection rules.

[0112] Step S145: Add a data storage layer to the personalized model architecture. The data storage layer is used to store scene element-related data, user interaction request-related data, user feedback-related data, iterative trajectory data of the dynamic adaptation and optimization link, and running status data of each layer, and generate data storage management rules.

[0113] In addition to the three core layers, a data storage layer is added. This layer contains multiple data storage areas. The historical scene element library stores scene element data from all historical interactions and is indexed by time, user, and other dimensions. The historical request library stores parsed user interaction request data. The feedback database stores all data in the scene feedback linkage set generated in step S126. The iterative trajectory library stores data in the iterative adjustment trajectory set generated in step S1364. The runtime status library records the runtime logs, performance metrics, and configuration parameters of the linkage adaptation layer, feedback processing layer, and iterative optimization layer. Access permissions, read / write policies, data retention periods, and backup / recovery mechanisms are defined for each data storage area, generating data storage management rules.

[0114] Step S146: Add additional functional layers to the model architecture, complete the architecture integration and testing, and generate a complete personalized model architecture.

[0115] For example, based on the existing layers, two additional functional layers are added. A collaborative coordination process is established between the personalized model architecture and the dynamic adaptation optimization link. This involves determining the adjustment logic of the dynamic adaptation optimization link and the collaborative working method of each layer of the personalized model architecture. This includes the iterative adjustment process of the dynamic adaptation optimization link triggering parameter updates at each layer of the personalized model architecture, and the feedback of the personalized model architecture's operational status to the dynamic adaptation optimization link to optimize iteration efficiency, generating a set of collaborative coordination rules. An input adaptation layer is added to the personalized model architecture. This layer receives real-time scene element data and user interaction request data from external sources, converts the data format to meet the receiving requirements of the linkage adaptation layer, and generates input data adaptation rules. An output adaptation layer is added to the personalized model architecture. This layer receives preliminary intelligent interaction response instructions generated by the linkage adaptation layer, optimizes the instruction format to meet the execution requirements of the intelligent interaction system, and generates output data adaptation rules. Integrating the linkage adaptation layer, feedback processing layer, iterative optimization layer, data storage layer, input adaptation layer, and output adaptation layer, and based on the set of layer connection rules, data storage management rules, collaborative coordination rules, input data adaptation rules, and output data adaptation rules, a personalized initial model architecture is constructed. Layer-by-layer collaboration testing is performed on the initial personalized model architecture to verify the smoothness of data transmission between layers, the effectiveness of logic adjustments, and the rationality of collaboration methods. Based on the test results, the functional boundaries and collaboration methods of each layer are optimized, generating a complete personalized model architecture including the linkage adaptation layer, feedback processing layer, and iterative optimization layer.

[0116] Step S150: Through the complete architecture of the personalized model, receive real-time scene elements and user interaction requests, and output scene-adaptive intelligent interaction response instructions.

[0117] Step S151: The input adaptation layer receives real-time scene element data and user interaction request data from external sources, performs format conversion and standardization processing on the data according to the input data adaptation rules, and generates adapted real-time scene element data and adapted user interaction request data.

[0118] The input adaptation layer continuously monitors external data source interfaces. When new real-time scene element data arrives, such as the latest environmental readings from the sensor network and user interaction request data, such as the user intent structure just parsed by the natural language understanding module, the input adaptation layer first checks the format of this raw data. If the data format is inconsistent with the input format required by the linkage adaptation layer, a conversion is performed. For example, time-series data with different sampling frequencies from sensors are unified to the same frequency through interpolation or thinning. Data with different dimensions are normalized, for example, light intensity values ​​are mapped to the range of 0 to 1, and temperature values ​​are also mapped to the range of 0 to 1 to eliminate the influence of dimensions. The text description in the user intent structure is converted into an enumeration type or one-hot encoding. After the conversion is completed, adapted real-time scene element data and adapted user interaction request data are generated. The formats of these two datasets are directly parsable by the linkage adaptation layer.

[0119] Step S152: Transmit the adapted real-time scene element data and the adapted user interaction request data to the linkage adaptation layer. The linkage adaptation layer analyzes the correspondence between the two types of data based on the final optimized scene intent linkage logic and generates preliminary intelligent interaction response instructions. The preliminary intelligent interaction response instructions include the initial response content, the initial response rhythm plan, and the initial interaction path plan.

[0120] The two sets of data generated in step S151 are transmitted to the linkage adaptation layer via the internal bus. Upon receiving the data, the linkage adaptation layer uses it as input and feeds it into the final optimized scene intent linkage logic deployed within it. This logic first executes the scene element combination rules, combining the current environmental elements, operational elements, and target elements into a scene state vector. Then, it executes the user interaction request matching rules, calculating the matching degree between the scene state vector and each preset request category, and selecting the request with the highest matching degree as the current user's core request. Next, based on the matched request and scene state, it calls the response logic module calling rules, selecting the most suitable response module combination and calling order. Finally, it executes the selected response module, generating specific response content, such as selecting a template from the content template library and filling in parameters extracted from the scene data. Simultaneously, based on the scene state and user historical data, it generates a response rhythm plan, such as determining the voice playback speed and the timeout duration for waiting for user feedback. It also generates an interaction path plan, such as determining whether to execute step A or step B first, and under what conditions certain steps can be skipped. All this information is encapsulated into a structure, namely the initial intelligent interaction response instruction.

[0121] Step S153: The linkage adaptation layer transmits the initial intelligent interaction response instructions and the corresponding adapted real-time scene element data to the feedback processing layer, and at the same time transmits the initial intelligent interaction response instructions to the output adaptation layer for initial format optimization.

[0122] After generating the initial instruction, the linkage adaptation layer performs two parallel data distribution operations. First, it packages a copy of the newly generated intelligent interactive response initial instruction, along with the adapted real-time scene element data used to generate the instruction, into a single data unit and actively pushes it to the input queue of the feedback processing layer via the internal bus. Second, it sends the intelligent interactive response initial instruction itself to the output adaptation layer via another internal bus. Upon receiving the instruction, the output adaptation layer immediately initiates a preliminary format optimization process, such as converting the internal data structure into a serialization format more suitable for network transmission, or preparing instruction copies in different formats for different output channels, such as the speech synthesis module and the application interface rendering module. However, it does not actually send these copies to the execution unit at this stage.

[0123] Step S154: The feedback processing layer receives feedback information from the user regarding the initial instructions of the intelligent interaction response and real-time scene element adaptation status information. According to the scene feedback linkage set processing logic, it analyzes the adjustment dimension indicated by the feedback information and the adaptation dimension indicated by the real-time scene element adaptation status information, establishes the correlation between the two types of information in the corresponding dimensions, and generates logical adjustment driving data.

[0124] The feedback processing layer continuously monitors its input queue. Upon receiving a data unit from the linkage adaptation layer, it begins to wait for and collect feedback information for this interaction. This includes direct feedback, such as the user's subsequent voice evaluation, and indirect feedback, such as the user's subsequent actions. Simultaneously, it obtains the scene element adaptation status information for this response from the adaptation status monitoring module, such as content consistency scores. The feedback processing layer invokes its core scene feedback linkage set processing logic to process all the collected information. This logic first associates the direct and indirect feedback, then binds them to the adaptation status information, generating a structured feedback event as described in step S126. Next, the logic analyzes this structured feedback event to determine which adjustment dimension it primarily points to—content, rhythm, or path. Finally, the logic outputs a formatted data packet, namely the logic adjustment driving data, which clearly indicates potential problems in the current basic logic, as well as relevant feedback evidence and adaptation status indicators.

[0125] Step S155: Transmit the logic adjustment driving data to the iterative optimization layer. Based on the iterative adjustment logic, the iterative optimization layer generates optimization adjustment instructions for the core logic of the linkage adaptation layer and transmits them to the linkage adaptation layer.

[0126] The feedback processing layer sends the logic adjustment driving data generated in step S154 to the iterative optimization layer via the internal bus. The iterative optimization layer uses this data as input to initiate its internal iterative adjustment logic. This logic first determines whether there is enough driving data to trigger an optimization adjustment. If the triggering condition is met, it calls the internal adjustment instruction generation module to generate a specific optimization adjustment instruction based on the driving data. Then, it calls the logic modification execution module to simulate modifying a copy of the core logic of the linkage adaptation layer. Next, it calls the optimization effect analysis module to evaluate the effect of the modification. If the evaluation passes, a formal optimization adjustment instruction is generated. The iterative optimization layer then sends the finally generated optimization adjustment instruction to the linkage adaptation layer via the internal bus.

[0127] Step S156: Based on the logic adjustment, drive data optimization response instructions and model core logic to complete intelligent interactive response output and continuous iteration.

[0128] After receiving optimization instructions from the iterative optimization layer, the linkage adaptation layer immediately parses the instructions and updates its core logic accordingly, such as modifying a matching threshold or adjusting the order of module calls. Simultaneously, for the currently processed interaction, the linkage adaptation layer may have already produced some outputs based on the initial instructions, or it may adjust parts that haven't yet been output based on the latest optimization instructions. Upon receiving the final instructions from the linkage adaptation layer, the output adaptation layer completes the final format optimization and outputs it as a scene-adaptive intelligent interaction response instruction to the execution unit of the intelligent interaction system, such as a speech synthesis engine or device control gateway, guiding the intelligent interaction system to adjust its response content, response rhythm, and interaction path. The data storage layer stores all data throughout the entire process in real time. The complete personalized model architecture receives the running status information after the intelligent interaction system executes the instructions through a collaborative coordination process, feeds it back to the dynamic adaptation optimization link, and initiates a new round of logic optimization, enabling the scene-adaptive intelligent interaction response instructions to continuously adapt to real-time changing scene elements and user interaction needs. The following details its sub-steps.

[0129] For example, in step S1561: the linkage adaptation layer adjusts the content, rhythm and path scheme of the initial intelligent interaction response according to the optimization adjustment instructions, generates intelligent interaction response optimization instructions, and updates its own core logic to adapt to subsequent interaction needs.

[0130] The linkage adaptation layer parses the optimization and adjustment instructions received from the iterative optimization layer. It extracts content adjustment parameters from these instructions, including text description modification requirements, data supplementation dimensions, and function option update instructions. Based on these parameters, it modifies the initial response content in the preliminary intelligent interaction response instructions, including replacing statements, supplementing missing data, and updating function options, generating an optimized response content version. It extracts rhythm adjustment parameters from the instructions, including response execution speed settings, stage interval duration adjustments, and user waiting prompt timing. Based on these parameters, it optimizes the initial response rhythm scheme in the preliminary intelligent interaction response instructions, adjusting the execution time allocation for each stage and the instruction sending time, generating an optimized response rhythm version. It identifies path adjustment parameters from the instructions, including stage addition / removal instructions, path branch switching requirements, and priority execution channel settings. According to these parameters, it modifies the initial interaction path scheme in the preliminary intelligent interaction response instructions, adding or removing redundant stages, switching path branches, and setting priority execution order, generating an optimized interaction path version. Finally, it integrates the optimized response content version, the optimized response rhythm version, and the optimized interaction path version to generate the initial intelligent interaction response optimization instruction. Simultaneously, the logical update requirements in the optimization and adjustment instructions are extracted, including supplementing the rules for combining scene elements, modifying the matching conditions for user interaction requests, and adjusting the priority of response module calls. Based on these requirements, the corresponding parts of the core logic in the linkage adaptation layer are modified, new rule entries are added, matching condition thresholds are adjusted, and the module call order is updated to generate an updated version of the core logic for handling future interaction requests. The feasibility of executing the initial intelligent interaction response optimization instructions is verified, and the compatibility of the instructions and logic is tested in conjunction with the updated version of the core logic. Based on the verification results, the details of the instructions are optimized to generate the final intelligent interaction response optimization instructions.

[0131] Step S1562: Transmit the intelligent interaction response optimization instruction to the output adaptation layer. The output adaptation layer performs final optimization of the instruction format according to the output data adaptation rules, and adjusts the instruction format to adapt to the execution requirements of the intelligent interaction system.

[0132] The linkage adaptation layer transmits the intelligent interactive response optimization instruction generated in step S1561 to the output adaptation layer via the internal bus. Upon receiving the instruction, the output adaptation layer invokes its internally defined output data adaptation rules. This rule set contains format conversion templates for different target execution units. For example, for text content to be sent to the speech synthesis engine, it is converted from the internal data structure into a plain text string, along with speech synthesis parameters such as speech rate and pitch. For instructions to be sent to the device control gateway, it is converted into a standard instruction frame format conforming to the gateway communication protocol, including device identifier, control code, and parameter checksum. For data to be displayed on the application interface, it is converted into a JSON data structure required by the interface rendering engine. After all format conversions are completed, the final executable instruction is generated.

[0133] Step S1563: Data storage layer: Real-time storage of adapted scene element data, adapted user interaction request data, intelligent interaction response initial instructions, user feedback information, logic adjustment driving data, optimization adjustment instructions, and intelligent interaction response optimization instructions.

[0134] Throughout the interaction, the data storage layer operates continuously. It captures all key data packets from the internal bus, including the adapted real-time scene element data and adapted user interaction request data generated in step S151, the initial intelligent interaction response instructions generated in step S152, the user feedback information collected in step S154 and the generated logic adjustment driving data, the optimization adjustment instructions generated in step S155, and the final intelligent interaction response optimization instructions output in step S1562. This data is stored according to a preset database table structure and associated with globally unique interaction event identifiers, forming a complete and traceable interaction lifecycle data record.

[0135] Step S1564: Output Adaptation Layer. The optimized intelligent interaction response instruction is output to the intelligent interaction system as a scene-adaptive intelligent interaction response instruction, guiding the intelligent interaction system to adjust the response content, response rhythm and interaction path.

[0136] The output adaptation layer sends the final executable instruction generated in step S1562 to the various execution components of the intelligent interaction system through the corresponding physical or logical interfaces. After receiving the text and parameters, the speech synthesis engine begins synthesizing speech and broadcasting it. Upon receiving the instruction frame, the device control gateway forwards it to the target home appliance. After receiving the data, the application interface updates its display. Through these operations, the intelligent interaction system presents the final interactive response to the user according to the instructions, with specific content, rhythm, and path.

[0137] Step S1565: The complete architecture of the personalized model receives the running status information after the intelligent interaction system executes the scene-adaptive intelligent interaction response command through the collaborative coordination process, feeds it back to the dynamic adaptation optimization link, and starts a new round of logic optimization. The scene-adaptive intelligent interaction response command can continuously adapt to the real-time changing scene elements and user interaction needs.

[0138] After the command is executed, each execution component of the intelligent interaction system returns execution status information, such as voice broadcast completion, successful device control, interface rendering completion, and any exceptions during execution. This status information is collected by the collaborative coordination process. The collaborative coordination process packages this status information, along with all records of the interaction from start to finish, and feeds it back to the dynamic adaptation and optimization link through a preset interface. Upon receiving this information, the dynamic adaptation and optimization link uses it as new input to evaluate the final effect of the response and may transform it into new feedback items, adding them to the scene feedback linkage set, thereby triggering another round of logical analysis and optimization iteration. Through this continuous closed-loop process, it is ensured that the core logic of the personalized model can dynamically evolve with changes in user habits and scenarios, enabling each output scene-adaptive intelligent interaction response command to more accurately meet the user's real needs in the current scenario.

[0139] Figure 2 This illustration shows a personalized model building system 100 for intelligent interaction provided in an embodiment of this application. It includes a processor 1001, a memory 1003, and program code stored in the memory 1003. The processor 1001 executes the program code to implement the steps of a personalized model building method for intelligent interaction. The processor 1001 and the memory 1003 are connected, for example, via a bus 1002. Optionally, the personalized model building system 100 for intelligent interaction may further include a transceiver 1004, which can be used for data interaction between this personalized model building system for intelligent interaction and other personalized model building systems for intelligent interaction, such as sending and / or receiving data. It should be noted that in actual scheduling, the transceiver 1004 is not limited to one, and the structure of this personalized model building system 100 for intelligent interaction does not constitute a limitation on the embodiments of this application. The memory 1003 is used to store the program code for executing the embodiments of this application, and its execution is controlled by the processor 1001. The processor 1001 is used to execute program code stored in the memory 1003 to implement the steps shown in the foregoing method embodiments.

[0140] This application provides a computer-readable storage medium storing program code, which, when executed by a processor, can implement the steps and corresponding content of the aforementioned method embodiments.

[0141] The above description is only an optional implementation method for some implementation scenarios of this application. It should be noted that for those skilled in the art, other similar implementation methods based on the technical concept of this application, without departing from the technical concept of this application, also fall within the protection scope of the embodiments of this application.

Claims

1. A method for constructing personalized models for intelligent interaction, characterized in that, The method includes: This process extracts scene elements and user interaction requests from intelligent interaction scenarios, analyzes the bidirectional linkage between scene elements and user interaction requests, and generates basic logic for scene intent linkage. The scene elements include interaction environment elements, user operation elements, and interaction target elements. The user interaction requests include information acquisition requests, function usage requests, opinion expression requests, and emotional communication requests. The bidirectional linkage is the triggering correspondence between scene elements and user interaction requests, and the adaptation requirement relationship between user interaction requests and scene elements. The basic logic for scene intent linkage is an executable logic set containing scene element combination rules, user interaction request matching rules, and intelligent interaction response invocation rules, used to match corresponding user interaction requests based on scene element combinations and generate adapted intelligent interaction responses. A nested feedback collection chain is constructed, embedding the basic logic of scene intent linkage into the core execution node of the nested feedback collection chain. Through the layered collection channels of the nested feedback collection chain, direct feedback information and indirect feedback information of users to intelligent interaction responses, as well as the adaptation and association information between scene elements and intelligent interaction responses, are collected to generate a scene feedback linkage set. The nested feedback collection chain is a layered data collection chain containing a direct feedback collection layer, an indirect feedback collection layer, and an adaptation status collection layer. Each layer is equipped with an independent collection channel and cross-layer data association rules. The scene feedback linkage set is a structured data set containing user feedback content, scene element adaptation status, feedback background information, and feedback type identifier. Using the aforementioned scene feedback linkage set as the driving source, a dynamic adaptation iteration link is constructed. Through multiple rounds of closed-loop iteration of the dynamic adaptation iteration link, the scene element combination rules and user interaction request matching rules in the basic logic of scene intent linkage are adjusted to generate a dynamic adaptation optimization link. The dynamic adaptation iteration link is a closed-loop iteration link that includes an adjustment instruction generation module, a logic modification execution module, an optimization effect analysis module, and a feedback loop input module. The dynamic adaptation optimization link is an executable iterative optimization logic set that includes the iteratively optimized scene intent linkage logic, iterative adjustment rules, and a feedback driving mechanism. A collaborative operation process is constructed between the personalized model architecture and the dynamic adaptation and optimization link. The adjustment logic of the dynamic adaptation and optimization link is embedded into the corresponding hierarchical structure of the personalized model architecture to generate a complete personalized model architecture. The complete personalized model architecture is a hierarchical model architecture that includes a linkage adaptation layer, a feedback processing layer, and an iterative optimization layer. Preset data transmission rules, triggering conditions, and interaction methods are set between each layer. The linkage adaptation layer is used to generate intelligent interactive responses, the feedback processing layer is used to process feedback data to generate adjustment driving data, and the iterative optimization layer is used to update the core logic of the model. Through the complete architecture of the personalized model, real-time scene elements and user interaction requests are received, and scene-adaptive intelligent interaction response instructions are output. The scene-adaptive intelligent interaction response instructions dynamically adjust the response content, response rhythm and interaction path of the intelligent interaction according to the real-time scene elements and user interaction requests, so as to complete the personalized adaptation of the intelligent interaction.

2. The personalized model construction method for intelligent interaction according to claim 1, characterized in that, The process of extracting scene elements and user interaction requests from intelligent interaction scenarios, analyzing the two-way linkage between scene elements and user interaction requests, and generating basic logic for scene intent linkage includes: Collect various environmental information related to user interaction behavior in intelligent interaction scenarios, filter and integrate them to form a set of interactive environment elements, bind each interactive environment element to the corresponding scene presentation details, and generate a set of interactive environment element detail descriptions. Each interactive environment element in the set of interactive environment elements corresponds to the specific environmental presentation of the intelligent interaction scenario, including the spatial attributes of the interaction, the specific state of the surrounding environment, and the interactive support conditions that the environment can provide. Record various operation-related information generated by users during intelligent interaction, filter and integrate them to form a set of user operation elements, associate each user operation element with the corresponding interaction action scenario, and generate a set of user operation element scenario associations. Each user operation element in the set of user operation elements corresponds to the user's specific interaction action, including the way the user starts the interaction, the execution order of the operation instructions, the pause state in each interaction stage, and the operation content that is repeatedly executed. Collect information related to various goals that users expect to achieve during intelligent interaction, filter and integrate them to form a set of interaction goal elements, bind each interaction goal element with its corresponding interaction purpose, and generate a set of interaction goal element purpose associations. Each interaction goal element in the set of interaction goal elements corresponds to the user's specific interaction purpose, including the user's core interaction needs, the hierarchical division of the needs, and the scenario support conditions required to achieve the goal. Determine the interaction relationships among the set of detailed descriptions of interactive environment elements, the set of scene associations of user operation elements, and the set of purpose associations of interactive target elements, and generate a total set of scene element associations. The interaction relationships include the way interactive environment elements affect user operation elements, the driving path of user operation elements on interactive target elements, and the adaptation requirements of interactive target elements to interactive environment elements. Collect various user interaction requests generated during the intelligent interaction process, describe the specific content of each user interaction request in detail, and generate a detailed set of user interaction requests. The detailed set of user interaction requests covers all possible types of user requests in the intelligent interaction scenario. Based on the overall set of scene elements and user interaction requests, scene request response rules are constructed and basic logic for scene intent linkage is generated.

3. The personalized model construction method for intelligent interaction according to claim 2, characterized in that, The process of constructing scene request response rules and generating basic logic for scene intent linkage based on the overall set of scene element associations and user interaction requests includes: Analyze the correspondence between various combinations of scene elements in the overall set of scene element associations and each user interaction request in the detailed set of user interaction requests, determine the combination of scene elements that can trigger each user interaction request, and generate a set of scene element request triggers. Each combination of scene element request triggers includes at least two elements from the following: interaction environment elements, user operation elements, and interaction target elements. Based on historical interaction data, interactive response logic rules are generated corresponding to each scenario element demand trigger combination, forming a scenario demand response rule set. The interactive response logic rules are formed based on the effective response process that can meet the corresponding user interaction demand in the historical interaction data, including the organization method of response content, the execution order of response instructions, and the connection form of interaction links. Extract universally applicable response logic modules from the set of scenario demand response rules, and integrate them to form a set of general response logic modules. The set of general response logic modules is suitable for responding to various scenario element demand trigger combinations. For specific scenario element trigger combinations in the set of scenario request response rules, supplement the exclusive response logic module to generate an exclusive response logic module set. The exclusive response logic module set can meet the personalized response needs in special scenarios. The set of general response logic modules, the set of dedicated response logic modules, and the set of scene element request triggers are associated to determine the calling method of the response logic module corresponding to each combination of scene element request triggers. The basic logic of scene intent linkage is generated, which is used to call the corresponding response logic module to generate intelligent interactive response based on the specific combination of scene elements.

4. The personalized model construction method for intelligent interaction according to claim 1, characterized in that, The construction of the nested feedback collection link involves embedding the basic logic of scene intent linkage into the core execution node of the nested feedback collection link. Through the layered collection channels of the nested feedback collection link, direct and indirect user feedback information to intelligent interaction responses, as well as the adaptation and association information between scene elements and intelligent interaction responses, are collected to generate a scene feedback linkage set, including: A hierarchical structure for nested feedback acquisition links is constructed, which has multiple levels, including a direct feedback acquisition layer, an indirect feedback acquisition layer, and an adaptation state acquisition layer. Each level is equipped with an independent acquisition channel and data transmission path, and the data acquired by each level can be transmitted independently and are interconnected. The core execution node of the nested feedback collection link embeds the basic logic of scene intent linkage into the core execution node. The core execution node can trigger the feedback collection operation at each level based on the intelligent interactive response generated by the basic logic of scene intent linkage. The core execution node establishes a real-time data interaction channel with each collection level. A feedback information collection unit is set up in the direct feedback collection layer to collect the user's direct expression after receiving the intelligent interaction response in real time, including the user's expression of approval of the response content, suggestions for adjusting the response form, expression of questions about the interaction process, and operation instructions to terminate the interaction. Each direct feedback information is bound to the corresponding intelligent interaction response to generate a direct feedback response association set. A behavior trajectory monitoring unit is set up in the indirect feedback collection layer to monitor the changes in the user's behavior trajectory in the intelligent interaction process in real time, including the behavior of restarting the interaction after the interaction is interrupted, the operation of actively adjusting the interaction path, and the behavior of repeatedly expressing the same request. Each indirect feedback information is associated with the corresponding scene elements to generate an indirect feedback scene association set. An adaptation status analysis unit is set up in the adaptation status acquisition layer to analyze the adaptation status of intelligent interaction response and current scene elements in real time, including the consistency between response content and interaction environment elements, the synchronization between response rhythm and user operation elements, and the conformity between response process and interaction target elements, and generate scene element response adaptation set. Based on the associated data collected at each level, a feedback data association system is constructed and a set of scenario feedback linkages is generated.

5. The personalized model construction method for intelligent interaction according to claim 4, characterized in that, The process of constructing a feedback data association system and generating a scenario feedback linkage set based on the associated data collected at each level includes: Construct a mapping relationship between the direct feedback response association set and the indirect feedback scenario association set, extract feedback information for the same intelligent interaction process from the direct feedback response association set and the indirect feedback scenario association set, and generate a comprehensive user feedback set. The mapping relationship is established based on the time node of the interaction and the combination of scenario elements. Construct the association between the comprehensive set of user feedback and the set of scene element response adaptation, bind each user feedback information with the corresponding scene element adaptation status, and generate a feedback adaptation association set. The feedback adaptation association set is used to reflect the adaptation background generated by user feedback. Supplement the background information in the set of related feedback adaptations, including the specific time when the feedback was generated, the corresponding intelligent interaction link, the user identifier of the user participating in the interaction, and the specific status data of the scene elements. Each feedback item contains context information. The feedback information in the feedback adaptation association set is classified and organized according to the feedback type into content optimization feedback, rhythm adjustment feedback, path modification feedback, and adaptation improvement feedback, generating a feedback category set; Integrate all feedback information in the feedback category set, along with associated adaptation status information and background information, to generate a scene feedback linkage set. Each feedback item in the scene feedback linkage set includes user feedback content, scene element adaptation status, feedback background information, and feedback type identifier.

6. The personalized model construction method for intelligent interaction according to claim 1, characterized in that, The process involves using the scenario feedback linkage set as the driving source to construct a dynamic adaptation iteration link. Through multiple rounds of closed-loop iteration of the dynamic adaptation iteration link, the scenario element combination rules and user interaction request matching rules in the basic logic of scenario intent linkage are adjusted to generate a dynamic adaptation optimization link, including: Extract the user feedback content and scene element adaptation status corresponding to each feedback item from the scene feedback linkage set, analyze the problem points in the scene intent linkage basic logic pointed to by each feedback item, and generate a logic adjustment target set. The logic adjustment target set includes the scene element combination rules and user interaction demand matching logic that need to be adjusted. Based on the set of logical adjustment directions, the iterative optimization direction of the dynamic adaptation iteration link is determined. The iterative optimization direction includes optimizing the accuracy of the correspondence between the combination of scene elements and user interaction requirements, improving the adaptation status of intelligent interaction response and scene elements, and improving the rationality of calling response logic modules, thereby generating a set of iterative optimization directions. The core architecture for building a dynamic adaptation and iteration chain includes an adjustment instruction generation module, a logic modification execution module, an optimization effect analysis module, and a feedback loop input module. Input the set of logical adjustment directions and the set of iterative optimization directions into the adjustment instruction generation module to generate specific adjustment instructions for matching the logic of scene element combination rules and user interaction requirements. The adjustment instructions include the specific content, adjustment method and adjustment scope of the adjustment, and generate a set of logical adjustment instructions. Input the set of logic adjustment instructions into the logic modification execution module, modify the corresponding content in the basic logic of scene intent linkage according to the adjustment instructions, including adjusting the composition of scene element combination, optimizing the matching relationship between user interaction requests and scene element combination, improving the calling conditions of response logic module, and generating preliminary optimized scene intent linkage logic. Based on the verification of optimization results, multiple rounds of iterative adjustments were completed, generating iterative trajectories and dynamically adapted optimization links.

7. The personalized model construction method for intelligent interaction according to claim 6, characterized in that, The process of completing multiple rounds of iterative adjustments based on optimization effect verification, generating iterative trajectories and dynamically adapted optimization links, includes: The module for analyzing the optimization effect of the initial optimization of the scene intent linkage logic input is combined with relevant feedback information in the scene feedback linkage set to analyze the improvement status of the logic after the initial optimization in terms of meeting user interaction needs and adapting to scene elements, and to generate optimization effect analysis results. Based on the optimization effect analysis results, the improvement status of the initial optimization scenario intent linkage logic in each direction of the iterative optimization direction set is judged in turn. If there is a direction that has not reached the preset improvement threshold, the analysis result of the corresponding direction is fed back to the feedback loop input module, which generates a supplementary adjustment direction for that direction and re-inputs the adjustment instruction generation module. Repeat the iterative process until the generated optimized scene intent linkage logic reaches the preset improvement threshold in all directions of the iterative optimization direction set, and generate the final optimized scene intent linkage logic. Record all adjustment information during the dynamic adaptation iteration process, including the feedback source, adjustment instruction content, modified logic, and optimization effect analysis results for each adjustment, and generate an iterative adjustment trajectory set; By integrating the final optimized scenario intent linkage logic with the iterative adjustment trajectory set, the patterns of iterative adjustments and the action paths of feedback information are extracted to generate a dynamically adapted optimization link.

8. The personalized model construction method for intelligent interaction according to claim 1, characterized in that, The collaborative operation process of constructing the personalized model architecture and the dynamic adaptation and optimization link embeds the adjustment logic of the dynamic adaptation and optimization link into the corresponding hierarchical structure of the personalized model architecture to generate a complete personalized model architecture, including: Extract the final optimized scenario intent linkage logic from the dynamic adaptation optimization link and use it as the core logic of the linkage adaptation layer of the personalized model architecture. The linkage adaptation layer is used to receive real-time scenario elements and user interaction requests, and generate corresponding intelligent interaction response initial instructions. Extract the scene feedback linkage set processing logic in the dynamic adaptation optimization link and use it as the core logic of the feedback processing layer of the personalized model architecture. The feedback processing layer is used to receive user feedback information and scene element adaptation status information and generate logic adjustment driving data. Extract the iterative adjustment logic from the dynamic adaptation optimization link and use it as the core logic of the iterative optimization layer of the personalized model architecture. The iterative optimization layer is used to receive the logic adjustment driving data generated by the feedback processing layer and optimize and adjust the core logic of the linkage adaptation layer. Construct a hierarchical connection process between the linkage adaptation layer, feedback processing layer, and iterative optimization layer, determine the data transmission format, triggering conditions, and interaction methods between each layer, including the linkage adaptation layer transmitting preliminary instructions for intelligent interactive response and corresponding scene element information to the feedback processing layer, the feedback processing layer transmitting logic adjustment driving data to the iterative optimization layer, the iterative optimization layer transmitting logic optimization instructions to the linkage adaptation layer, and generating a set of hierarchical connection rules. Add a data storage layer to the personalized model architecture. The data storage layer is used to store data related to scene elements, data related to user interaction requests, data related to user feedback, iterative trajectory data of the dynamic adaptation and optimization link, and running status data of each layer, and generate data storage management rules. Add additional functional layers to the model architecture, complete the architecture integration and testing, and generate a complete personalized model architecture.

9. The personalized model construction method for intelligent interaction according to claim 1, characterized in that, The process, through the complete architecture of the personalized model, receives real-time scene elements and user interaction requests, and outputs scene-adaptive intelligent interaction response instructions, including: The input adaptation layer receives real-time scene element data and user interaction request data from external sources, performs format conversion and standardization processing on the data according to the input data adaptation rules, and generates adapted real-time scene element data and adapted user interaction request data. The adapted real-time scene element data and adapted user interaction request data are transmitted to the linkage adaptation layer. The linkage adaptation layer analyzes the correspondence between the two types of data based on the final optimized scene intent linkage logic, and generates the initial intelligent interaction response instructions. The initial intelligent interaction response instructions include the initial response content, the initial response rhythm plan, and the initial interaction path plan. The linkage adaptation layer transmits the initial intelligent interaction response instructions and the corresponding adapted real-time scene element data to the feedback processing layer, and at the same time transmits the initial intelligent interaction response instructions to the output adaptation layer for initial format optimization. The feedback processing layer receives feedback information from the user regarding the initial instructions of the intelligent interaction response and real-time scene element adaptation status information. According to the scene feedback linkage set processing logic, it analyzes the adjustment dimension indicated by the feedback information and the adaptation dimension indicated by the real-time scene element adaptation status information, establishes the correlation between the two types of information in the corresponding dimensions, and generates logical adjustment driving data. The logic adjustment driving data is transmitted to the iterative optimization layer. Based on the iterative adjustment logic, the iterative optimization layer generates optimization adjustment instructions for the core logic of the linkage adaptation layer and transmits them to the linkage adaptation layer. Based on logic adjustment, the data optimization response instructions and model core logic are optimized to complete intelligent interactive response output and continuous iteration.

10. A personalized model building system for intelligent interaction, characterized in that, The method includes a processor and a computer-readable storage medium storing machine-executable instructions that, when executed by the processor, implement the personalized model construction method for intelligent interaction as described in any one of claims 1-9.

Images