Intent recognition and feedback generation method and apparatus, and device and medium

Abstract

The present application relates to the technical field of artificial intelligence and digital medical treatment. Disclosed are an intent recognition and feedback generation method and apparatus, and a device and a medium. For the technical problem of inaccurate intent recognition in a dialog system, the last valid input is extracted and subjected to matching with intent identifiers in an intent definition library, and when the matching fails, different intent recognition models are used on the basis of an interaction stage to execute an information query operation, a stage adjustment operation or a guidance operation, so as to realize multi-stage adaptive semantic understanding and feedback generation. The present application can be used in intelligent interaction scenarios such as medical consultation and financial consultation.

Description

Intent recognition and feedback generation method, device, equipment and medium

[0001] The present application claims priority to the Chinese patent application No. 202510141897.8, filed on February 8, 2025, and entitled "Intent recognition and feedback generation method, device, equipment and medium", the whole content of which is incorporated herein by reference. TECHNICAL FIELD

[0002] The present application relates to the field of artificial intelligence and digital medical technology, in particular to an intent recognition and feedback generation method, device, equipment and storage medium. BACKGROUND

[0003] With the deep integration of medical systems and artificial intelligence technology, intelligent robot-assisted diagnosis has gradually become one of the important research directions in the field of medical health. The current medical diagnosis robot mainly undertakes the task of pre-collection of patient disease information, and helps doctors to reduce the work pressure of routine diagnosis through preliminary dialogue with patients. However, the inventors realize that in the interaction process between patients and diagnosis robots, due to the diversification and uncertainty of the way patients describe their illness, the existing diagnosis robot is prone to inaccurate answers, repeated questions or answers beyond the scope of knowledge, resulting in poor diagnosis experience and affecting the satisfaction and trust of patients.

[0004] Traditional medical diagnosis robots rely on static rule matching methods. When the patient's question exceeds the preset rules, the robot often cannot identify the patient's actual needs, and even appears the phenomenon of "random answer" or circular questioning. In addition, the robot lacks effective dynamic adjustment and escape mechanism, and cannot identify abnormal situations and end the dialogue in time, which makes the patient's diagnosis process become long and incoherent, and cannot effectively promote the subsequent doctor diagnosis link.

[0005] In the medical service scene, after completing the interaction with the robot, the doctor usually needs to take over the diagnosis. However, the existing system lacks in-depth analysis and arrangement of the patient's intent when transferring the patient's interaction record to the doctor, resulting in the doctor still needing to reconfirm the patient's illness information and needs during the actual diagnosis. This lack of auxiliary decision support mechanism increases the workload of doctors and reduces the diagnosis efficiency in a multi-patient scenario, especially in the remote diagnosis of Internet hospitals, this problem is particularly prominent.

[0006] In addition, the current diagnosis system cannot dynamically learn and optimize the patient's intent using historical interaction data. Due to the lack of continuous model updating and rule adjustment capability, the intent recognition accuracy of the diagnosis robot is difficult to improve with the change of patient behavior, which leads to the weak response ability of the robot when facing new problems, and the robot cannot provide accurate medical assistance support.

[0007] Overall, the existing intelligent consultation robots have obvious deficiencies in intent recognition, dynamic adjustment, auxiliary decision-making and continuous optimization. How to recognize the intent of each conversation input by the patient and perform corresponding operations according to the different intents of the patient has become a key technical direction to improve the efficiency of medical consultation and improve the patient experience. This not only improves the interactive experience of the consultation robot, but also effectively supports the diagnosis and decision-making process of the doctor, helping the digital and intelligent transformation of medical services. SUMMARY

[0008] The main purpose of the present application is to provide an intent recognition and feedback generation method, device, equipment and storage medium, aiming to solve the technical problem that the existing medical consultation system lacks dynamic intent judgment ability for different consultation stages in the intent recognition process, and cannot effectively distinguish the different needs of human-machine interaction and human-doctor interaction.

[0009] To achieve the above purpose, the present application provides an intent recognition and feedback generation method, comprising:

[0010] receiving input conversation information, and extracting the last input content from the conversation information, taking the last input content as the to-be-processed data;

[0011] matching the to-be-processed data with the intent identifier in the intent definition library, and judging whether there is an intent identifier matched with the to-be-processed data;

[0012] when there is an intent identifier matched with the to-be-processed data, performing the operation corresponding to the intent identifier, and generating feedback information according to the execution result, and outputting the feedback information to the conversation interaction end;

[0013] when there is no intent identifier matched with the to-be-processed data, determining the current interaction stage, the interaction stage including the human-machine interaction stage and the human-human interaction stage;

[0014] if the current interaction stage is the human-machine interaction stage, inputting the to-be-processed data into the conversation intent recognition model, and performing the corresponding operation including information query operation, interaction stage adjustment operation or guidance operation based on the recognition result of the conversation intent recognition model.

[0015] Further, to achieve the above purpose, the present application provides an intent recognition and feedback generation device, comprising:

[0016] an input data management module, configured to receive input conversation information, and extract the last input content from the conversation information, taking the last input content as the to-be-processed data;

[0017] The intent matching module is used to match the data to be processed with intent identifiers in the intent definition library to determine whether there is an intent identifier that matches the data to be processed.

[0018] The feedback generation module is used to execute the operation corresponding to the intent identifier when there is an intent identifier that matches the data to be processed, generate feedback information based on the execution result, and output the feedback information to the dialogue interaction terminal.

[0019] The interaction phase management module is used to determine the current interaction phase when there is no intent identifier that matches the data to be processed. The interaction phase includes the human-machine interaction phase and the human-human interaction phase.

[0020] The intent recognition module is used to input the data to be processed into the dialogue intent recognition model if the current interaction stage is a human-machine interaction stage, and to perform corresponding operations, including information query operation, interaction stage adjustment operation or guidance operation, based on the recognition result of the dialogue intent recognition model.

[0021] Furthermore, to achieve the above objectives, this application also provides a computer device, the computer device including a memory, a processor, and an intent recognition and feedback generation program stored in the memory and executable on the processor, wherein when the intent recognition and feedback generation program is executed by the processor, it implements the steps of the intent recognition and feedback generation method as described above:

[0022] Receive input dialogue information, extract the last input content from the dialogue information, and use the last input content as data to be processed;

[0023] The data to be processed is matched with the intent identifiers in the intent definition library to determine whether there is an intent identifier that matches the data to be processed.

[0024] When an intent identifier that matches the data to be processed exists, the operation corresponding to the intent identifier is executed, and feedback information is generated based on the execution result and output to the dialogue interaction terminal.

[0025] When no intent identifier matches the data to be processed, the current interaction stage is determined, which includes human-machine interaction and human-human interaction.

[0026] If the current interaction stage is a human-machine interaction stage, the data to be processed is input into the dialogue intent recognition model, and corresponding operations, including information query operation, interaction stage adjustment operation, or guidance operation, are executed based on the recognition result of the dialogue intent recognition model.

[0027] Furthermore, to achieve the above objectives, this application also provides a non-volatile computer-readable storage medium storing an intent recognition and feedback generation program, which, when executed by a processor, implements the steps of the intent recognition and feedback generation method as described above.

[0028] Beneficial Effects: This application relates to the fields of artificial intelligence and digital healthcare technology, and discloses an intent recognition and feedback generation method, including: receiving input dialogue information, extracting the last input content as data to be processed, matching the data to be processed with intent identifiers, and executing corresponding operations. When no rule is matched, the interaction stage is determined; in the human-machine interaction stage, an intent recognition model is input, and operations are executed based on the recognition results. This application achieves rapid determination of user intent by matching input data with a dynamic rule base, and adjusts the intent recognition strategy according to the interaction stage for cases of mismatch; by distinguishing intent recognition models at different stages, incorrect answers and repeated questioning are avoided, improving the user interaction experience; simultaneously, executing corresponding operations based on the recognition results optimizes the coherence and accuracy of the interaction process, reduces the error rate, and improves system response efficiency. Attached Figure Description

[0029] The present application will be further described below with reference to the accompanying drawings and embodiments. In the accompanying drawings:

[0030] Figure 1 is a schematic diagram of an application environment for the intent recognition and feedback generation method in one embodiment of this application;

[0031] Figure 2 is a flowchart illustrating an embodiment of the intent recognition and feedback generation method of this application;

[0032] Figure 3 is a schematic diagram of the functional modules of a preferred embodiment of the intent recognition and feedback generation device of this application;

[0033] Figure 4 is a structural schematic diagram of a computer device according to an embodiment of this application;

[0034] Figure 5 is another structural schematic diagram of a computer device according to one embodiment of this application. Detailed Implementation

[0035] It should be understood that the specific embodiments described herein are for illustrative purposes only and are not intended to limit the scope of this application.

[0036] The intent recognition and feedback generation method provided in this application embodiment can be applied in the application environment shown in Figure 1, where the user terminal communicates with the server via a network. The server can receive input dialogue information from the user terminal, extract the last input content as data to be processed, match the data to be processed with intent identifiers, and execute corresponding operations. When no rule is matched, the interaction stage is determined, and the intent recognition model is input during the human-machine interaction stage. Operations are executed based on the recognition results. This application achieves rapid determination of user intent by matching input data with a dynamic rule base, and adjusts the intent recognition strategy according to the interaction stage for cases of no match. By distinguishing intent recognition models at different stages, incorrect answers and repeated questioning are avoided, improving the user interaction experience. At the same time, executing corresponding operations based on recognition results optimizes the coherence and accuracy of the interaction process, reduces the error rate, and improves system response efficiency. The user terminal can be, but is not limited to, various personal computers, laptops, smartphones, tablets, and portable wearable devices. The server can be implemented using a separate server or a server cluster composed of multiple servers. The following is a detailed description of this application through specific embodiments.

[0037] Please refer to Figure 2, which is a flowchart illustrating an embodiment of the intent recognition and feedback generation method provided in this application. It should be noted that although a logical order is shown in the flowchart, in some cases, the steps shown or described may be performed in a different order than that shown here.

[0038] As shown in Figure 2, the intent recognition and feedback generation method proposed in this application includes the following steps:

[0039] S10, receive the input dialogue information, extract the last input content from the dialogue information, and use the last input content as data to be processed;

[0040] In this embodiment, the system receives dialogue information input by the user through an interactive terminal. The user's dialogue information is typically input into the consultation platform or intelligent customer service platform in text format. After the user inputs, the system automatically captures the entire input information stream and temporarily stores it as a data stream to be processed. The process of receiving input information includes verifying the integrity of the input data, deduplication, and formatting, ensuring that the system can receive input normally in different interactive scenarios.

[0041] The process of receiving input information is typically accomplished through interaction between the client and server interfaces. Specifically, the client provides the user's input dialogue content via text input boxes, and the system receives the user's input data stream in real time via an API interface. During the receiving process, the system performs preliminary validity validation on the input data, eliminating empty inputs or incorrect inputs containing special characters. Simultaneously, to ensure data continuity, the system appends user identification information and a timestamp to each input, ensuring accurate identification of the source and order of each input during subsequent processing.

[0042] Extracting the last input from the dialogue ensures that the system uses only the most recent data in subsequent processing. Since users may input multiple times when interacting with a medical consultation platform or intelligent customer service, the system needs to use timestamps to determine and extract the user's most recent input as the most timely data to be processed.

[0043] After receiving all dialogue information from the user, the system arranges this information in chronological order and extracts the record with the latest timestamp. During the extraction process, the system checks the completeness of each input to ensure that no valid data is missed. The extracted content can be structured text or natural language text, and is temporarily stored in a processing queue. If the input received by the system contains multiple lines or segments of information, only the record with the latest timestamp is extracted as the target data for subsequent processing.

[0044] The system extracts the latest input from the dialogue information as the data to be processed. This data is the core data content for the system to perform intent recognition. This data usually requires further preprocessing, such as removing spaces, special characters, and format conversion, to ensure that the data conforms to the system's standard input format.

[0045] When storing the last piece of input as data to be processed, the system performs data formatting. First, the input text is checked for syntax and spelling to ensure its correctness. Second, the system converts the data into a standardized structured format, such as JSON or XML, to facilitate subsequent processing and matching operations. Finally, the system stores the formatted data in a temporary cache queue, preparing it for invocation by the intent recognition model. Throughout the process, the system also retains the original input content for backtracking and review when necessary.

[0046] By receiving user dialogue information and extracting the last input as the data to be processed, the system achieves accurate acquisition of the latest user input, avoiding erroneous judgments caused by using expired or invalid data when processing user intent. Extracting the last input using a timestamp ensures system real-time performance and data accuracy, improves the user experience, and avoids duplicate questions or incorrect handling.

[0047] S20, Match the data to be processed with the intent identifiers in the intent definition library to determine whether there is an intent identifier that matches the data to be processed;

[0048] In this embodiment, the system compares the data to be processed extracted from the user's dialogue with the intent definition library within the system to quickly determine whether the user's input matches the defined intent identifiers. The intent definition library is a database containing multiple preset intent identifiers, with each rule corresponding to an intent category, used to describe common user needs and problem scenarios. The core of the matching process lies in semantic analysis of the data to be processed, comparing it one by one with the rules in the intent definition library to find the most suitable data match.

[0049] The matching process is typically accomplished through an exact matching module or a rule-based matching algorithm. First, the system breaks down the data to be processed into multiple keywords or phrases and compares them item by item according to preset rules in the intent definition library. The matching process can employ full matching, partial matching, or fuzzy matching strategies. For full matching, the system strictly adheres to the format in the rule library; while for fuzzy matching, the system allows for some deviation between the input content and the rule library, such as synonym replacement or word order adjustment.

[0050] For example, the system's intent definition library may contain the rule "I want to consult [department name]". When a user enters "I want to consult the neurology department", the system can recognize "neurology department" as the department name and match the corresponding intent identifier.

[0051] After completing the matching operation, the system needs to determine whether there is an intent identifier that matches the data to be processed. If a matching rule exists, the system considers that the user's intent category has been identified and can execute the corresponding operation based on the matched intent identifier; if no rule is matched, the system needs to further call the intent recognition model for deeper recognition.

[0052] After the matching operation is completed, the system returns a matching result set. If one or more matches exist in the result set, the system will select the most suitable match as the final result according to the priority of the intent identifier. If no rule is matched, the system will hand over the data to be processed to a deep intent recognition model, which will use more complex algorithms such as large language models to predict the intent of the data. Through this two-step matching strategy, the system can ensure both processing efficiency and improve the accuracy of intent recognition.

[0053] By matching the data to be processed with intent identifiers in the intent definition library, rapid identification of user intents is achieved, reducing the frequency of system calls to the deep model and thus improving system response speed and processing efficiency. Simultaneously, through a dynamic update mechanism for preset intent identifiers, the system can promptly adjust rule content according to changes in actual business scenarios, improving the accuracy of user intent identification and the system's flexibility.

[0054] S30, when an intent identifier that matches the data to be processed exists, the operation corresponding to the intent identifier is executed, and feedback information is generated based on the execution result, and the feedback information is output to the dialogue interaction terminal;

[0055] In this embodiment, after matching the data to be processed with rules in the intent definition library, if a rule is identified as matching the data, the system will trigger the operation instruction associated with that rule. The operation instructions corresponding to the intent identifier include a series of preset functional tasks, such as querying user information, recommending services, jumping to a specified department, or executing a fixed response. The key to this process is selecting the specific operation instruction based on the matched intent identifier, ensuring that the system can respond correctly to the user's input.

[0056] The system first loads the operation instructions associated with the matched intent identifier from the operation instruction library. For example, when a user enters "I want to consult a neurologist," and the system matches the intent identifier as "consult a department," the system will load the "recommend department information" operation instruction and provide the user with relevant information about neurology. The operation instructions can be simple reply messages or calls to external service API interfaces to achieve more complex operation tasks.

[0057] In addition, the system monitors the execution status of the operation tasks to ensure that the instructions are executed successfully. If any abnormal situation occurs (such as interface call failure or network problems), the system will automatically record the abnormal status and generate corresponding prompt information.

[0058] After an operation task is completed, the system will generate feedback information based on the execution result. The content of the feedback information depends on the specific execution of the operation task and typically includes the operation result, status information, and prompts. The feedback information can be user-visible text or an execution log recorded internally by the system for subsequent processing.

[0059] When generating feedback information, the system first extracts the execution status and results of the task and converts them into text content that is easy for users to understand. For example, if the user's intention is to query department information, the system's feedback might be "You have successfully selected neurology, and a doctor will see you as soon as possible." The process of generating feedback information typically involves formatting and data mapping to ensure that the output content conforms to the system's preset text format requirements.

[0060] Outputting the generated feedback to the interactive interface means that the system returns the processing results to the user in a visual format. The interactive interface can be a web page, a mobile application, or other types of user interface. By outputting feedback, the system can achieve real-time interaction with the user, enhancing the user experience.

[0061] The system connects to the client via an interface, outputting feedback information to the user interface in the form of text, images, or buttons. During the output process, the system encodes and transmits the feedback information to ensure it is displayed correctly on the user's device. For example, in an intelligent consultation platform, when a patient consults a specific department, the system displays recommended department information and a list of doctors on the user interface. Simultaneously, the system records the output status of this feedback information to ensure the stability and reliability of the information transmission process.

[0062] By matching intent identifiers and executing corresponding actions, the system automates the processing of user intents, improving response speed and interaction efficiency. Feedback information is generated based on the execution results, ensuring a visual display of the system's operation outcomes and enhancing user experience. Furthermore, by outputting feedback information in real time, the system effectively avoids user dissatisfaction caused by operation delays or missing information, thus enhancing system stability and reliability.

[0063] S40, when there is no intent identifier that matches the data to be processed, determine the current interaction stage, which includes the human-machine interaction stage and the human-human interaction stage;

[0064] In this embodiment, after the system matches the data to be processed with intent identifiers in the intent definition library, if no matching intent identifier is found, the system proceeds to the next judgment step. At this point, the system considers that the user's input content could not be quickly matched according to preset rules, which may belong to a more complex or unexpected scenario, and therefore requires further processing.

[0065] The matching process typically includes two methods: quick matching and fuzzy matching. If the system fails to find a matching intent identifier in either of these stages, it determines that the user's input has no matching rules. In this case, the system does not immediately terminate the conversation but instead enters a dynamic intent recognition process, selecting different intent recognition strategies based on the current interaction stage.

[0066] For example, when a user enters "Can I buy medicine online?" but there is no matching rule in the rule base, the system will enter the judgment process of the interaction stage to determine the next processing method.

[0067] Determining the current interaction stage is a crucial step for the system when no matching rules exist. The system needs to determine whether the current user interaction is in the human-machine interaction stage (user interacting with the consultation robot) or the human-human interaction stage (user interacting with a doctor). The different interaction stages determine the intent recognition model and processing strategies that the system needs to adopt.

[0068] The system determines the current interaction stage by analyzing the dialogue context or the user's identity status. For example, when a user first enters the consultation platform and interacts with the robot, the system automatically determines it to be in the human-machine interaction stage; when the user's dialogue has been taken over by a doctor, it is determined to be in the human-human interaction stage. The determination of the interaction stage can be implemented in the following ways:

[0069] User role recognition: Determine the interaction stage based on the user's identity status (such as patient or doctor).

[0070] Dialogue context analysis: Based on the keywords and language style of the dialogue content, infer whether the current interaction is with a robot or a doctor.

[0071] Status identifiers: The current interaction stage is recorded through system status identifiers, which are automatically updated when the status changes.

[0072] When determining the interaction phase, the system mainly distinguishes between two scenarios: human-machine interaction and human-human interaction.

[0073] The human-machine interaction stage: When users interact with the consultation robot, the system mainly processes user input through preset rules and intent recognition models to avoid the robot giving incorrect answers or asking repeated questions.

[0074] Human-to-human interaction stage: When users interact with doctors, the system provides doctors with auxiliary decision support based on the dialogue records from the human-computer stage and the user's latest input, such as identifying the user's intention to purchase medication or examination needs.

[0075] The intent recognition models used in these two stages differ, as do the operational strategies. The human-machine stage focuses more on preventing the robot from giving incorrect answers, while the human-human stage focuses more on identifying the user's actual needs and assisting doctors in improving diagnostic efficiency.

[0076] By determining the interaction stage in the absence of matching rules, the system achieves dynamic adjustment capabilities in different scenarios. The system can distinguish between human-machine interaction stages and human-to-human interaction stages, preventing the robot from getting stuck in an infinite loop or providing incorrect answers when it cannot answer questions. Simultaneously, by providing doctors with historical user interaction records, the system can assist doctors in more accurately identifying user needs during human-to-human interactions, improving diagnostic efficiency and user satisfaction.

[0077] S50, if the current interaction stage is a human-machine interaction stage, the data to be processed is input into the dialogue intent recognition model, and corresponding operations, including information query operation, interaction stage adjustment operation or guidance operation, are executed based on the recognition result of the dialogue intent recognition model.

[0078] In this embodiment, when the system determines that the user's current interaction stage is a human-machine interaction stage, the system will adopt a specific processing strategy for human-machine interaction. During human-machine interaction, users typically interact with medical consultation robots or intelligent customer service systems. Since medical consultation robots may encounter situations such as incomplete information or questions exceeding their knowledge scope when answering user questions, the system needs to use a specific model to recognize the user's intent in order to avoid the robot providing incorrect or repetitive answers.

[0079] The system analyzes the context of the user's identity and the dialogue content to determine whether the current interaction stage is a human-machine interaction stage. If the user's input has not yet been taken over by a doctor or explicitly transferred to human customer service, the system identifies the current stage as a human-machine interaction stage. The system marks the user's latest input data as pending data, preparing it for subsequent model recognition.

[0080] After determining the current stage of human-machine interaction, the system inputs the user's data to be processed into the dialogue intent recognition model. The dialogue intent recognition model is a model based on Large Language Model (LLM) or other machine learning algorithms, specifically designed to analyze the user's dialogue content and identify the user's actual intent.

[0081] During the model recognition process, the system first preprocesses the data to be processed, such as removing spaces, special characters, and noisy data. Then, the system inputs the formatted text data into the dialogue intent recognition model. Based on the semantic features, keywords, and contextual information of the input data, the model outputs a prediction result for one or more intent categories. To improve recognition accuracy, the model is typically fine-tuned and trained in a specialized field to ensure it can accurately identify common intents in medical or financial domains.

[0082] For example, in a medical consultation scenario, if a user enters "I want to consult a cardiologist", the system will input this text data into the dialogue intent recognition model, and the model will identify the user's intent category as "specific department".

[0083] After receiving the output of the dialogue intent recognition model, the system will execute corresponding operations based on the recognition results. Each recognized intent category corresponds to a specific operation task, such as recommending department information, transferring to a doctor, or providing a fixed response. The system automatically calls the corresponding operation instruction library based on the recognition results and executes the corresponding task in the background.

[0084] The system performs intent mapping on the model's output, assigning the identified intent categories to predefined operational tasks. For example:

[0085] When the intent category is "Specified Department", the system will automatically recommend a list of departments and doctors that the user should consult.

[0086] When the intent category is "other rights and interests", the system provides a fixed response to guide the user to the next step of consultation.

[0087] When the intent category is "sensitive issue", the system will automatically terminate the current interaction and proceed to the next stage of human customer service.

[0088] During the execution of the operation task, the system will monitor the execution status and result of the operation task, and generate feedback information based on the task result and output it to the user terminal.

[0089] By inputting the data to be processed into the dialogue intent recognition model and executing corresponding operations based on the recognition results, the system achieves automated analysis and accurate processing of user input. During the human-machine interaction phase, the application of the intent recognition model effectively avoids issues such as incorrect robot responses and repetitive questioning, improving the user's interactive experience. Simultaneously, dynamic operation command invocation based on the recognition results ensures the timeliness and accuracy of the system's response, reducing user discomfort during robot interaction and improving overall service quality.

[0090] This application relates to the fields of artificial intelligence and digital healthcare technology, and discloses an intent recognition and feedback generation method, including: receiving input dialogue information, extracting the last input content as data to be processed, matching the data to be processed with intent identifiers, and executing corresponding operations. If no rule is matched, the interaction stage is determined; in the human-machine interaction stage, an intent recognition model is input, and operations are executed based on the recognition results. This application achieves rapid determination of user intent by matching input data with a dynamic rule base, and adjusts the intent recognition strategy according to the interaction stage for cases of mismatch; by distinguishing intent recognition models at different stages, incorrect answers and repeated questioning are avoided, improving the user interaction experience; simultaneously, executing corresponding operations based on the recognition results optimizes the coherence and accuracy of the interaction process, reduces the error rate, and improves system response efficiency.

[0091] In one embodiment, after S40 above, the following is also included:

[0092] S401, if the current interaction stage is a human-to-human interaction stage, input the data to be processed into the intent recognition model, and determine the service request category based on the recognition result of the intent recognition model;

[0093] S402, when the service request category is information query, provide information feedback corresponding to the service request category;

[0094] S403, when the service request type is an operation instruction, generate corresponding prompt information, and prompt the corresponding execution entity to perform the operation corresponding to the operation instruction according to the prompt information;

[0095] S404, the identification result and information feedback or the operation information performed are stored as an interaction record.

[0096] In this embodiment, when the system cannot find a rule matching the data to be processed in the intent identifier library, it needs to determine the current interaction stage by analyzing the interaction context. This is to select different processing strategies and ensure that the system can provide appropriate responses based on the actual interaction situation. The system distinguishes between human-machine interaction stages and human-human interaction stages to dynamically switch between human-machine interaction and manual intervention.

[0097] The system can determine the interaction stage through the following dimensions:

[0098] User identity status: Determine the current interaction status based on the user's role (patient or doctor).

[0099] Dialogue history analysis: By analyzing the dialogue content and contextual information, determine whether the current interaction object is a robot or a doctor.

[0100] System identifier: The system updates the interaction state identifier at each interaction phase transition to ensure that each operation is executed in the correct interaction phase.

[0101] During the human-to-human interaction phase, the system needs to analyze the user's specific service needs based on the user's latest input and historical interaction records. The data to be processed is input into the human intent recognition model, which performs semantic analysis on the user's input to identify the user's actual intent category, i.e., the service request category.

[0102] The Renren Intent Recognition Model, fine-tuned for specific vertical domains, can identify various user service requests, including information queries and operation commands. After inputting the data to be processed, the model outputs one or more possible service request categories and selects the category with the highest confidence as the final recognition result.

[0103] For example, if a user enters "I want to know about the side effects of medicines", the model will identify the user's service request category as an information query.

[0104] When the system identifies a service request as an information query, it will automatically generate corresponding feedback. For example, if a user inquires about the side effects of a medication or a doctor's schedule, the system will provide detailed text feedback.

[0105] The system extracts content related to the service request category through the information feedback module and generates feedback information in a standardized text format. Feedback information can be retrieved directly from the database or through external API calls. For example:

[0106] Check your rights: The system provides information on the user's medical rights.

[0107] Prescription Query: The system displays the user's historical prescription records.

[0108] When the identified service request is classified as an operation instruction, the system generates operation prompts to guide the doctor or other executor to complete the corresponding operation. For example, when a user indicates that they need a prescription or examination order, the system generates a prompt to remind the doctor to complete the ordering process.

[0109] The system generates prompts related to the operation instructions through the instruction generation module and transmits this information to the doctor's terminal. For example:

[0110] When a user intends to issue a prescription, the system prompts "Please issue the required medication prescription for the patient".

[0111] When a user intends to issue an examination order, the system prompts "Please arrange the necessary examination items for the patient".

[0112] After completing intent recognition and operation execution, the system stores the recognition results and generated feedback or operation information as interaction logs. These interaction logs can be used for subsequent user behavior analysis, model optimization, and rule updates.

[0113] The process of storing interaction records includes the following steps: recording the intent recognition results, including the identified service request category and confidence value; recording information feedback or operation execution status, including the generated feedback content and execution status; and storing the interaction records in a log database for subsequent analysis and optimization.

[0114] This embodiment achieves accurate identification and efficient processing of user service requests by dynamically determining the interaction stage when no matching rules are available, and executing different operation tasks based on the intent recognition results during the human-to-human interaction stage. In information query scenarios, the system can quickly provide the feedback information needed by the user; in operation command scenarios, the system can automatically generate command prompts to help doctors complete relevant operation tasks. By recording interaction data, the system also has self-optimization capabilities, improving the accuracy of intent recognition and response speed.

[0115] In one embodiment, after S404 above, the following is further included:

[0116] S405, After the human-to-human interaction phase ends, the interaction record is transmitted to the log database;

[0117] S406, Based on the interaction data stored in the log database, adjust the parameter weights of the intent recognition model to optimize the recognition capability of the intent recognition model;

[0118] S407, Update the intent identifier in the intent definition library based on the interaction record and the feedback information received during the human-to-human interaction phase.

[0119] In this embodiment, after the human-to-human interaction phase is completed, the system transmits the recognition results, information feedback, or executed operation information generated during the current interaction as an interaction record to the log database. The main function of the log database is to store the user's interaction data with the system for subsequent analysis and model optimization.

[0120] The system automatically triggers the storage operation of the interaction record when the interaction ends. The interaction record typically includes:

[0121] User-input data to be processed; intent recognition results (such as information queries, operation instructions, etc.); executed operation tasks and their status; generated feedback information.

[0122] The system transmits these interaction records to the log database via API interfaces or batch data transfer. The log database can be a local database or a cloud database, and it has efficient data query and management capabilities.

[0123] The system dynamically adjusts the parameter weights of the intent recognition model by analyzing interaction data in the log database to improve the model's recognition accuracy. The main purpose of adjusting the model parameter weights is to optimize the model's intent classification ability and confidence judgment, ensuring that the system can more accurately identify user intents in different scenarios.

[0124] Interaction data from the log database is used for model training and fine-tuning. The specific implementation process is as follows:

[0125] Data extraction: The system extracts interaction records from the log database, including user input, recognition results, and feedback information.

[0126] Model training: The extracted data is used as training samples and input into the intent recognition model to adjust the model's parameter weights.

[0127] Weight update: Based on the model's performance on new data, dynamically adjust the model's weight parameters to improve the model's ability to recognize different intent categories.

[0128] For example, if the system frequently makes mistakes in recognizing the intent to "issue a prescription", the system will adjust the confidence weight of that intent category based on the interaction data in the log database, thereby improving the accuracy of recognition.

[0129] The system dynamically updates intent identifiers in the intent definition library by analyzing interaction logs and user feedback. Updating intent identifiers includes adding rules, modifying existing rules, and deleting invalid rules, ensuring the system can adapt to changing user needs.

[0130] The system updates the intent identifier based on the user's actual feedback, specifically including the following steps:

[0131] Rule analysis: Analyze unmatched user input in the interaction log to identify new user intents.

[0132] Added rule: For user intents that do not match, generate a new intent identifier and add it to the intent definition library.

[0133] Rule optimization: Adjust existing intent identifiers, such as optimizing keywords and updating matching conditions.

[0134] Rule deletion: Delete intent identifiers that have not been triggered or have expired for a long time, ensuring that the rules in the intent definition library always remain valid.

[0135] For example, if a user enters "How to view my inspection report" multiple times, but the system fails to match the corresponding intent identifier, the system will automatically add a "Inspection Report Query" rule and optimize and adjust that rule.

[0136] This embodiment achieves the system's self-optimization capability by storing interaction records in a log database and dynamically adjusting the parameter weights of the intent recognition model based on the log data. By analyzing interaction records and user feedback, the system can dynamically update the rules of the intent definition library, ensuring that the system's intent recognition capability always adapts to changes in user needs, thereby improving the system's recognition accuracy and response speed, and reducing user dissatisfaction and complaints.

[0137] In one embodiment, the above S50 includes:

[0138] S501, if the current interaction stage is the human-machine interaction stage, input the data to be processed into the dialogue intent recognition model, and determine whether the dialogue intent recognition model recognizes the specific intent.

[0139] S502, when the dialogue intent recognition model identifies a specific intent, it executes the operation corresponding to the specific intent;

[0140] S503, when the dialogue intent recognition model fails to recognize a specific intent, or when the input data to be processed exceeds the processing range, the current interaction stage is switched to the human-to-human interaction stage.

[0141] S504, during the human-to-human interaction phase, based on historical interaction data, identify the actual needs and execute auxiliary operations corresponding to the actual needs.

[0142] In this embodiment, during the human-machine interaction phase, the system processes the user-input data. The system transmits the user-input data to the dialogue intent recognition model and determines whether the model can identify the user's specific intent category. The core function of the intent recognition model is to determine the user's true needs through semantic analysis, thereby providing corresponding operational guidance.

[0143] The data to be processed first undergoes data cleaning and formatting to remove redundant characters, special symbols, spaces, and other noise. The cleaned data is then input into the dialogue intent recognition model. Based on semantic understanding, keyword matching, and contextual analysis, the model predicts the user's intent category. The system receives the model's output and determines whether a specific intent has been identified. If the model outputs a clear intent category, the system considers the user's specific intent successfully identified.

[0144] For example, if a user enters "I want to make an appointment with the neurology department", the model identifies the intent category as "appointment consultation", and the system can then execute the corresponding operation task.

[0145] After the model identifies the user's specific intent, the system will invoke the corresponding operation command based on the identification result. Each intent category is pre-associated with one or more operation tasks, and the system will execute these tasks according to preset logic to meet the user's needs.

[0146] The system extracts operation instructions corresponding to the identified intent categories from the operation instruction library. The system then invokes these instructions to perform tasks such as querying, recommending, navigating, or responding. The system monitors the execution status of these tasks and generates feedback information based on the results.

[0147] For example, when a user's intent is "to inquire about rights and benefits", the system will load the "inquire about user rights and benefits" operation command and return the query results to the user.

[0148] If the dialogue intent recognition model fails to identify a specific intent category, or if the user's input exceeds the model's processing capacity, the system will automatically switch to a human-to-human interaction phase. This switching mechanism ensures that the user does not get stuck in a loop due to questions the robot cannot answer, while also preventing the system from providing incorrect or incomplete answers.

[0149] When the model fails to recognize a specific intent, the system triggers a no-match processing logic. The system automatically switches to a human-to-human interaction stage, transferring the user's input to a doctor or customer service representative. The system transmits the user's historical interaction data and current input to the assigned doctor or representative to ensure smooth subsequent conversations.

[0150] For example, when a user asks, "Can you recommend a good nutritionist?" but the model cannot recognize the intent, the system will transfer the user's conversation record to a human customer service representative for handling the request.

[0151] During the human-to-human interaction phase, the system analyzes the user's historical interaction data and current input to identify the user's specific needs and execute corresponding auxiliary operations. These auxiliary operations include recommending services, providing benefits information, and generating prescriptions or examination orders, helping doctors or customer service personnel to meet the user's needs more efficiently.

[0152] The system analyzes users' historical interaction data to identify behavioral patterns and potential needs. Based on the analysis results, the system automatically generates auxiliary operation prompts and recommends corresponding operation tasks to doctors or customer service personnel.

[0153] This embodiment introduces a dialogue intent recognition model into the human-machine interaction stage, ensuring the system can quickly identify the user's specific intent and execute corresponding tasks. When the model fails to recognize a specific intent, the system automatically switches to the human-to-human interaction stage, preventing users from experiencing negative experiences due to questions the robot cannot answer. In the human-to-human interaction stage, the system analyzes the user's historical interaction data and current input to provide precise auxiliary operation prompts, improving the work efficiency of doctors or customer service personnel and reducing users' waiting time and communication costs.

[0154] In one embodiment, S20 above includes:

[0155] S201, Load intent identifiers from the intent definition library and adjust the priority of intent identifiers based on historical interaction data;

[0156] S202, the data to be processed is input into the precise matching module, and the precise matching module determines whether there is an intent identifier that matches the data to be processed;

[0157] S203, when multiple matching intent identifiers exist, the intent identifier with the highest priority is selected as the final matching result based on priority;

[0158] S204, when there is no matching intent identifier, the data to be processed is input into the fuzzy matching module, and matching is performed according to the fuzzy matching conditions to determine whether there is an intent identifier that is fuzzily matched with the data to be processed.

[0159] In this embodiment, the system loads a preset set of intent identifiers from an intent definition library. These rules typically include keyword matching rules, semantic matching rules, etc. To ensure matching accuracy, the system dynamically adjusts the priority of intent identifiers based on historical interaction data. The rule priority adjustment process is based on user input frequency, success rate, and feedback, thereby enabling the system to prioritize matching the most frequently used and most effective intent identifiers.

[0160] Upon startup, the system automatically loads all intent identifiers from the intent definition library and caches these rules in memory to improve matching efficiency. The system periodically analyzes historical interaction data, including user input, matched intent identifiers, and actual feedback results. Based on this data, the system dynamically adjusts rule priorities. For example, if a rule is frequently triggered and provides positive feedback, its priority is increased; conversely, its priority is decreased.

[0161] The exact match module is one of the core modules of the system used for efficient matching of intent identifiers. This module is mainly for intent identifiers with fixed formats or keyword matching, such as specific department names or drug names. Through the exact match module, the system can quickly determine whether the data to be processed meets the matching conditions in the intent identifier.

[0162] The system inputs the pre-processed data to the precise matching module. The precise matching module uses methods such as keyword matching, regular expression matching, or rule tree matching to quickly scan the data and determine if a matching intent identifier exists. If the precise matching module finds a matching intent identifier, it outputs the matching result; otherwise, the data is passed to the next step, the fuzzy matching module.

[0163] Example: In a financial services scenario, when a customer enters "I want to check my credit card limit", the precise matching module recognizes the keyword "credit card limit", quickly matches the intent marker "check limit", and provides the corresponding operation instructions.

[0164] If the precise matching module identifies multiple matching intent identifiers, the system will select the highest priority intent identifier as the final matching result by prioritizing them. This process ensures that when faced with multiple candidate rules, the system can prioritize the rule that best meets the user's needs, avoiding incorrect matching results.

[0165] The system sorts multiple matching intent identifiers according to priority. The system selects the intent identifier with the highest priority from the sorted results as the final match. The system then executes the corresponding operation task based on the selected intent identifier and generates feedback information.

[0166] If the exact matching module fails to identify any matching intent identifier, the system will pass the data to be processed to the fuzzy matching module. The fuzzy matching module is mainly used to handle cases where the user input is incomplete, contains typos, or is ambiguous, ensuring that the system can still recognize the user's intent when faced with complex or abnormal input.

[0167] The system inputs unmatched data to the fuzzy matching module. The fuzzy matching module uses methods such as semantic analysis, word vector matching, or similarity calculation to determine whether the data meets the fuzzy matching criteria. If the fuzzy matching module identifies an intent marker that fuzzily matches the data, it outputs the matching result; otherwise, the system will indicate that there is no matching result and guide the user to re-enter the data.

[0168] This embodiment achieves dynamic optimization and accurate matching of intent identifiers by loading intent identifiers from an intent definition library and dynamically adjusting the priority of rules based on historical interaction data. By introducing precise matching and fuzzy matching modules, the system can simultaneously meet the intent recognition requirements of both standardized and abnormal inputs, improving the accuracy of intent recognition and the system's response speed. In the case of matching multiple intent identifiers, the optimal rule is selected through a priority ranking mechanism, avoiding incorrect rule matching and improving the user's interactive experience.

[0169] In one embodiment, S10 includes:

[0170] S101, Receive input dialogue information from the dialogue interaction terminal, and add a timestamp identifier to each input content in the dialogue information;

[0171] S102, Remove blank inputs, noisy data, or duplicate information from the dialogue information to obtain valid information;

[0172] S103, extract the latest valid input content with the timestamp from the valid information;

[0173] S104, the latest valid input content identified by the timestamp is formatted to generate text data that meets the preset data format requirements, and the text data is stored as data to be processed.

[0174] In this embodiment, when the system receives dialogue information input by the user, a timestamp is appended to each piece of input for subsequent time sorting and data processing. The timestamp is typically represented in a time format (such as Unix timestamp or ISO format) to ensure that the system can accurately identify the chronological order of each input.

[0175] The system receives text input from users via API interfaces or data streams from interactive endpoints (such as chat windows, online consultation platforms, etc.). Upon receiving input, the system immediately generates a timestamp and appends it to each piece of input for subsequent sorting and extraction. The timestamp uses a time format accurate to milliseconds or seconds to ensure the system can effectively process multiple input data within a short period.

[0176] In a medical consultation scenario, a patient continuously inputs "I want to register" and "Internal medicine doctor" through the consultation platform. The system adds a timestamp to each input, such as:

[0177] "I want to register" - Timestamp: 2025-01-07 10:00:01;

[0178] "Internal Medicine Doctor" - Timestamp: 2025-01-07 10:00:03.

[0179] When processing user-input dialogue information, the system removes blank input, noisy data, and duplicate information to ensure that it only processes valid dialogue content. This step aims to reduce data redundancy and the impact of erroneous input, thereby improving the accuracy of subsequent data processing.

[0180] The system detects and removes blank or invalid characters from user input. It identifies and removes noisy data from user input, such as meaningless character combinations and special symbols, through keyword detection or regular expression matching. The system also detects whether the user has entered the same content multiple times and removes duplicates, ensuring that each piece of information is retained only once.

[0181] In medical consultation scenarios, patients may enter "internal medicine doctor" multiple times, or enter invalid characters "###". The system will remove this noisy data and duplicate content, retaining only the valid information.

[0182] Valid information: "I want to make an appointment"; "Internal medicine doctor".

[0183] The system sorts valid information according to its timestamp and extracts the latest valid input as the most recent user input. This process ensures that the system always processes the user's most recent input, avoiding interference from older data in subsequent intent recognition and operation execution.

[0184] The system sorts valid information chronologically based on the attached timestamps. It extracts the latest valid input entry with the newest timestamp and treats it as the latest data to be processed.

[0185] This embodiment achieves accurate extraction and standardized storage of the latest user input data by attaching a timestamp to each piece of user input and performing noise reduction and formatting on the input content. By extracting the most recent valid input content, the system can effectively avoid interference from old data, ensuring that subsequent intent recognition and operation execution are based on the latest user input, thereby improving the system's response speed and recognition accuracy.

[0186] In one embodiment, S30 includes:

[0187] S301, Based on the matched intent identifier, determine the operation instruction corresponding to the intent identifier, and load the corresponding operation instruction from the operation instruction library;

[0188] S302, invoke the operation instruction to execute the corresponding operation task, and monitor the execution status and result of the operation task;

[0189] S303, Based on the execution status and execution result of the operation task, generate feedback information including operation result, status information and prompt content;

[0190] S304, The feedback information is formatted and converted into text data that meets the preset format requirements;

[0191] S305 outputs the formatted feedback information to the dialogue interaction terminal.

[0192] In this embodiment, after the system successfully matches an intent identifier, it needs to determine the operation instruction to be executed based on the intent identifier. The operation instruction is a set of specific operation steps or tasks pre-configured by the system, such as data query, service recommendation, form generation, etc. The system will load the operation instruction corresponding to the intent identifier from the operation instruction library to ensure the accuracy and efficiency of subsequent task execution.

[0193] Once the system identifies the user's intent category, it automatically searches for the corresponding operation instructions. It then extracts the set of operation instructions matching the intent category from the operation instruction library. This library typically uses key-value pairs or rule mapping tables to ensure fast instruction lookup and loading. The system uses these loaded operation instructions as input for the task, preparing for subsequent operations.

[0194] In a medical consultation scenario, when a user enters "I want to book an appointment with an internal medicine doctor," the system identifies the intent category as "appointment department." The system loads the "appointment with an internal medicine doctor" operation instruction from the operation instruction library and prepares to execute the appointment process.

[0195] The system calls the operation instructions loaded from the operation instruction library and executes the corresponding operation tasks. During task execution, the system monitors the execution status and results of the tasks in real time to ensure smooth task execution and generate valid feedback information.

[0196] The system invokes the corresponding services or modules to execute tasks based on the loaded operation instructions. During task execution, the system monitors the execution status (such as success, failure, timeout) and execution results (such as query results, update results). If an error occurs during task execution, the system triggers an error handling mechanism, records an error log, and provides feedback on the error information.

[0197] In a medical consultation scenario, once the system loads the "book an internal medicine doctor" command, it calls the appointment service interface to execute the appointment task and monitors its execution status. If the appointment is successful, the system generates an "appointment successful" result; if the appointment fails, it returns an error message, such as "no doctor available for appointment."

[0198] The system generates feedback information based on the task's execution status and results. This feedback information typically includes:

[0199] Operation result (e.g., success or failure); status information (e.g., whether the task was completed, whether any exceptions occurred); prompts (e.g., instructions for the next step, error messages, etc.).

[0200] The system generates feedback information based on the task execution result, such as "Reservation successful" or "Reservation failed." The system monitors the task execution status and adds relevant information (such as execution time and whether there were any anomalies) to the feedback. Based on the task execution status, the system generates prompts to guide the user through subsequent operations.

[0201] In a medical consultation scenario, when a user successfully schedules an appointment with an internist, the system generates the feedback message: "Your appointment has been successful. Please schedule your appointment at the designated time." If the appointment fails, the system generates the feedback message: "No doctors are currently available for appointment. Please select another time slot."

[0202] To ensure that feedback information is displayed in a standardized format, the system performs data formatting on the generated feedback information. Data formatting includes operations such as text structure adjustment, encoding conversion, and format verification to ensure that the output information conforms to preset format requirements.

[0203] The system adjusts the text structure of the feedback information to ensure clear logic and ease of reading. It converts the encoding format of the feedback information to UTF-8 or other common formats to ensure compatibility across different systems. Finally, the system verifies the formatted feedback information to ensure it conforms to preset display standards.

[0204] This embodiment automates task execution and feedback generation by loading operation instructions, executing operation tasks, generating feedback information, and outputting the formatted feedback information to the interactive interface. During feedback generation, the system provides real-time and accurate feedback based on the task execution status, thereby improving user experience and system interaction efficiency. Furthermore, through data formatting and multi-platform compatibility, the system ensures that feedback information can be displayed correctly on different platforms, further optimizing the user's interactive experience.

[0205] In one embodiment, an intent recognition and feedback generation device is provided, which corresponds one-to-one with the intent recognition and feedback generation methods in the above embodiments. Referring to Figure 3, Figure 3 is a functional block diagram of a preferred embodiment of the intent recognition and feedback generation device of this application. The device comprises an input data management module 10, an intent matching module 20, a feedback generation module 30, an interaction stage management module 40, and an intent recognition module 50. Detailed descriptions of each functional module are as follows:

[0206] The input data management module 10 is used to receive input dialogue information, extract the last input content from the dialogue information, and use the last input content as data to be processed.

[0207] The intent matching module 20 is used to match the data to be processed with intent identifiers in the intent definition library to determine whether there is an intent identifier that matches the data to be processed.

[0208] The feedback generation module 30 is used to execute the operation corresponding to the intent identifier when there is an intent identifier that matches the data to be processed, generate feedback information based on the execution result, and output the feedback information to the dialogue interaction terminal.

[0209] The interaction phase management module 40 is used to determine the current interaction phase when there is no intent identifier that matches the data to be processed. The interaction phase includes the human-machine interaction phase and the human-human interaction phase.

[0210] The intent recognition module 50 is used to input the data to be processed into the dialogue intent recognition model if the current interaction stage is a human-machine interaction stage, and to perform corresponding operations, including information query operation, interaction stage adjustment operation or guidance operation, based on the recognition result of the dialogue intent recognition model.

[0211] For specific limitations regarding the intent recognition and feedback generation device, please refer to the foregoing limitations regarding the intent recognition and feedback generation method, which will not be repeated here. Each module in the aforementioned intent recognition and feedback generation device can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in a computer device in hardware form, or stored in the memory of a computer device in software form, so that the processor can call and execute the operations corresponding to each module.

[0212] In one embodiment, a computer device is provided, which may be a server, and its internal structure diagram is shown in Figure 4. The computer device includes a processor, memory, a network interface, and a database connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile and / or volatile storage media and internal memory. The non-volatile storage media stores an operating system, computer programs, and a database. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage media. The network interface of the computer device is used to communicate with external user terminals via a network connection. When the computer program is executed by the processor, it implements the functions or steps of an intent recognition and feedback generation method on the server side.

[0213] In one embodiment, a computer device is provided, which can be a user terminal, and its internal structure diagram can be as shown in Figure 5. The computer device includes a processor, memory, network interface, display screen, and input device connected via a system bus. The processor of the computer device provides computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and internal memory. The non-volatile storage medium stores an operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage medium. The network interface of the computer device is used to communicate with an external server via a network connection. When the computer program is executed by the processor, it implements the functions or steps of an intent recognition and feedback generation method on the user terminal side.

[0214] In one embodiment, a computer device is provided, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to perform the following steps:

[0215] Receive input dialogue information, extract the last input content from the dialogue information, and use the last input content as data to be processed;

[0216] The data to be processed is matched with the intent identifiers in the intent definition library to determine whether there is an intent identifier that matches the data to be processed.

[0217] When an intent identifier that matches the data to be processed exists, the operation corresponding to the intent identifier is executed, and feedback information is generated based on the execution result and output to the dialogue interaction terminal.

[0218] When no intent identifier matches the data to be processed, the current interaction stage is determined, which includes human-machine interaction and human-human interaction.

[0219] If the current interaction stage is a human-machine interaction stage, the data to be processed is input into the dialogue intent recognition model, and corresponding operations, including information query operation, interaction stage adjustment operation, or guidance operation, are executed based on the recognition result of the dialogue intent recognition model.

[0220] In one embodiment, a non-volatile computer-readable storage medium is provided, which may be non-volatile or volatile, and stores a computer program thereon. When the computer program is executed by a processor, it performs the following steps:

[0221] Receive input dialogue information, extract the last input content from the dialogue information, and use the last input content as data to be processed;

[0222] The data to be processed is matched with the intent identifiers in the intent definition library to determine whether there is an intent identifier that matches the data to be processed.

[0223] When an intent identifier that matches the data to be processed exists, the operation corresponding to the intent identifier is executed, and feedback information is generated based on the execution result and output to the dialogue interaction terminal.

[0224] When no intent identifier matches the data to be processed, the current interaction stage is determined, which includes human-machine interaction and human-human interaction.

[0225] If the current interaction stage is a human-machine interaction stage, the data to be processed is input into the dialogue intent recognition model, and corresponding operations, including information query operation, interaction stage adjustment operation, or guidance operation, are executed based on the recognition result of the dialogue intent recognition model.

[0226] It should be noted that the functions or steps that can be implemented by the computer-readable storage medium or computer device described above can be referred to the relevant descriptions on the server side and user side in the foregoing method embodiments. To avoid repetition, they will not be described one by one here.

[0227] It should be noted that any software tools or components not belonging to this company appearing in the embodiments of this application are merely illustrative examples and do not represent actual use. All user personal information involved in the embodiments of this application has been authorized (with the knowledge and consent) by the relevant parties or has been fully authorized by all parties, and the executing entity can obtain it through various open, legal, and compliant means. The collection, storage, use, processing, transmission, provision, and disclosure of the information, data, and signals involved all comply with the relevant laws and regulations of the relevant countries and regions, and do not violate public order and good morals.

Claims

1. An intent recognition and feedback generation method, wherein, Includes the following steps: Receive input dialogue information, extract the last input content from the dialogue information, and use the last input content as data to be processed; The data to be processed is matched with the intent identifiers in the intent definition library to determine whether there is an intent identifier that matches the data to be processed. When an intent identifier that matches the data to be processed exists, the operation corresponding to the intent identifier is executed, and feedback information is generated based on the execution result and output to the dialogue interaction terminal. When no intent identifier matches the data to be processed, the current interaction stage is determined, which includes human-machine interaction and human-human interaction. If the current interaction stage is a human-machine interaction stage, the data to be processed is input into the dialogue intent recognition model, and corresponding operations, including information query operation, interaction stage adjustment operation, or guidance operation, are executed based on the recognition result of the dialogue intent recognition model.

2. The intent recognition and feedback generation method as described in claim 1, wherein, When no intent identifier matches the data to be processed, the current interaction stage is determined. This interaction stage includes both human-machine interaction and human-to-human interaction stages, and further includes: If the current interaction stage is a human-to-human interaction stage, the data to be processed is input into the intent recognition model, and the service request category is determined based on the recognition result of the intent recognition model. When the service request category is information query, provide information feedback corresponding to the service request category; When the service request type is an operation instruction, a corresponding prompt message is generated, and the corresponding execution entity is prompted to perform the operation corresponding to the operation instruction based on the prompt message; The identification results and information feedback or the operation information performed are stored as an interaction record.

3. The intent recognition and feedback generation method as described in claim 2, wherein, If the current interaction stage is a human-to-human interaction stage, the data to be processed is input into the intent recognition model, and the service request category is determined based on the recognition result of the intent recognition model, including: If the current interaction stage is a human-to-human interaction stage, relevant contextual information is extracted from historical interaction records based on the semantics of the data to be processed. The context information is concatenated with the data to be processed to generate an enhanced input sequence; The enhanced input sequence is input into the intent recognition model to generate a recognition result; Based on the identification results, the service request category is determined.

4. The intent recognition and feedback generation method as described in claim 2, wherein, After storing the recognition results and information feedback or the executed operation information as an interaction record, the method further includes: After the human-to-human interaction phase ends, the interaction records are transmitted to the log database; Based on the interaction data stored in the log database, the parameter weights of the intent recognition model are adjusted to optimize the recognition capability of the intent recognition model; The intent identifiers in the intent definition library are updated based on the interaction records and feedback information received during the human-to-human interaction phase.

5. The intent recognition and feedback generation method as described in claim 1, wherein, If the current interaction stage is a human-machine interaction stage, the data to be processed is input into the dialogue intent recognition model. Based on the recognition result of the dialogue intent recognition model, corresponding operations are performed, including information query operations, interaction stage adjustment operations, or guidance operations, including: If the current interaction stage is the human-machine interaction stage, the data to be processed is input into the dialogue intent recognition model, and it is determined whether the dialogue intent recognition model recognizes the specific intent. When the dialogue intent recognition model identifies a specific intent, it executes the operation corresponding to the specific intent. When the dialogue intent recognition model fails to recognize a specific intent, or when the input data to be processed exceeds the processing range, the current interaction stage is switched to the human-to-human interaction stage. During the human-to-human interaction phase, the actual needs are identified based on historical interaction data, and auxiliary operations corresponding to those actual needs are executed.

6. The intent recognition and feedback generation method as described in claim 5, wherein, Based on historical interaction data, identify actual needs and execute auxiliary operations corresponding to those needs, including: Feature extraction is performed on users' historical interaction data to identify user behavior patterns and potential needs; The behavioral patterns and potential needs are semantically fused with the current data to be processed to generate a comprehensive contextual feature vector. The comprehensive context feature vector is input into the demand classification model to determine the user's actual demand type. Based on the actual need type, generate auxiliary operation prompts and execute the auxiliary operation corresponding to the auxiliary operation prompts.

7. The intent recognition and feedback generation method as described in claim 1, wherein, The process involves matching the data to be processed with intent identifiers in the intent definition library to determine whether an intent identifier matches the data to be processed, including: Load intent identifiers from the intent definition library and adjust the priority of intent identifiers based on historical interaction data; The data to be processed is input into the precise matching module, which determines whether there is an intent identifier that matches the data to be processed. When multiple matching intent identifiers exist, the intent identifier with the highest priority is selected as the final matching result based on priority. When no matching intent identifier exists, the data to be processed is input into the fuzzy matching module, and matching is performed according to the fuzzy matching conditions to determine whether there is an intent identifier that fuzzily matches the data to be processed.

8. The intent recognition and feedback generation method as described in claim 7, wherein, Load intent identifiers from the intent definition library and adjust their priority based on historical interaction data, including: Load intent identifiers from the intent definition library; Based on the historical interaction data, extract the user input frequency, success matching rate and feedback score corresponding to each intent identifier; Based on the user input frequency, success rate, and feedback score, determine the priority adjustment value for each intent identifier; Based on the priority adjustment value, the priority of all intent identifiers in the intent definition library is dynamically updated.

9. The intent recognition and feedback generation method as described in claim 7, wherein, When no matching intent identifier exists, the data to be processed is input into the fuzzy matching module, and matching is performed according to the fuzzy matching conditions to determine whether there is an intent identifier that fuzzily matches the data to be processed, including: When no matching intent identifier exists, the unmatched data to be processed is processed into semantic feature vectors to generate corresponding semantic feature vectors. The semantic feature vector is compared with the feature vectors corresponding to each intent identifier pre-stored in the intent definition library to obtain at least one similarity value; Determine whether the similarity value exceeds a preset fuzzy matching threshold; If a similarity value exceeds the preset fuzzy matching threshold, a fuzzy match is determined to exist, and the intent identifier with the highest similarity value is identified as the fuzzy matching result. If none of the similarity values ​​exceed the preset fuzzy matching threshold, a no-match message will be output.

10. The intent recognition and feedback generation method as described in claim 1, wherein, Receive input dialogue information, extract the last input content from the dialogue information, and use the last input content as data to be processed, including: Receive input dialogue information from the dialogue interaction terminal and add a timestamp to each input content in the dialogue information; Remove blank inputs, noisy data, or duplicate information from the dialogue information to obtain valid information; Extract the latest valid input from the valid information; The latest valid input content identified by the timestamp is formatted to generate text data that meets the preset data format requirements, and the text data is stored as data to be processed.

11. The intent recognition and feedback generation method as described in claim 1, wherein, When an intent identifier matching the data to be processed is found, the operation corresponding to the intent identifier is executed, and feedback information is generated based on the execution result. The feedback information is then output to the dialogue interaction terminal, including: Based on the matched intent identifier, determine the operation instruction corresponding to the intent identifier, and load the corresponding operation instruction from the operation instruction library; The operation instructions are invoked to execute the corresponding operation tasks, and the execution status and results of the operation tasks are monitored. Based on the execution status and result of the operation task, generate feedback information including operation result, status information and prompt content; The feedback information is formatted and converted into text data that meets the preset format requirements; The formatted feedback information is output to the interactive interface.

12. The intent recognition and feedback generation method as described in claim 11, wherein, Based on the matched intent identifier, the operation instruction corresponding to the intent identifier is determined, and the corresponding operation instruction is loaded from the operation instruction library, including: Based on the matched intent identifier, a key-value mapping strategy is used to search the operation instruction library to obtain one or more candidate operation instructions. Determine the number of candidate operation instructions retrieved; If there is only one candidate operation instruction, then the candidate operation instruction is used as the operation instruction to be loaded; If there are multiple candidate operation instructions, the candidate operation instruction with the highest priority is selected as the operation instruction to be loaded based on the priority label attached to each operation instruction. If no candidate operation instructions are found, the operation instruction update mechanism is triggered to generate an instruction update request; According to the instruction update request, the operation instruction corresponding to the intent identifier is synchronized from the remote instruction library, and the operation instruction is loaded into the local cache as the final loaded operation instruction.

13. An intent recognition and feedback generation apparatus, wherein, The intent recognition and feedback generation device includes: The input data management module is used to receive input dialogue information, extract the last input content from the dialogue information, and use the last input content as data to be processed. The intent matching module is used to match the data to be processed with intent identifiers in the intent definition library to determine whether there is an intent identifier that matches the data to be processed. The feedback generation module is used to execute the operation corresponding to the intent identifier when there is an intent identifier that matches the data to be processed, generate feedback information based on the execution result, and output the feedback information to the dialogue interaction terminal. The interaction phase management module is used to determine the current interaction phase when there is no intent identifier that matches the data to be processed. The interaction phase includes the human-machine interaction phase and the human-human interaction phase. The intent recognition module is used to input the data to be processed into the dialogue intent recognition model if the current interaction stage is a human-machine interaction stage, and to perform corresponding operations, including information query operation, interaction stage adjustment operation or guidance operation, based on the recognition result of the dialogue intent recognition model.

14. A computer device, wherein, The computer device includes a memory, a processor, and an intent recognition and feedback generation program stored in the memory and executable on the processor. When the intent recognition and feedback generation program is executed by the processor, it implements the following steps of the intent recognition and feedback generation method: Receive input dialogue information, extract the last input content from the dialogue information, and use the last input content as data to be processed; The data to be processed is matched with the intent identifiers in the intent definition library to determine whether there is an intent identifier that matches the data to be processed. When an intent identifier that matches the data to be processed exists, the operation corresponding to the intent identifier is executed, and feedback information is generated based on the execution result and output to the dialogue interaction terminal. When no intent identifier matches the data to be processed, the current interaction stage is determined, which includes human-machine interaction and human-human interaction. If the current interaction stage is a human-machine interaction stage, the data to be processed is input into the dialogue intent recognition model, and corresponding operations, including information query operation, interaction stage adjustment operation, or guidance operation, are executed based on the recognition result of the dialogue intent recognition model.

15. The computer device as claimed in claim 14, wherein, When no intent identifier matches the data to be processed, the current interaction stage is determined. This interaction stage includes both human-machine interaction and human-to-human interaction stages, and further includes: If the current interaction stage is a human-to-human interaction stage, the data to be processed is input into the intent recognition model, and the service request category is determined based on the recognition result of the intent recognition model. When the service request category is information query, provide information feedback corresponding to the service request category; When the service request type is an operation instruction, a corresponding prompt message is generated, and the corresponding execution entity is prompted to perform the operation corresponding to the operation instruction based on the prompt message; The identification results and information feedback or the operation information performed are stored as an interaction record.

16. The computer device as claimed in claim 14, wherein, After storing the recognition results and information feedback or the executed operation information as an interaction record, the method further includes: After the human-to-human interaction phase ends, the interaction records are transmitted to the log database; Based on the interaction data stored in the log database, the parameter weights of the intent recognition model are adjusted to optimize the recognition capability of the intent recognition model; The intent identifiers in the intent definition library are updated based on the interaction records and feedback information received during the human-to-human interaction phase.

17. The computer device as claimed in claim 14, wherein, If the current interaction stage is a human-machine interaction stage, the data to be processed is input into the dialogue intent recognition model. Based on the recognition result of the dialogue intent recognition model, corresponding operations are performed, including information query operations, interaction stage adjustment operations, or guidance operations, including: If the current interaction stage is the human-machine interaction stage, the data to be processed is input into the dialogue intent recognition model, and it is determined whether the dialogue intent recognition model recognizes the specific intent. When the dialogue intent recognition model identifies a specific intent, it executes the operation corresponding to the specific intent. When the dialogue intent recognition model fails to recognize a specific intent, or when the input data to be processed exceeds the processing range, the current interaction stage is switched to the human-to-human interaction stage. During the human-to-human interaction phase, the actual needs are identified based on historical interaction data, and auxiliary operations corresponding to those actual needs are executed.

18. The computer device as claimed in claim 14, wherein, The process involves matching the data to be processed with intent identifiers in the intent definition library to determine whether an intent identifier matches the data to be processed, including: Load intent identifiers from the intent definition library and adjust the priority of intent identifiers based on historical interaction data; The data to be processed is input into the precise matching module, which determines whether there is an intent identifier that matches the data to be processed. When multiple matching intent identifiers exist, the intent identifier with the highest priority is selected as the final matching result based on priority. When no matching intent identifier exists, the data to be processed is input into the fuzzy matching module, and matching is performed according to the fuzzy matching conditions to determine whether there is an intent identifier that fuzzily matches the data to be processed.

19. The computer device as claimed in claim 14, wherein, Receive input dialogue information, extract the last input content from the dialogue information, and use the last input content as data to be processed, including: Receive input dialogue information from the dialogue interaction terminal and add a timestamp to each input content in the dialogue information; Remove blank inputs, noisy data, or duplicate information from the dialogue information to obtain valid information; Extract the latest valid input from the valid information; The latest valid input content identified by the timestamp is formatted to generate text data that meets the preset data format requirements, and the text data is stored as data to be processed.

20. A non-volatile computer-readable storage medium, wherein, The storage medium stores an intent recognition and feedback generation program, which, when executed by a processor, implements the steps of the intent recognition and feedback generation method as described in any one of claims 1-12.

Images