Smart home interaction processing method and device

By using a small-parameter intent recognition model for pre-classification and differentiated routing in smart home systems, the efficiency and accuracy issues of existing systems in handling diverse interaction requests are solved, achieving efficient and low-latency smart home interaction processing.

CN123089682APending Publication Date: 2026-11-13SHANGHAI XIAODU TECHNOLOGY CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202611328786.9
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2026-08-28
Publication Date
2026-11-13

AI Technical Summary

Technical Problem

Existing smart home interaction systems suffer from problems such as high computing power consumption, high response latency, and susceptibility to misjudgments in device control scenarios when handling diverse user interaction requests, especially in remote control and non-device control business processing.

Method used

A small-parameter intent recognition model is used for pre-semantic understanding and intent classification. Through a differentiated routing and distribution mechanism, interaction requests are allocated to the device side, application side, or open domain for processing, and are handled by the corresponding modules to realize device control, service response, and open domain dialogue generation.

Benefits of technology

It reduces overall computing power consumption, decreases response latency, improves the efficiency and accuracy of interactive processing, and supports users in conveniently handling diverse business needs remotely.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN123089682A_ABST
    Figure CN123089682A_ABST
Patent Text Reader

Abstract

This disclosure provides a smart home interaction processing method and apparatus, relating to the field of artificial intelligence technology, and particularly to the fields of smart home human-computer interaction, natural language intent recognition, and multi-path routing distribution technology. The method includes: employing an intent recognition model to classify intents based on user-inputted smart home service-oriented interaction requests, determining intent categories, wherein the number of parameters in the intent recognition model is less than a preset parameter threshold; performing differentiated routing distribution processing based on intent categories to obtain processing results; and sending the processing results to the user.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This disclosure relates to the field of artificial intelligence technology, and in particular to the fields of smart home human-computer interaction, natural language intent recognition, and multi-path routing and distribution technology. Background Technology

[0002] As the number of smart home devices continues to grow, users' needs for device control, usage analysis, content retrieval, and convenient interaction are becoming increasingly diverse. Direct device interaction is the basic interaction mode for smart homes and can well meet near-field usage scenarios. In practical applications, users often find themselves far from the device, or their business needs exceed the device's own capabilities. Interaction requests include various types such as device control, report retrieval, customer service consultation, and knowledge Q&A.

[0003] Currently, there are various implementation methods in the field of smart home interaction. Mainstream smart home applications can handle device discovery, network configuration, status display, and remote control command forwarding; non-device control functions are implemented through independent pages or third-party entry points. Some smart assistants use a single, high-parameter model to process all natural language input, relying on this model to autonomously parse and respond to user requests. Smart speakers, smart screens, and other terminals are equipped with local voice recognition modules, using preset command templates to match commands and achieve home interaction and device control. Summary of the Invention

[0004] This disclosure presents a smart home interaction processing method and apparatus.

[0005] In a first aspect, embodiments of this disclosure propose a smart home interaction processing method, comprising: employing an intent recognition model to classify intents based on user-inputted smart home service-oriented interaction requests, determining intent categories, wherein the number of parameters in the intent recognition model is less than a preset parameter threshold; performing differentiated routing distribution processing based on the intent category to obtain processing results; and sending the processing results to the user; wherein performing differentiated routing distribution processing based on the intent category to obtain processing results includes: in response to determining that the intent category is a device-side processing intent, performing device control-type service processing to obtain device execution results; in response to determining that the intent category is an application-side processing intent, performing service processing on the application hub side to obtain service response results; and in response to determining that the intent category is a default intent, performing open-domain dialogue generation processing to obtain open-domain response results.

[0006] Secondly, embodiments of this disclosure propose a smart home interaction processing device, comprising: an intent classification module configured to use an intent recognition model to classify intents based on user-inputted smart home service-oriented interaction requests, and determine intent categories, wherein the number of parameters of the intent recognition model is less than a preset parameter threshold; a routing distribution module configured to perform differentiated routing distribution processing based on intent categories to obtain processing results; and a result sending module configured to send the processing results to the user; wherein the routing distribution module is further configured to: in response to determining that the intent category is a device-side processing intent, perform device control-type service processing to obtain device execution results; in response to determining that the intent category is an application-side processing intent, perform service processing on the application hub side to obtain service response results; and in response to determining that the intent category is a default intent, perform open-domain dialogue generation processing to obtain open-domain response results.

[0007] Thirdly, embodiments of this disclosure provide an electronic device, including: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to enable the at least one processor to perform the method as described in the first aspect.

[0008] Fourthly, embodiments of this disclosure provide a non-transitory computer-readable storage medium storing computer instructions for causing a computer to perform the methods described in the first aspect.

[0009] Fifthly, embodiments of this disclosure provide a computer program product including a computer program that, when executed by a processor, implements the method described in the first aspect.

[0010] The key or essential features of the embodiments disclosed herein are not intended to limit the scope of this disclosure. Other features of this disclosure will become readily apparent from the following description. Attached Figure Description

[0011] Other features, objects, and advantages of this disclosure will become more apparent from the following detailed description of non-limiting embodiments with reference to the accompanying drawings. The drawings are provided for a better understanding of the invention and are not intended to limit the scope of this disclosure. Wherein: Figure 1 This is a flowchart of one embodiment of the smart home interaction processing method according to the present disclosure; Figure 2 This is a flowchart of yet another embodiment of the smart home interaction processing method according to the present disclosure; Figure 3 This is a schematic diagram of the overall process of the smart home interaction processing method according to an embodiment of the present disclosure; Figure 4This is a schematic diagram of the routing distribution for the first-level classification of the pre-identification model in this embodiment of the present disclosure; Figure 5 This is a schematic diagram of the routing distribution for the second-level classification of the pre-identification model in this embodiment of the present disclosure; Figure 6 This is a schematic diagram of the internal processing mechanism of the requirement satisfaction model in the embodiments of this disclosure; Figure 7 This is a schematic diagram of the structure of an embodiment of the smart home interaction processing device according to the present disclosure; Figure 8 This is a block diagram of an electronic device used to implement the smart home interaction processing method of the embodiments of this disclosure. Detailed Implementation

[0012] The exemplary embodiments of this disclosure are described below with reference to the accompanying drawings, including various details of the embodiments to aid understanding, and should be considered merely exemplary. Therefore, those skilled in the art will recognize that various changes and modifications can be made to the embodiments described herein without departing from the scope and spirit of this disclosure. Similarly, for clarity and brevity, descriptions of well-known functions and structures are omitted in the following description.

[0013] It should be noted that, unless otherwise specified, the embodiments and features described in this disclosure can be combined with each other. This disclosure will now be described in detail with reference to the accompanying drawings and embodiments.

[0014] Figure 1 A flowchart 100 of an embodiment of a smart home interaction processing method according to the present disclosure is shown. The smart home interaction processing method includes the following steps: Step 101: Using an intent recognition model, intents are classified based on user-inputted interaction requests for smart home services to determine the intent category.

[0015] This step serves as the business entry point for the interactive processing flow. It receives the original interactive requests for smart home services, categorizes the intents to determine the intent type, and uses this as the basis for subsequent routing and distribution. This eliminates the constraints of the physical location of smart devices, allowing users to initiate smart home-related service requests anytime, anywhere via mobile devices.

[0016] The user interaction layer is deployed in the smart home app (Application), which is a mobile presentation layer. It is responsible for receiving user voice or text input and feeding back the processing results to the user in the form of text, cards, multiple images, etc. It is the unified interaction entry point for users.

[0017] Interaction requests are voice or text commands input by users for smart home services. These can be device control commands, such as "turn off the bedroom light," or non-control commands, such as "view this week's speaker usage report" or "inquire about device malfunctions."

[0018] Intent classification is the process of semantically parsing a user's original interaction request, identifying the user's true business needs, and outputting intent category labels.

[0019] The intent category is a business label obtained by classifying user business requests. In this solution, it is at least divided into device-side processing intent, application-side processing intent, and default intent.

[0020] The intent recognition model has fewer parameters (e.g., 1.4B) than a preset parameter threshold, classifying it as a small-parameter pre-recognition model trained on a scenario-based dataset from the smart home domain. The preset parameter threshold distinguishes between small and large models. This model is specifically responsible for pre-recognition semantic understanding and intent classification, and does not handle complex content generation tasks. It should be noted that other intent recognition models trained on smart home datasets can also achieve this function; this embodiment does not limit its implementation. Unlike existing solutions that directly use large models to process all requests, using a small model specifically for intent classification results in faster inference speeds and lower computational consumption, making it suitable as a pre-filtering unit.

[0021] In some optional implementations of this embodiment, the interaction request is standardized to obtain standardized input information; the standardized input information is then semantically understood and classified using an intent recognition model to determine the intent category.

[0022] The interaction standardization layer, which belongs to the front-end access layer, is used to normalize and standardize user input from different sources and in different formats, and to shield the format differences caused by different input channels.

[0023] Standardized input information is text information that has been formatted, verbal redundancy removed, and noise reduced, making it usable by the intent recognition model. For example, a user's spoken input, "Uh, can you check how my child is using the speaker this week?", after standardization, yields the formatted text "Check how my child is using the smart speaker this week," facilitating subsequent semantic recognition by the model.

[0024] In some optional implementations of this embodiment, the text request input by the user is standardized to obtain standardized user input information; or the voice request input by the user is converted into a corresponding text request, and the text request is standardized to obtain standardized user input information.

[0025] The speech parsing module, belonging to the speech processing layer, is responsible for converting user speech input into standardized text. When the user inputs speech, speech-to-text conversion is performed first, followed by text normalization; when the user directly inputs text, speech-to-text conversion is skipped, and normalization is performed directly. For example, when a user presses the record button in the smart home app to output a voice command, the speech parsing module completes speech recognition and outputs the original text, while the interaction normalization layer removes interjections and meaningless filler words, outputting standardized input information.

[0026] Step 102: Perform differentiated route distribution processing based on intent category to obtain the processing result.

[0027] This step is executed collaboratively by the intent understanding and routing layer and the agent scheduling layer. Based on the identified intent categories, user requests are allocated to matching business processing links, employing a three-way distribution mechanism: device-side processing link, application central-side processing link, and default fallback link. This step addresses the technical problems of high computational overhead, high response latency, and susceptibility to misjudgments in device control scenarios caused by a single large model directly processing all requests. By classifying and routing intents beforehand, different requests are handled by appropriate links, balancing response speed, computational cost, and response quality.

[0028] Among them, the intent understanding and routing layer belongs to the semantic understanding module. After completing the intent parsing, it coordinates task routing and modules to execute branch routing decisions on the device side or the application side.

[0029] The agent scheduling layer belongs to the task execution layer. Based on the routing results, it calls the corresponding agent, demand fulfillment model, or business module to carry out business processing.

[0030] Routing and distribution refers to assigning user requests to matching business processing links based on the identified intent category.

[0031] Specifically, in response to determining that the intent category is a device-side processing intent, device control-type business processing is executed to obtain the device execution result; in response to determining that the intent category is an application-side processing intent, business processing is executed on the application hub side to obtain the business response result; in response to determining that the intent category is a default intent, open domain dialogue generation processing is executed to obtain the open domain response result.

[0032] For device-side processing intents, such business requests require issuing instructions to physical smart devices for execution. Typical business scenarios include turning lights on / off and adjusting air conditioning temperature. Instructions are issued to the target smart device through a device control link or a dedicated smart agent. The device execution result includes an instruction execution status identifier, the updated device status, or error messages. For example, the output might be "Execution successful, bedroom light is off" or "Execution failed, device offline." For instance, if a user inputs "Turn off the bedroom light," this is recognized as a device-side processing intent, and device control-related business processing is executed.

[0033] For application-side processing intents, this type of service does not require controlling smart hardware devices. The service is completed directly within the smart home app's central processing loop. This includes subcategories such as usage report queries, smart customer service consultations, information Q&A, casual conversation, and app function activation. The service response is generated based on a demand fulfillment model or the corresponding business module, without needing to issue control commands to the smart devices. For example, if a user enters "Check my child's smart speaker usage this week," this is recognized as an application-side processing intent, and an analysis report is directly generated at the app's central processing level.

[0034] For default intents, if the user input does not match any of the preset device-side or application-side proprietary services, the open-domain general model will be used to generate an open-domain response, ensuring system availability across all scenarios. For example, if a user inputs "Tell me a joke," it is recognized as the default intent and enters the open-domain dialogue generation process.

[0035] Step 103: Send the processing result to the user.

[0036] This step is executed by the user interaction layer, and the processing result can be any of the following: device execution feedback, business response result, or open domain reply result. Output formats support text, cards, and structured feedback with multiple graphs. For complex statistical results such as usage reports, pie charts, line charts, and other graphs combined with cards can be used to fully utilize mobile screens and improve readability. For example, a speaker usage report might use cards to display a text summary, along with multiple graphs showing duration distribution and content percentage.

[0037] The smart home interaction processing method provided in this disclosure uses a smaller intent recognition model as a front-end recognition unit to classify the semantic intent of interaction requests, and then performs differentiated three-way routing based on this classification. For different intent categories, a division-of-labor processing mode is adopted, which includes device link execution, application central-side call demand fulfillment model, and open-domain general-purpose large model as a fallback, enabling collaborative work between large and small models. Using a smart home app as the intelligent central hub, it uniformly receives user voice and text interaction requests, eliminating the need for all requests to be processed by a large model, reducing overall computing power overhead and response latency. Finally, it uniformly returns results to the user, overcoming the technical shortcomings of traditional solutions such as fragmented interaction entry points, the need to be physically close to the device to complete operations, and the potential for misjudgment of device control scenarios due to a single large model processing all requests.

[0038] Figure 2 A flow 200 is shown as yet another embodiment of the smart home interaction processing method according to the present disclosure.

[0039] This embodiment relies on an intelligent agent application development platform as its underlying technical support. This platform is used to support the integration of large model components, intelligent agent orchestration, and rapid application deployment. Based on this platform, applications oriented towards large language models can be quickly built, and the deployment, invocation, and joint debugging of intent recognition models and demand fulfillment models can be completed, supporting the implementation of AI (Artificial Intelligence) butler business capabilities in smart home apps.

[0040] This smart home interaction processing method includes the following steps: Step 201: Standardize the interaction request to obtain standardized input information.

[0041] This step corresponds to the processing logic of the interaction standardization layer. Regardless of whether the user's original input is text obtained from speech transcription or text entered directly by the user, the input is standardized in a unified manner to eliminate spoken redundancy and format differences, and output standardized input information with a unified format as input material for the downstream intent recognition model.

[0042] Step 202: Use the intent recognition model to perform semantic understanding and intent classification on the standardized input information to determine the intent category.

[0043] The intent recognition model has fewer parameters (e.g., 1.4B) than a preset parameter threshold, classifying it as a small-parameter pre-recognition model trained on a scenario-based dataset from the smart home domain. The preset parameter threshold distinguishes between small and large models. This model is specifically responsible for pre-recognition semantic understanding and intent classification, and does not handle complex content generation tasks. It should be noted that other intent recognition models trained on smart home datasets can also achieve this function; this embodiment does not limit its implementation. Unlike existing solutions that directly use large models to process all requests, using a small model specifically for intent classification results in faster inference speeds and lower computational consumption, making it suitable as a pre-filtering unit.

[0044] In some optional implementations of this embodiment, standardized input information is input into the intent recognition model, and the confidence score of each candidate intent category is output; in response to determining that the highest confidence score is greater than or equal to a preset confidence threshold, the candidate intent category corresponding to the highest confidence score is determined as the intent category; in response to determining that all confidence scores are less than the preset confidence threshold, it is determined as the default intent.

[0045] The confidence score is the probability score output by the intent recognition model, representing that the input belongs to the corresponding intent category. For example, the preset confidence threshold is 0.8. This mechanism is used to control the misclassification rate and avoid incorrectly distributing user requests with uncertain identification to dedicated business links. For example, if the highest confidence score is 0.85, which is greater than the threshold of 0.8, then this candidate intent is adopted; if the confidence scores of all candidate intents are only 0.6 and 0.55, which are both lower than the threshold, then they are determined to be the default intent and handled by the open-domain large model as a fallback.

[0046] In some optional implementations of this embodiment, the standardized input information is classified at the first level using an intent recognition model to obtain a first-level intent category; in response to determining that the first-level intent category is an application-side processing intent, the standardized input information is classified at the second level using an intent recognition model to obtain a second-level intent category. The second-level intent category includes, but is not limited to, at least one of the following: usage report, function activation, intelligent customer service, information Q&A, casual chat, etc.

[0047] This embodiment employs a two-level hierarchical classification mechanism, where each level can be independently configured with a confidence threshold to achieve hierarchical fault tolerance. The first level categorizes the main categories. These first-level intent categories include, but are not limited to, device-side processing intents, application-side processing intents, and default intents. Application-side processing intents include, but are not limited to, requests suitable for direct processing at the App's central processing unit, such as usage reports, feature activation, intelligent customer service, information Q&A, and casual conversation. Device-side processing intents include, but are not limited to, requests that can be processed by the device control link or a dedicated intelligent agent, such as device control and weather. Default intents represent requests that do not match a specific business category and require entry into the open domain fallback processing link. Once an application-side processing intent is identified, the second level of subdivision is executed, distinguishing between sub-intents such as usage reports, feature activation, and intelligent customer service. Different sub-intents are matched with specific prompts and business logic.

[0048] Table 1 shows examples of intent categories, exemplary identification identifiers, and corresponding follow-up processing methods:

[0049] It should be noted that Table 1 is only an exemplary embodiment of this disclosure. The intent category and identification identifier can be added, deleted or adjusted according to the actual business scenario, and this disclosure does not limit them.

[0050] Step 203: In response to determining that the intent category is a device-side processing intent, the device control link or a dedicated intelligent agent is invoked to send control commands to the target intelligent device and receive the device execution results returned by the target intelligent device after executing the control commands.

[0051] The device control link is the communication link between the App's intelligent hub and the physical intelligent device. A dedicated intelligent agent is an execution unit encapsulated for specific device capabilities. Control commands carry the target device ID and specific action parameters. For example, if a user inputs the text "Turn off the bedroom light," the device's processing intent is recognized, and the intelligent agent scheduling layer sends a control command containing the device ID and the action of turning off the light to the bedroom light via the device control link. After the device completes the action, it sends back the execution result via a long connection, informing the App whether the execution was successful or failed.

[0052] Step 204: In response to determining that the intent category is an application-side processing intent, the requirement satisfaction model is invoked to perform a business response and obtain the business response result.

[0053] The demand fulfillment model is a high-parameter model (e.g., 20 bytes of parameters), fine-tuned and trained using proprietary business data. For example, the demand fulfillment model targets various sub-businesses on the App side, loading corresponding sub-category-specific prompts to generate content, covering usage report summaries, customer service Q&A, and casual conversation responses. The demand fulfillment model is trained using both Supervised Fine-Tuning (SFT) and Retrieval-Augmented Generation (RAG) techniques. SFT is used for supervised training and prompt optimization in smart home business scenarios, guiding the model to generate results that meet business requirements and improving the accuracy and generalization ability in tasks such as report summaries, resource analysis, and customer service Q&A. RAG retrieves information related to the input content from knowledge bases, device manuals, or relevant text resources, and combines this information with the generation model to organize and output the results, thereby meeting the high-quality answer requirements in open-ended question-and-answer and knowledge-based scenarios. For example, if the sub-intent is identified as a usage report, 7 days of offline statistical device historical interaction data is retrieved and fed into the demand fulfillment model to generate a summary result including usage duration and content distribution.

[0054] In some optional implementations of this embodiment, the standardized input information is converted into a vector representation; candidate text fragments that meet the preset conditions for similarity with the vector representation are retrieved from the pre-built knowledge base vector index; prompt words are constructed based on the candidate text fragments and the standardized input information; and the prompt words are input into the demand fulfillment model to obtain the business response result.

[0055] The knowledge base vector index stores structured and unstructured text resources such as device manuals, customer service tickets, and FAQs (Frequently Asked Questions). Preset conditions can be, for example, Top N similarity or similarity exceeding a preset similarity threshold. The RAG execution flow is as follows: After the user query is vectorized, candidate text fragments meeting the similarity threshold are retrieved from the vector index. These retrieved fragments are then concatenated with the original user request to construct prompts, which are then fed into the demand fulfillment model. The model outputs an answer based on real knowledge base materials, ensuring that the answer content is based on traceable knowledge sources and reducing fabricated illusion questions. This process is often used in intelligent customer service consultation services.

[0056] Step 205: In response to determining the intent category as the default intent, the open domain general model is invoked to perform semantic understanding and dialogue content generation, and the open domain response result is obtained.

[0057] Among them, the open-domain general-purpose model is specifically designed to handle default intents that do not match the preset proprietary business logic, enabling general knowledge-based question answering and casual conversation as a fallback. For example, if a user enters "Tell me a joke," it is determined to be the default intent, and the open-domain general-purpose model generates a casual conversation response, ensuring the system's availability across various scenarios.

[0058] Step 206: Send the processing result to the user.

[0059] In this embodiment, the specific operation of step 206 has been described. Figure 1 Step 103 in the illustrated embodiment is described in detail and will not be repeated here.

[0060] The smart home interaction processing method provided in this disclosure fully demonstrates the complete chain from small model intent recognition, confidence level judgment, and two-level intent classification, to the execution of device control, demand fulfillment model business response, and open domain large model fallback, and finally the return of results. This reflects a collaborative processing mechanism of small and large models, with the small model handling pre-recognition and the large model and intelligent agent responsible for fine-grained business processing. This disclosure upgrades the smart home application from a traditional device remote control to an intelligent hub with independent semantic understanding capabilities, constructing a unified interaction entry point capable of intelligently distributing and closing-loop processing diverse user requests such as device control, usage analysis, customer service consultation, and casual conversation. For complex business output content such as reports and statistics, a structured format combining cards and multiple images can be used to provide feedback to users, fully adapting to the display characteristics of mobile devices and improving information readability. The entire processing architecture balances business response efficiency and computing power consumption, ensuring the accuracy of dedicated business processing while also being compatible with generalized question-and-answer scenarios, thus improving the overall human-computer interaction experience in smart home scenarios.

[0061] The actual execution process of the method in this embodiment will be further explained below with reference to several exemplary business implementation scenarios.

[0062] The first exemplary scenario is device control services via the device side path. Users can issue device control requests in a smart home app using natural language. After the system completes intent recognition, parameter extraction, and action summarization, the device control link sends instructions to the target smart device. Users can control the device without having to approach the physical device or manually operate it, achieving more natural and convenient remote control.

[0063] Example scenario description: A user is in the living room but wants to turn off the bedroom lights.

[0064] Input: The user enters the text "Turn off the bedroom light" in the app.

[0065] Processing procedure: Since the input is text, the speech parsing module is skipped, and the process goes directly to the interaction standardization layer.

[0066] The intent understanding and routing layer identifies the intent as a "device control intent" in the device-side processing intent and completes parameter extraction: target device = bedroom light, action = turn off.

[0067] The intelligent agent scheduling layer sends control commands (device ID, action type) to the target intelligent device through the device control link.

[0068] After the smart device performs the action of turning off the lights, it will send back the result (success / failure) via a long connection.

[0069] The user interaction layer provides the user with a text message indicating that "the bedroom light is off".

[0070] The technical effect of this scenario is that it verifies the improved convenience of users controlling devices without having to approach them, simply by inputting text through the app.

[0071] The second exemplary scenario is the usage report service on the application side. Based on offline-generated user historical data and combined with a demand fulfillment model, the system summarizes, analyzes, and generates multi-dimensional information such as usage duration, application usage, resource consumption, and device interactions, and allows users to continue asking follow-up questions within the app. This service is processed entirely in a closed loop on the application side, requiring no physical smart device involvement, and is applicable to scenarios such as parental care, usage analysis, and behavioral insights. In one embodiment, the demand fulfillment model can be a 20-parameter model.

[0072] Example scenario description: A user wants to know about their child's recent use of the smart speaker.

[0073] Input: Users can use voice input in the smart home app to say, "Let's see what my child has listened to on the smart speaker this week."

[0074] Processing procedure: The voice parsing module converts speech into standardized text: "See what your child listened to on the smart speaker this week."

[0075] The interaction standardization layer standardizes the text, removes colloquial and redundant expressions, and outputs standardized input.

[0076] The intent understanding and routing layer calls the intent recognition model to identify that the request belongs to the "user device dialogue history" category in the application-side processing intent, with the corresponding identifier report_query. When the confidence level reaches a preset threshold (such as 0.85), it is routed to the cloud core processing module.

[0077] The intelligent agent scheduling layer combines the user interaction history data of the smart speaker generated offline over the past 7 days, calls the demand satisfaction model (20B parameter scale) to summarize and generate the output, which includes: the distribution of content types played by the speaker this week (children's songs, stories, Q&A, etc.), cumulative usage time, and distribution of active time periods.

[0078] The results are presented to users through a combination of structured cards and multiple charts, such as pie charts showing the proportion of content types and line charts showing daily usage time. This feedback is provided to users through a user interaction layer, and users are encouraged to ask follow-up questions, such as "Why did you use it for a longer time on Wednesday?"

[0079] The technical effect of this scenario is that the entire process does not require the user to operate a physical device. The closed loop from natural language questioning to structured report generation is completed on the application side, which verifies the feasibility of independent processing capabilities on the application side.

[0080] The third exemplary scenario is intelligent customer service on the application side. The system combines device manuals, knowledge bases, and retrieval-enhanced generation technology to allow users to inquire about device usage details, fault descriptions, and operation instructions via natural language at the user interaction layer. The system then directly generates answers from the demand fulfillment model and provides feedback at the intelligent hub, reducing the cost of manual customer service and improving service response efficiency.

[0081] The fourth exemplary scenario is information Q&A and casual chat services on the application side. The system supports information retrieval-based Q&A and companion-style casual chat scenarios, responding to open questions and role-based dialogue requests under a unified interaction entry point. For questions that fall within the scope of proprietary business functions, the intelligent hub side calls the corresponding business module or requirement-satisfying model for processing; for open questions that do not fall within a clearly proprietary intent, the system calls a general-purpose open-domain model to generate an answer, thereby enabling the entire process of processing and feedback to be completed on the application side, improving user retention and satisfaction.

[0082] Example scenario description: A user casually asks an open question that is unrelated to smart home.

[0083] Input: The user inputs via voice: "Tell me a joke."

[0084] Processing procedure: After the speech parsing module converts the speech into text, the intent recognition model determines that the input does not match any of the preset categories on the application side or the device side, and identifies it as default_intent.

[0085] The intent understanding and routing layer routes requests to the open domain common big model.

[0086] The open-domain general large model generates a joke as the reply content.

[0087] The user interaction layer displays replies in text form and preserves the conversation context to support follow-up casual conversations and questions.

[0088] The technical effect of this scenario is that it verifies that the system can still maintain availability and full-scenario coverage even when it does not hit a specific business scenario, thus avoiding the negative user experience caused by "incomprehensibility".

[0089] The fifth exemplary scenario is the function access service via the application side path. This service allows users to directly state the function name or trigger internal functions of the smart home app through descriptive language. After recognizing the intent, the system directly completes the function jump and invocation within the app, achieving a lower learning cost for app function access.

[0090] The sixth exemplary scenario is a hybrid intent multi-turn session service that coordinates between the device side and the application side. This scenario demonstrates the ability of different routing and distribution logics to work together within the same session.

[0091] Scenario: The user first issues a device control command, and then asks for device usage suggestions, demonstrating the collaborative capability of dual-route distribution within the same session.

[0092] Input 1: User voice input: "Set the living room air conditioner to 26 degrees."

[0093] Processing flow 1: The intent recognition model identifies the intent as a device control intent; The system will send the control parameters (device: living room air conditioner; action: temperature adjustment; target value: 26 degrees Celsius) to the target device through the device control link; After the device executes the command, it sends the result ("The living room air conditioner has been set to 26 degrees") to the App, and a confirmation card is displayed in the user interaction layer.

[0094] Input 2 (continued from the same session): The user then asked, "Does this temperature consume a lot of electricity? Are there any energy-saving suggestions?"

[0095] Processing flow 2: The intent recognition model identifies that the query does not match either the device control intent or the known proprietary business intent, and determines it to be the default intent. This question belongs to open-ended knowledge-based questions and answers, and is outside the preset business scope. The system will route the request to the Open Domain General Large Model; The open-domain general model, combined with basic air conditioning knowledge, generates the following answer: "26 degrees is a relatively energy-efficient setting temperature. Compared with temperatures below 22 degrees, it can reduce power consumption by about 10%-15%. It is recommended to use a timer or sleep mode to further save power." The results are returned to the user interaction layer in text format.

[0096] The technical effect of this scenario is that, within the same natural dialogue flow, the system automatically distinguishes between device control intents and open domain question-and-answer intents, without requiring users to switch applications or function entry points, thus verifying the coherent processing capability of the layered routing architecture in real multi-turn interaction scenarios.

[0097] Figure 3 This diagram illustrates the overall flow of the smart home interaction processing method according to an embodiment of the present disclosure. The diagram fully demonstrates the process starting with user input of a query, passing through a pre-recognition model, and flowing through three main branches: application function intent, device capability intent, and default intent. The application function intent further branches into usage report generation, function activation processing, and intelligent customer service processing; the device capability intent branch includes device control execution and existing capability processing such as weather; the default intent flows to the open domain general model. After each branch is processed, all results are aggregated and returned to the user interaction layer. This diagram macroscopically illustrates the overall architecture of the multi-path routing distribution of the present invention.

[0098] Figure 4 This diagram illustrates the routing and distribution logic of the first-level classification of the pre-recognition model in this embodiment. The diagram shows the routing and distribution logic after the pre-recognition model completes the first-level intent classification. User input queries are sent to the pre-recognition model, which outputs three first-level intents: application function intents, device capability intents, and default intents. Application function intents are distributed to the cloud-based core processing module for user report generation, function activation, and intelligent customer service processing; device capability intents are distributed to the device control link or dedicated intelligent agent for device control execution and weather-related capability processing; default intents are distributed to the open-domain general-purpose large model. After all link processing is completed, the results are uniformly returned to the user interaction layer. This diagram illustrates the first-level distribution logic of the collaboration between the large and small models in this embodiment, where the small model completes the pre-processing, assigning different services to different processing units, rather than all being processed by the large model.

[0099] Figure 5This diagram illustrates the routing and distribution of the second-level classification in the pre-identification model of this embodiment. The diagram also shows the internal processing mechanism of the requirement-fulfilling model, demonstrating that it supports three processing mechanisms: supervised fine-tuning, retrieval-enhanced generation, and open-domain fallback. The supervised fine-tuning mechanism performs supervised training and prompt word optimization for specific scenarios, improving the accuracy of report summaries, resource analysis, and customer service Q&A tasks. The retrieval-enhanced generation mechanism retrieves relevant text resources from the knowledge base and equipment manuals, extracts relevant information, and combines it with the generation model to organize the output results for knowledge-based Q&A scenarios. The open-domain fallback mechanism calls a general open-domain model to generate answers for default inputs, ensuring usability in generalized scenarios. After processing by all three mechanisms, the results are returned to the user interaction layer. Supervised fine-tuning and retrieval-enhanced generation are the core training and inference methods of the requirement-fulfilling model in this embodiment.

[0100] Figure 6 This diagram illustrates the internal processing mechanism of the requirement fulfillment model in this embodiment. It fully demonstrates the execution chain of the layered architecture. User input is first fed into a small model to complete pre-processing intent recognition, outputting a structured intent classification result, which is then passed to the intent understanding and routing layer, and further distributed to the agent scheduling layer. The agent scheduling layer distinguishes three processing chains: the device capability intent processing chain calls the device control chain or a dedicated agent to complete device control; the application function intent processing chain calls the requirement fulfillment model or business module to generate business results such as reports and customer service Q&A; the default intent processing chain calls the open domain general model to generate casual conversation or generalized Q&A responses; after the three chains are completed, the results are uniformly returned to the user interaction layer. This diagram fully reproduces the business flow relationship of the five-layered architecture in this embodiment, which consists of: user interaction layer, voice parsing module, interaction standardization layer, intent understanding and routing layer, and agent scheduling layer.

[0101] Further reference Figure 7 As an implementation of the methods shown in the above figures, this disclosure provides an embodiment of a smart home interaction processing device, which is similar to... Figure 1 Corresponding to the method embodiments shown, this device can be specifically applied to various electronic devices.

[0102] like Figure 7As shown, the smart home interaction processing device 700 of this embodiment may include: an intent classification module 701, a routing distribution module 702, and a result sending module 703. The intent classification module 701 is configured to use an intent recognition model to classify intents based on user-inputted smart home service-oriented interaction requests, determining the intent category. The number of parameters in the intent recognition model is less than a preset parameter threshold. The routing distribution module 702 is configured to perform differentiated routing distribution processing based on the intent category to obtain a processing result. The result sending module 703 is configured to send the processing result to the user. The routing distribution module 702 is further configured to: in response to determining that the intent category is a device-side processing intent, perform device control-type service processing to obtain a device execution result; in response to determining that the intent category is an application-side processing intent, perform service processing on the application hub side to obtain a service response result; and in response to determining that the intent category is a default intent, perform open-domain dialogue generation processing to obtain an open-domain response result.

[0103] In this embodiment, the specific processing of the intent classification module 701, the route distribution module 702, and the result sending module 703 in the smart home interaction processing device 700, and their resulting technical effects, can be found in reference to [reference needed]. Figure 1 The relevant descriptions of steps 101-103 in the corresponding embodiments will not be repeated here.

[0104] In some optional implementations of this embodiment, the intent classification module 701 is further configured to: standardize the interaction request to obtain standardized input information; and use an intent recognition model to perform semantic understanding and intent classification on the standardized input information to determine the intent category.

[0105] In some optional implementations of this embodiment, the intent classification module 701 is further configured to: standardize the text request input by the user to obtain standardized user input information; or convert the voice request input by the user into a corresponding text request, standardize the text request, and obtain standardized user input information.

[0106] In some optional implementations of this embodiment, the intent classification module 701 is further configured to: input standardized input information into the intent recognition model and output the confidence score of each candidate intent category; in response to determining that the highest confidence score is greater than or equal to a preset confidence threshold, determine the candidate intent category corresponding to the highest confidence score as the intent category; in response to determining that all confidence scores are less than the preset confidence threshold, determine it as the default intent.

[0107] In some optional implementations of this embodiment, the intent classification module 701 is further configured to: perform a first-level classification on the standardized input information using an intent recognition model to obtain a first-level intent category, the first-level intent category including at least one of the following: device-side processing intent, application-side processing intent, and default intent; in response to determining that the first-level intent category is an application-side processing intent, perform a second-level classification on the standardized input information using the intent recognition model to obtain a second-level intent category, the second-level intent category including at least one of the following: usage report, function activation, intelligent customer service, information Q&A, and casual conversation.

[0108] In some optional implementations of this embodiment, the routing distribution module 702 is further configured to: invoke the device control link or a dedicated intelligent agent to send control commands to the target intelligent device, and receive the device execution results returned by the target intelligent device after executing the control commands.

[0109] In some optional implementations of this embodiment, the routing distribution module 702 is further configured to: invoke the demand satisfaction model to perform a service response and obtain the service response result.

[0110] In some optional implementations of this embodiment, the routing distribution module 702 is further configured to: convert standardized input information into vector representation; retrieve candidate text fragments that meet preset conditions for similarity with the vector representation from a pre-built knowledge base vector index; construct prompt words based on the candidate text fragments and standardized input information; and input the prompt words into the demand fulfillment model to obtain the business response result.

[0111] In some optional implementations of this embodiment, the routing distribution module 702 is further configured to: call the open domain general large model to perform semantic understanding and dialogue content generation, and obtain the open domain response result.

[0112] The collection, storage, use, processing, transmission, provision, and disclosure of any type of information, such as user personal information, in this technical solution comply with relevant laws and regulations and do not violate public order and good morals.

[0113] According to embodiments of this disclosure, this disclosure also provides an electronic device, a readable storage medium, and a computer program product.

[0114] Figure 8A schematic block diagram of an example electronic device 800 that can be used to implement embodiments of the present disclosure is shown. The electronic device is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device may also represent various forms of mobile devices, such as personal digital processors, cellular phones, smartphones, wearable devices, and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely illustrative and are not intended to limit the implementation of the present disclosure described and / or claimed herein.

[0115] like Figure 8 As shown, device 800 includes a computing unit 801, which can perform various appropriate actions and processes based on a computer program stored in read-only memory (ROM) 802 or a computer program loaded from storage unit 808 into random access memory (RAM) 803. RAM 803 may also store various programs and data required for the operation of device 800. The computing unit 801, ROM 802, and RAM 803 are interconnected via bus 804. Input / output (I / O) interface 805 is also connected to bus 804.

[0116] Multiple components in device 800 are connected to I / O interface 805, including: input unit 806, such as keyboard, mouse, etc.; output unit 807, such as various types of monitors, speakers, etc.; storage unit 808, such as disk, optical disk, etc.; and communication unit 809, such as network card, modem, wireless transceiver, etc. Communication unit 809 allows device 800 to exchange information / data with other devices through computer networks such as the Internet and / or various telecommunications networks.

[0117] The computing unit 801 can be a variety of general-purpose and / or special-purpose processing components with processing and computing capabilities. Some examples of the computing unit 801 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various special-purpose artificial intelligence (AI) computing chips, various computing units running machine learning model algorithms, a digital signal processor (DSP), and any suitable processor, controller, microcontroller, etc. The computing unit 801 performs the various methods and processes described above, such as smart home interaction processing methods. For example, in some embodiments, the smart home interaction processing method can be implemented as a computer software program tangibly contained in a machine-readable medium, such as storage unit 808. In some embodiments, part or all of the computer program can be loaded and / or installed on device 800 via ROM 802 and / or communication unit 809. When the computer program is loaded into RAM 803 and executed by the computing unit 801, one or more steps of the smart home interaction processing method described above can be performed. Alternatively, in other embodiments, the computing unit 801 can be configured to perform smart home interaction processing methods by any other suitable means (e.g., by means of firmware).

[0118] Various embodiments of the systems and techniques described above herein can be implemented in digital electronic circuit systems, integrated circuit systems, field-programmable gate arrays (FPGAs), application-specific integrated circuits (ASICs), application-specific standard products (ASSPs), systems-on-a-chip (SoCs), payload-programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various embodiments may include implementations in one or more computer programs that can be executed and / or interpreted on a programmable system including at least one programmable processor, which may be a dedicated or general-purpose programmable processor, capable of receiving data and instructions from a storage system, at least one input device, and at least one output device, and transmitting data and instructions to the storage system, the at least one input device, and the at least one output device.

[0119] The program code used to implement the methods of this disclosure may be written in any combination of one or more programming languages. This program code may be provided to a processor or controller of a general-purpose computer, special-purpose computer, or other programmable data processing apparatus, such that when executed by the processor or controller, the program code causes the functions / operations specified in the flowcharts and / or block diagrams to be implemented. The program code may be executed entirely on a machine, partially on a machine, as a standalone software package partially on a machine and partially on a remote machine, or entirely on a remote machine or server.

[0120] In the context of this disclosure, a machine-readable medium can be a tangible medium that may contain or store a program for use by or in conjunction with an instruction execution system, apparatus, or device. A machine-readable medium can be a machine-readable signal medium or a machine-readable storage medium. A machine-readable medium can be, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination of the foregoing. More specific examples of machine-readable storage media include electrical connections based on one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination of the foregoing.

[0121] To provide interaction with a user, the systems and techniques described herein can be implemented on a computer having: a display device for displaying information to the user (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor); and a keyboard and pointing device (e.g., a mouse or trackball) through which the user provides input to the computer. Other types of devices can also be used to provide interaction with the user; for example, feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form (including sound input, voice input, or tactile input).

[0122] The systems and technologies described herein can be implemented in computing systems that include backend components (e.g., as a data server), or computing systems that include middleware components (e.g., an application server), or computing systems that include frontend components (e.g., a user computer with a graphical user interface or web browser through which a user can interact with implementations of the systems and technologies described herein), or any combination of such backend, middleware, or frontend components. The components of the system can be interconnected via digital data communication of any form or medium (e.g., a communication network). Examples of communication networks include local area networks (LANs), wide area networks (WANs), and the Internet.

[0123] Computer systems can include clients and servers. Clients and servers are generally located far apart and typically interact via communication networks. Client-server relationships are created by computer programs running on the respective computers and having a client-server relationship with each other. Servers can be cloud servers, distributed system servers, or servers incorporating blockchain technology.

[0124] It should be understood that the various forms of processes shown above can be used to rearrange, add, or delete steps. For example, the steps described in this disclosure can be executed in parallel, sequentially, or in different orders, as long as the desired result of the technical solution provided in this disclosure can be achieved, and this is not limited herein.

[0125] The specific embodiments described above do not constitute a limitation on the scope of protection of this disclosure. Those skilled in the art should understand that various modifications, combinations, sub-combinations, and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this disclosure should be included within the scope of protection of this disclosure.

Claims

1. A smart home interaction processing method, comprising: An intent recognition model is used to classify intents based on user-inputted interaction requests for smart home services and determine the intent category. The number of parameters in the intent recognition model is less than a preset parameter threshold. Differential routing distribution is performed based on the intent category to obtain the processing result; The processing result is sent to the user; The step of performing differentiated route distribution processing based on the intent category to obtain the processing result includes: In response to determining that the intent category is a device-side processing intent, device control-related business processing is performed to obtain the device execution result; In response to determining that the intent category is an application-side processing intent, business processing is performed on the application central side to obtain a business response result; In response to determining that the intent category is the default intent, open domain dialogue generation processing is performed to obtain an open domain response result.

2. The method according to claim 1, wherein, The intent recognition model is used to classify intents based on user-inputted smart home service-oriented interaction requests, and to determine intent categories, including: The interaction request is standardized to obtain standardized input information; The intent recognition model is used to perform semantic understanding and intent classification on the standardized input information to determine the intent category.

3. The method according to claim 2, wherein, The standardization process for the interaction request to obtain standardized input information includes: The text request input by the user is standardized to obtain the standardized user input information; or The user's voice input request is converted into a corresponding text request, and the text request is standardized to obtain the standardized user input information.

4. The method according to claim 2, wherein, The step of using an intent recognition model to perform semantic understanding and intent classification on the standardized input information, and determining the intent category, includes: The standardized input information is input into the intent recognition model, and the confidence score of each candidate intent category is output. In response to determining that the highest confidence score is greater than or equal to a preset confidence threshold, the candidate intent category corresponding to the highest confidence score is determined as the intent category; In response to the determination that all confidence scores are less than the preset confidence threshold, the default intent is determined.

5. The method according to claim 2, wherein, The step of using an intent recognition model to perform semantic understanding and intent classification on the standardized input information, and determining the intent category, includes: The standardized input information is classified at the first level using the intent recognition model to obtain a first-level intent category, which includes at least one of the following: device-side processing intent, application-side processing intent, and default intent. In response to determining that the first-level intent category is an application-side processing intent, the standardized input information is classified in a second-level manner using the intent recognition model to obtain a second-level intent category. The second-level intent category includes at least one of the following: usage report, function activation, intelligent customer service, information Q&A, and casual conversation.

6. The method according to claim 1, wherein, The execution of device control-related business processes yields device execution results, including: The device control link or a dedicated intelligent agent is invoked to send control commands to the target intelligent device, and the execution result of the target intelligent device after executing the control commands is received.

7. The method according to claim 1, wherein, The process of performing business processing at the application central side to obtain a business response result includes: The requirement-satisfaction model is invoked to provide a business response, and the business response result is obtained.

8. The method according to claim 7, wherein, The call request satisfies the model to respond to the business request and obtain the business processing result, including: The standardized input information is converted into a vector representation; Retrieve candidate text fragments that meet preset conditions in the pre-constructed knowledge base vector index and whose similarity to the vector representation is satisfied; Construct prompt words based on the candidate text fragments and the standardized input information; The prompt words are input into the demand fulfillment model to obtain the business response result.

9. The method according to claim 1, wherein, The process of generating open-domain dialogues to obtain open-domain response results includes: The open-domain general model is invoked for semantic understanding and dialogue content generation to obtain the open-domain response result.

10. A smart home interaction processing device, comprising: The intent classification module is configured to use an intent recognition model to classify intents based on user-inputted smart home business-oriented interaction requests and determine intent categories. The number of parameters in the intent recognition model is less than a preset parameter threshold. The routing distribution module is configured to perform differentiated routing distribution processing based on the intent category to obtain the processing result; The result sending module is configured to send the processing result to the user; The routing distribution module is further configured to: In response to determining that the intent category is a device-side processing intent, device control-related business processing is performed to obtain the device execution result; In response to determining that the intent category is an application-side processing intent, business processing is performed on the application central side to obtain a business response result; In response to determining that the intent category is the default intent, open domain dialogue generation processing is performed to obtain an open domain response result.

11. An electronic device, comprising: At least one processor; as well as A memory communicatively connected to the at least one processor; wherein, The memory stores instructions that can be executed by the at least one processor to enable the at least one processor to perform the method of any one of claims 1-9.

12. A non-transitory computer-readable storage medium storing computer instructions for causing the computer to perform the method of any one of claims 1-9.

13. A computer program product comprising a computer program that, when executed by a processor, implements the method according to any one of claims 1-9.