Agent-oriented user dialogue task processing method, system and electronic device

Abstract

Embodiments of the present application disclose an agent-oriented user dialogue task processing method and system and an electronic device. The method introduces a solution mechanism, encapsulates slot rules and dialogue strategies matching dialogue intents into a solution package, recalls the solution package matching the dialogue intent in the dialogue request based on retrieval enhancement technology after intent recognition and slot extraction on the dialogue request, generates prompt words for the agent based on a large language model according to the slot rules and dialogue strategies in the solution package, drives the large language model to execute the corresponding task process, can realize closed-loop task processing of understanding-decision-execution based on user dialogue, improves dialogue task processing efficiency and user experience, ensures the same processing flow for similar problems, reduces task processing failure caused by expression ambiguity, and further improves the dialogue task processing capability of the agent.

Description

Technical Field

[0001] This invention relates to the field of artificial intelligence technology, and more specifically to a user dialogue task processing method, system, and electronic device for intelligent agents. Background Technology

[0002] In the current user service system of internet platforms, after initiating a dialogue request, users typically need to manually fill in key information about their needs. This process heavily relies on the user's understanding and accurate operation of the platform rules. However, in practice, situations frequently arise where requests cannot be processed promptly and effectively due to inconsistencies between the request tags and the request descriptions. This not only prolongs the processing cycle of dialogue tasks but also significantly reduces user satisfaction. Furthermore, traditional user service systems often use static help centers, fixed forms, or simple keyword matching question-answering robots to handle dialogue tasks. This is essentially a "passive response mode," which can only provide general answers and cannot proactively identify the user's true needs, nor can it achieve a complete automated closed-loop processing chain of "understanding the need - making a decision - executing the decision." Summary of the Invention

[0003] In view of this, the purpose of this invention is to provide a user dialogue task processing method, system and electronic device for intelligent agents, so as to realize closed-loop task processing based on user dialogue, improve dialogue task processing efficiency and user experience.

[0004] In a first aspect, embodiments of the present invention aim to provide a user dialogue task processing method for intelligent agents, the method comprising: Receive user dialogue requests; The dialogue request is subjected to intent recognition and slot extraction to determine the corresponding semantic parsing result, which includes dialogue intent and slot information. The solution package that matches the dialogue intent is recalled by retrieval enhancement technology. The solution package includes slot rules and dialogue strategies corresponding to the intent category. The slot rules include at least one slot, and the dialogue strategy is used to describe the task processing flow corresponding to the dialogue intent. The prompt words input to the agent based on the large language model are determined according to the slot rules and the dialogue strategy. The prompt words drive the large language model to execute the corresponding task processing flow.

[0005] Secondly, embodiments of the present invention aim to provide a user dialogue task processing system for intelligent agents, the system comprising: The client is configured to send conversation requests to the user; The intelligent assistant server is configured to receive the dialogue request; perform intent recognition and slot extraction on the dialogue request to determine the corresponding semantic parsing result, the semantic parsing result including dialogue intent and slot information; recall a solution package matching the dialogue intent through retrieval enhancement technology, the solution package including slot rules and dialogue strategies corresponding to the intent category, the slot rules including at least one slot, the dialogue strategy describing the task processing flow corresponding to the dialogue intent; determine prompt words input to the agent based on the large language model according to the slot rules and the dialogue strategy; and drive the large language model to execute the corresponding task processing flow according to the prompt words.

[0006] Thirdly, embodiments of the present invention aim to provide a user dialogue task processing device for intelligent agents, the device comprising: The input module is used to receive user dialogue requests; The intent recognition module is used to perform intent recognition and slot extraction on the dialogue request, and determine the corresponding semantic parsing result, the semantic parsing result including dialogue intent and slot information; The solution configuration module is used to recall a solution package that matches the dialogue intent through retrieval enhancement technology. The solution package includes slot rules and dialogue strategies corresponding to the intent category. The slot rules include at least one slot, and the dialogue strategy is used to describe the task processing flow corresponding to the dialogue intent. The task execution module is used to determine the prompt words input to the agent based on the large language model according to the slot rules and the dialogue strategy; and to drive the large language model to execute the corresponding task processing flow according to the prompt words.

[0007] Fourthly, embodiments of the present invention aim to provide an electronic device, including a memory and a processor, wherein the memory is used to store one or more computer program instructions, wherein the one or more computer program instructions are executed by the processor to implement the method as described in any of the preceding claims.

[0008] Fifthly, embodiments of the present invention aim to provide a computer program product, the computer program product including a computer program / instruction, which, when executed by a processor, implements the method described in any of the preceding claims.

[0009] Sixthly, embodiments of the present invention aim to provide a computer-readable storage medium storing a computer program that, when executed by a processor, implements the method described in any of the preceding claims.

[0010] The technical solution of this invention, upon receiving a user's dialogue request, performs intent recognition and slot extraction to determine the corresponding dialogue intent and slot information. It then uses retrieval enhancement technology to recall solution packages matching the dialogue intent, and determines prompt words for the agent based on a large language model based on the slot rules and dialogue strategies within the solution packages. These prompt words drive the large language model to execute the corresponding task processing flow, thereby achieving a closed-loop processing of understanding the request, making a decision, and executing the decision—that is, closed-loop task processing based on user dialogue. This improves the agent's dialogue task processing capabilities, efficiency, and user experience. Simultaneously, by introducing a solution mechanism to unify the slot rules and dialogue strategies related to task processing under the dialogue intent, and by using retrieval enhancement technology to recall solution packages matching the dialogue intent and driving the agent to execute the corresponding task processing flow based on these solution packages, it ensures that similar problems follow the same processing flow, reducing task processing failures caused by ambiguity and further enhancing the agent's dialogue task processing capabilities. Attached Figure Description

[0011] The above and other objects, features and advantages of the present invention will become clearer from the following description of embodiments of the invention with reference to the accompanying drawings, in which: Figure 1 This is a schematic diagram of a task processing system according to an embodiment of the present invention; Figure 2 This is a flowchart of a task processing method according to an embodiment of the present invention; Figure 3 This is a flowchart of the intent clarification method according to an embodiment of the present invention; Figure 4 This is a flowchart of the slot information integrity detection method according to an embodiment of the present invention; Figure 5 This is a schematic diagram of the slot information supplement page according to an embodiment of the present invention; Figure 6 This is a flowchart of the application and claim task processing according to an embodiment of the present invention; Figure 7 This is a schematic diagram of a claim application page according to an embodiment of the present invention; Figure 8 This is a flowchart of the task distribution interaction page configuration data according to an embodiment of the present invention; Figure 9 This is a schematic diagram of the problem description page according to an embodiment of the present invention; Figure 10 This is a schematic diagram of the reason description page of the embodiment of the present invention; Figure 11 This is a schematic diagram of the evidence upload page according to an embodiment of the present invention; Figure 12This is a schematic diagram of a claim application page according to an embodiment of the present invention; Figure 13 This is a flowchart of the query claim task processing according to an embodiment of the present invention; Figure 14 This is a flowchart of the manual task transfer process according to an embodiment of the present invention; Figure 15 This is a schematic diagram of the intelligent assistant server architecture according to an embodiment of the present invention; Figure 16 This is a schematic diagram of the interaction process in a claim scenario according to an embodiment of the present invention; Figure 17 This is a schematic diagram of a task processing device according to an embodiment of the present invention; Figure 18 This is a schematic diagram of an electronic device according to an embodiment of the present invention. Detailed Implementation

[0012] The present application is described below based on embodiments, but it is not limited to these embodiments. In the detailed description of the present application below, certain specific details are described in detail. Those skilled in the art can fully understand the present application without these details. To avoid obscuring the substance of the present application, well-known methods, processes, flows, elements, and circuits are not described in detail.

[0013] Furthermore, those skilled in the art should understand that the accompanying drawings provided herein are for illustrative purposes only and are not necessarily drawn to scale.

[0014] Unless the context explicitly requires it, words such as "including" or "contains" throughout the application should be interpreted as including rather than exclusive or exhaustive; that is, meaning "including but not limited to".

[0015] In the description of this application, it should be understood that the terms "first," "second," etc., are used for descriptive purposes only and should not be construed as indicating or implying relative importance. Furthermore, in the description of this application, unless otherwise stated, "a plurality of" means two or more.

[0016] The solutions described in this specification and embodiments, if involving the processing of personal information, will be processed only on the premise of having a legal basis (such as obtaining the consent of the personal information subject, or being necessary for the performance of a contract), and will only be processed within the scope stipulated or agreed upon. A user's refusal to process personal information beyond what is necessary for basic functions will not affect the user's use of basic functions.

[0017] In this embodiment, the process of handling the dialogue related to the claim initiated by the merchant in the merchant service system of the food delivery platform is used as an example. However, it should be understood that the method in this embodiment is also applicable to other application scenarios, such as the formulation of merchant marketing and event operation strategies, consumer logistics complaints, etc. In different application scenarios, appropriate solution packages can be used to provide users with a better service experience. There are no restrictions on the application scenarios of the solution or the specific content of the solution package.

[0018] In the current merchant service system of food delivery platforms, when merchants initiate after-sales claims (such as deductions due to rider delivery errors, food spillage, or system misjudgments), they typically need to manually fill in key information such as order number, claim amount, claim type (tag), problem description, and supporting images. This process heavily relies on merchants' understanding and accurate operation of platform rules. However, in practice, claim review failures due to inconsistencies between the problem description and the claim tag are frequent and account for a high percentage of all claim failures. This not only prolongs the processing cycle of dialogue tasks but also significantly reduces user satisfaction, causing some merchants to repeatedly call for consultation or submit repeated claims, directly increasing the platform's customer costs and the Customer Pain Point Index (CPO, used to measure the intensity of negative experiences for merchants in claim processes). Furthermore, traditional food delivery merchant service systems mostly use static help centers, fixed forms, or simple keyword matching Q&A robots, which are essentially "passive response" models. They can only provide general answers and cannot proactively identify the merchant's true needs, dynamically follow up on specific information, or have the ability to invoke business systems to perform operations. Even with the introduction of basic natural language understanding modules, these systems are often limited to single-turn intent judgment, lacking the ability to coherently track multi-turn dialogue contexts and make strategic decisions. This makes it difficult to support end-to-end closed-loop processing of complex tasks such as "food compensation" and "delivery anomaly appeals." Therefore, this invention proposes a user dialogue task processing system and method for intelligent agents, capable of achieving closed-loop task processing based on user dialogue, thereby improving dialogue task processing efficiency and user experience.

[0019] Figure 1 This is a schematic diagram of a task processing system according to an embodiment of the present invention. Figure 1As shown, the task processing system in this embodiment includes a client 10 and a smart assistant server 20, which interact via a communication connection. The client 10 is the dialogue task initiator, configured to send dialogue requests from users. The smart assistant server 20 is the dialogue task processor, configured to execute task processing methods upon receiving the dialogue request from the client 10, thereby completing the dialogue task processing. In the merchant claim scenario, the client 10 is the merchant's terminal device, such as a smartphone or computer. The smart assistant server 20 includes one or more agents, each deployed on a server or related terminal device (such as the merchant's client or the platform reviewer's terminal device) according to its function in the dialogue task processing. Therefore, when merchants have claims-related processing requests such as application or inquiry, they can initiate a dialogue request to the smart assistant server through the client. Based on the interaction process between the client and the smart assistant server, as well as between the client and the user (i.e., the merchant), the corresponding dialogue task processing is realized, thereby helping users complete the claim processing matters, realizing closed-loop task processing based on user dialogue, improving the efficiency of dialogue task processing and user experience.

[0020] Figure 2 This is a flowchart of a task processing method according to an embodiment of the present invention. This task processing method is applied to a smart assistant server and performs corresponding dialogue task processing based on the user's conversation. Figure 2 As shown, the task processing method in this embodiment includes the following processing steps.

[0021] In step S210, a user's dialogue request is received.

[0022] In this embodiment, the user sends a dialogue request to the intelligent assistant server through the client and triggers a dialogue task. The dialogue request includes explicit information input by the user through text, voice, and / or other input methods. Explicit information includes the user's dialogue intent (e.g., requesting a claim), key entities (e.g., order number, time), factual descriptions (e.g., "damaged food," "how to request a claim," "system misjudgment led to deduction, requesting a claim"), desired outcome (e.g., "partial refund"), and / or other relevant information. For example, the dialogue request might be: "The rider damaged food while delivering order A, requesting a claim of x yuan."

[0023] In step S220, the dialogue request is subjected to intent recognition and slot extraction to determine the corresponding semantic parsing result. The semantic parsing result includes the dialogue intent and slot information. The slot information includes the slot and the entity corresponding to the slot.

[0024] In this embodiment, after receiving a dialogue request from the client, the intelligent assistant server performs intent recognition and slot extraction on the dialogue request. Intent recognition determines the user's intent in the dialogue request, i.e., the dialogue intent of the dialogue task, representing the purpose and need behind the user's dialogue request, such as applying for a claim or querying the claim result. Slot extraction extracts the data slots and corresponding entities from the dialogue request. A slot refers to a key information unit in the dialogue request, such as order identifier, time, claim reason, etc., which is the information necessary to fulfill the user's needs. An entity is the specific numerical value or text field corresponding to the slot, i.e., the fill value of the slot, such as order identifier A and claim reason "spoiled food". For example, for the dialogue request "The rider damaged food while delivering order A and applies for a claim of x yuan", after intent recognition and slot extraction, the dialogue intent of this dialogue request can be determined as "apply for a claim". The slot information of the dialogue request includes the following slots and corresponding entities: order identifier - order A, claim reason - damaged food, scenario - rider delivery time, claim request - refund x yuan.

[0025] Optionally, after receiving a dialogue request from the client, the intelligent assistant server in this embodiment will automatically obtain the implicit context of the dialogue request and determine the dialogue intent and slot information corresponding to the dialogue request by performing intent recognition and slot extraction on the dialogue request and its implicit context. The implicit context is supplementary information for the dialogue request, including front-end collected information, dialogue link metadata, back-end associated data, and / or other dialogue-related data. Front-end collected information includes merchant identity (e.g., merchant ID), dialogue time, device environment (e.g., client device identifier, network status), operation trajectory (e.g., "just finished browsing the order details page → clicked the claim icon"), and / or other front-end collected information. Dialogue link metadata includes the current interaction page / scene context (e.g., data obtained through front-end tracking data—the merchant clicked the claim control in the order details page to initiate a dialogue), historical dialogue memory (e.g., data retrieved from the database / cachment—the merchant's most recent N rounds of dialogue records, the current multi-round dialogue state machine state, etc.), and / or other dialogue link related information. Backend associated data includes merchant profiles (such as merchant onboarding duration and credit score), business status (such as order pickup / receipt), historical claim records (such as whether a claim has been filed in the past month and whether the claim was approved), and / or other backend related information. Therefore, this embodiment, by simultaneously utilizing real-time information in the dialogue request and information related to the dialogue request in the implicit context, can automatically complete missing information in the dialogue request and eliminate ambiguity, compressing cumbersome multi-turn question-and-answer sessions into a streamlined conversation. This not only significantly improves the interaction efficiency and accuracy of dialogue task processing but also provides personalized services based on merchant type, enabling the intelligent assistant server to transform dialogue tasks from "passive response" to "active perception," effectively improving the intelligent assistant server's task processing capabilities and user experience.

[0026] Optionally, the intelligent assistant server in this embodiment includes an intent recognition agent, which is used to perform intent recognition and slot extraction on the dialogue request. Further, after acquiring the dialogue request and its corresponding implicit context, the intent recognition agent in this embodiment converts the implicit context into text of the target format and concatenates it into a prompt for intent recognition and slot extraction. This prompt then drives the large language model in the intent recognition agent to perform intent recognition and slot extraction on the dialogue request, thereby outputting the dialogue intent and slot information corresponding to the dialogue request.

[0027] It should be noted that the method for determining dialogue intent and slot information by combining a large language model with dialogue requests and implicit context given in this embodiment is only an example. Other available intent recognition and slot extraction methods can also be used in this embodiment. For example, the dialogue request and implicit context can be converted into corresponding word vectors first, and then the two word vectors can be concatenated through the neural network input layer or attention layer and the influence of context on semantics can be learned to output the dialogue intent and slot information in the dialogue request.

[0028] In step S230, a solution package matching the dialogue intent is recalled by using retrieval enhancement technology. The solution package includes slot rules and dialogue strategies corresponding to the intent category. The slot rules include at least one slot, and the dialogue strategy is used to describe the task processing flow corresponding to the dialogue intent.

[0029] In this embodiment, the solution package is a skill package (agent skills) obtained by uniformly configuring the knowledge required for dialogue task processing under at least one type of problem (scenario). The solution package includes slot rules and dialogue strategies corresponding to intent categories. Slot rules include at least one slot, used to define the parameters (such as order number, intent category), constraints (such as parameter type, enumeration value), and / or extraction priority (whether to extract from the user's input dialogue request first or from the implicit context). The dialogue strategy is used to describe the task processing flow corresponding to the dialogue intent, including defining the state machine of the current task, defining the goals of different states (such as confirmation information, execution action), and defining the processing logic (such as asking for missing slots when the slot is incomplete, triggering intent clarification logic when there are multiple dialogue intents, displaying interactive cards, calling the tool to submit a claim or end the task after the intent is clear and the slot is complete, etc.) and / or other relevant information. It should also be understood that the solution packages corresponding to different intent categories are usually different, and the slot rules and dialogue strategies in the solution packages corresponding to different intent categories may be partially or completely different, which can be configured according to the business processing requirements of the application scenario.

[0030] For example, when the intent category is to apply for a claim, the corresponding slot rules include order identifier, order type (e.g., instant order / pre-order, platform delivery order / self-delivery order), reason for claim (e.g., food spillage / system misjudgment), and channel type (representing the entry category for merchant claims, such as online platform appeal / offline negotiation). The processing logic in the corresponding dialogue strategy includes intent clarification, slot integrity diagnosis, claim submission, and / or other related processing logic. When the intent category is to query a claim, the corresponding slot rules include order identifier, and the processing logic in the corresponding dialogue strategy includes querying the claim processing results for the order.

[0031] Optionally, the solution package in this embodiment, in addition to the slot rules and dialogue strategies corresponding to the intent category, may also include prompt words and / or knowledge-based question-and-answer scripts corresponding to the intent category. Prompt words are used to drive the corresponding large language model to perform the appropriate processing operation (such as submitting a claim or transferring to a human agent). Knowledge-based question-and-answer scripts include at least one set of pre-configured scripts corresponding to the processing operation (such as intent clarification scripts and slot information follow-up scripts) to be pushed to the client.

[0032] Furthermore, after configuration, the solution package is stored in the knowledge base. Upon triggering a dialogue task, Retrieval-Augmented Generation (RAG) technology is used to recall solution packages that match the intent and dialogue intent in real time, driving multi-turn dialogues and tool execution. Therefore, this embodiment, by configuring solution packages, enables unified configuration of task processing strategies for similar problems, supporting version management, canary releases, effect evaluation, and rapid iteration. It possesses advantages of operability and sustainable iterative upgrades, avoiding the inconsistencies and high modification costs caused by business strategies scattered across code or FAQ (Frequently Asked Questions) databases. When executing dialogue task processing, only the solution package corresponding to the intent category needs to be called to determine the business processing strategy for the dialogue task, ensuring the executability of the business strategy and ensuring that similar problems follow the same process (SOP) and interpretation, reducing arbitrary model behavior, increasing stability, and thus improving the accuracy and efficiency of dialogue task processing.

[0033] Optionally, the solution package in this embodiment can be automatically generated by intelligent operation (AI operation) assisted intelligent agent, and can also be combined with manual review to drive the intelligent assistant server to automatically iterate and upgrade, so that the solution package can be continuously iterated and upgraded, further improving the task processing capability of the intelligent assistant server and bringing a better user experience.

[0034] Furthermore, in this embodiment, each dialogue intent corresponds to at least one solution package. Simultaneously, when multiple dialogue intents are identified in the user's input dialogue request, in one optional implementation, the intelligent assistant server in this embodiment simultaneously recalls the solution packages corresponding to multiple dialogue intents, and subsequently executes multiple dialogue intent clarification processing logics to confirm the dialogue intent to be processed through interaction with the client user; finally, based on the user-confirmed dialogue intent and the corresponding solution package, the corresponding task processing flow continues to be executed. In another optional implementation, the intelligent assistant service in this embodiment first executes multiple dialogue intent clarification processing logics to determine the dialogue intent to be processed; then recalls the solution package corresponding to the dialogue intent to be processed; and then executes the corresponding task processing flow based on the solution package corresponding to the dialogue intent, thereby completing the dialogue task processing. Therefore, by providing different methods to handle dialogue tasks under multiple dialogue intents, this embodiment can further improve the flexibility and accuracy of dialogue task processing, meeting the user's dialogue processing needs.

[0035] In step S240, prompt words are determined for input to the agent based on the large language model, according to slot rules and dialogue strategy.

[0036] In this embodiment, after obtaining the solution package corresponding to the dialogue intent in the dialogue request, the prompt words to be input to the agent based on the large language model are determined according to the slot rules and dialogue strategy in the solution package. The prompt words include the instructions input when interacting with the large language model, the question context information and / or other information used to drive the large language model in the agent to perform operations.

[0037] Optionally, when determining the prompt words input to the intelligent agent based on the large language model, this embodiment first determines the keywords of the dialogue task according to the slot rules and dialogue strategy corresponding to the user's dialogue intent, and then generates the corresponding prompt words according to the keywords of the dialogue task, thereby constructing the driving instructions of the intelligent agent with high controllability and high stability, and improving the task processing capability of the intelligent assistant terminal.

[0038] Furthermore, when generating prompts based on keywords, this embodiment can first load a base prompt template based on the keywords. The base prompt template is a predefined text framework in the intelligent assistant server, containing a fixed structure and dynamic placeholders; it serves as a bridge connecting business logic and the capabilities of the large language model. The final prompt is generated after specific data (such as slot status and dialogue context) is filled in at runtime. The base prompt template includes structural information such as role definition (e.g., "You are a professional food delivery merchant handling claims, with a friendly and professional tone, skilled at handling claims"), task background (e.g., merchant profile, claim rules), and dynamic strategy instructions (e.g., asking for confirmation of slot, confirming intent, and submitting a claim). Then, the determined keywords, slot information from the user's dialogue request, and the corresponding implicit context are injected into the base prompt template to generate prompts for the dialogue task. Therefore, in this embodiment, the method described above can decouple business decision-making logic from natural language generation. While generating the prompts required for dialogue task processing, it can avoid the illusion and uncertainty of large language models in process decision-making, ensuring the controllability and accuracy of the dialogue task processing process, thereby improving the dialogue task processing capability of the intelligent assistant terminal.

[0039] Optionally, when the dialogue task includes multiple rounds of conversation, since the task content to be processed is usually different for each round of conversation in the dialogue task, the aforementioned method can be used in this embodiment to generate corresponding prompt words for each round of conversation, so as to continuously iterate and optimize the prompt words used by the intelligent assistant server during the dialogue task, continuously improve the accuracy and reliability of the prompt words, thereby continuously optimizing the task processing capability of the intelligent assistant server and bringing a better user experience.

[0040] In step S250, the corresponding task processing flow is executed based on the prompt words that drive the large language model.

[0041] In this embodiment, by inputting prompt words into the large language model, the large language model can be driven to execute the corresponding task processing flow, thereby completing the task dialogue processing.

[0042] Optionally, in this embodiment, the intelligent assistant server, when performing dialogue processing tasks, also takes snapshots of the conversation information in the dialogue task processing flow. The conversation information includes conversation state, conversation content, and / or conversation results. The conversation state includes task progress (e.g., clarifying intent, asking follow-up questions, transferring to human assistant, task completed), slot information, intent category, dialogue round identifier, control identifiers (e.g., exception markers, retry count), user emotional state (e.g., angry, calm), and / or other state information, serving as a dynamic context environment that is maintained and updated in real time to ensure task continuity. The conversation content includes records of interactions between the client and the intelligent assistant server (e.g., user input information, intelligent assistant server responses; information formats can be text / images / interactive cards, etc.), used as input material for the large language model to facilitate semantic understanding, response generation, and subsequent human quality checks and model optimization. The conversation result refers to the business output or execution conclusion generated after the dialogue task ends, including the business execution status (e.g., successful claim, transfer to human agent), outputs (e.g., proof of successful claim – claim identifier), statistical indicators (e.g., task time, number of interaction rounds), and / or other relevant information. Therefore, by taking snapshots of the conversation information in the dialogue task processing flow in this embodiment, the intelligent assistant server can accurately track the dialogue context information, record user intent, changes in slot values, and a complete dialogue history, ensuring that the intelligent assistant server has a deeper level of understanding and response capabilities. Furthermore, after each round of dialogue, it can correctly update the dialogue status and maintain information consistency, providing an information foundation for the accurate processing of subsequent dialogue tasks.

[0043] Optionally, in this embodiment, when the intelligent assistant server executes a dialogue processing task, it can trigger the execution of the business processing logic corresponding to the business logic interface by calling the business logic interface, and receive the processing result obtained after the business processing logic is executed. Further, in this embodiment, before the dialogue task processing flow is executed, the business processing logic involved in the dialogue task processing (such as order query, claim submission, voucher verification, etc.) is encapsulated into a standardized tool, and the corresponding business logic interface of the standardized tool is provided to the intelligent assistant server for calling. This allows the intelligent assistant server to complete cross-domain operations simply by calling the business logic interface, triggering the corresponding standardized tool to execute the corresponding business processing logic, and quickly obtaining the processing result, achieving the engineering goal of stable main process and horizontally scalable processing capabilities.

[0044] Optionally, the business logic interface can adopt the MCP (Model Context Protocol) interface. The intelligent assistant server triggers the execution of the corresponding business processing logic by calling the business logic MCP interface (also known as the MCP client).

[0045] Optionally, when the large language model is driven to execute the corresponding task processing flow based on the prompt words, this embodiment will first execute the intent clarification processing logic so that the task processing is completed based on the clear dialogue intent and the processing logic corresponding to the dialogue intent category, thereby improving the accuracy and reliability of dialogue task processing.

[0046] Figure 3 This is a flowchart of an intent clarification method according to an embodiment of the present invention. Figure 3 As shown, the intent clarification method in this embodiment includes the following steps.

[0047] In step S310, the dialogue intent obtained after intent recognition of the dialogue request is parsed, and the intent parsing result is determined. The intent parsing result includes the number of dialogue intents in the dialogue request, the confidence level of each dialogue intent, and / or the acknowledgment status.

[0048] In step S320, it is determined whether the intent of the dialogue needs clarification.

[0049] In this embodiment, when a dialogue request includes multiple dialogue intentions, the confidence level of the dialogue intentions is lower than the confidence level threshold, or the dialogue intentions are in a pending confirmation state, it is determined that the dialogue intentions of the dialogue request need to be clarified, and step S330 is continued; while when the confidence level of all dialogue intentions in the dialogue request reaches the confidence level threshold or has been confirmed, it is determined that the dialogue intentions are clear and no further clarification is needed, and step S340 is executed to end the clarification of intentions.

[0050] In step S330, in response to the need for clarification of the dialogue intent, clarification content is sent.

[0051] In this embodiment, when the dialogue intent of a dialogue request needs clarification, the intelligent assistant server will send clarification content to the client. The clarification content can be displayed to the user through cards, buttons, or other display formats, so that the user can confirm the identified dialogue intents based on the displayed clarification content. After receiving the user's confirmation of the dialogue intent, this embodiment will return to step S310 until the dialogue intent is confirmed to be clear, at which point the clarification of intent will end.

[0052] Therefore, in this embodiment, by parsing the dialogue request and obtaining the dialogue intent after intent recognition, determining whether the dialogue intent needs clarification based on the intent parsing result, and issuing clarification content when the dialogue intent needs clarification to guide the client user to confirm or correct the dialogue intent in the dialogue request, the dialogue task processing can be made to have a clear dialogue intent, thereby improving the accuracy and reliability of the dialogue task processing.

[0053] Optionally, this embodiment also performs integrity detection and dynamic tracking on the slot information in the dialogue request, so as to execute the corresponding task processing flow based on the complete slot information, improve the robustness and generalization ability of the intelligent assistant service terminal, and ensure the accuracy and efficiency of dialogue task processing.

[0054] Figure 4 This is a flowchart of the slot information integrity detection method according to an embodiment of the present invention. Figure 4 As shown, in this embodiment, the integrity of slot information in a dialog request is determined by the following method.

[0055] In step S410, the slot information that matches the slot rule is obtained from the slot information based on the prompt words corresponding to the dialogue intent.

[0056] In this embodiment, after determining the dialogue intent in the dialogue request, the system determines the slot information in the corresponding dialogue request slot information that matches the slot rules based on the prompt words corresponding to the dialogue intent and the slot rules. Optionally, after receiving the prompt words corresponding to the dialogue intent, the large language model matches the slots in the dialogue request slot information with the slots in the dialogue intent slot rules to determine the slots covered by the dialogue intent slot information (i.e., the slots involved in both the dialogue intent slot information and the slot rules) and the entities covering the slots.

[0057] For example, the slot information of dialogue request 1 includes slot a, slot b, and slot c, and the entities corresponding to slot a, slot b, and slot c are A, B, and C, respectively; the slot rules corresponding to the dialogue intent of dialogue request 1 include slot a and slot b. After matching the slot information of dialogue request 1 with the slot rules, it can be determined that the slot information in the slot information of dialogue request 1 that matches the slot rules includes slot a-entity A and slot b-entity B. That is, the slot information of dialogue request 1 covers all the slots in the slot rules.

[0058] For example, the slot information of dialogue request 2 includes slot a and slot b, and the entities corresponding to slot a and slot b are A and B, respectively; the slot rules corresponding to the dialogue intent of dialogue request 2 include slot a and slot c. After matching the slot information of dialogue request 2 with the slot rules, it can be determined that the slot information in the slot information of dialogue request 2 that matches the slot rules includes slot a - entity A. That is, the slot information of dialogue request 2 covers some slots in the slot rules.

[0059] In step S420, it is determined whether the slot information in the dialogue request is complete.

[0060] In this embodiment, if the slot information in the dialogue request covers all slots in the slot rule, the slot information in the dialogue request is determined to be complete, and step S430 is executed to end the slot information integrity check. However, if the slot information in the dialogue request does not cover all slots in the slot rule or the slot information does not match the slots in the slot rule (e.g., type / format mismatch, logical conflict), the slot information in the dialogue request is determined to be incomplete, and step S440 is executed.

[0061] In step S440, a follow-up question is sent to guide the user to supplement and / or confirm the slot information that matches the slot rules.

[0062] In this embodiment, when the slot information in the dialogue request is incomplete, the smart assistant server will send follow-up questions to the client to guide the user to supplement and / or confirm the slot information that matches the slot rules.

[0063] Optionally, in this embodiment, the follow-up questioning script in the solution package can be used as follow-up questioning information and sent to the client; alternatively, the slot information supplementation page configuration data can be used as follow-up questioning information. By sending the slot information supplementation page configuration data, the client can be driven to display the corresponding slot information filling page, thereby facilitating the user to supplement or confirm the missing slot information.

[0064] Figure 5 This is a schematic diagram of the slot information supplement page according to an embodiment of the present invention. When configuring data for the slot information supplement page in follow-up questions, such as... Figure 5 As shown, after receiving follow-up information, the client will display a slot information supplement page to the user. The slot information supplement page includes a slot fill control 51 and a slot submit control 52. The slot fill control is used to input slot information. The slot submit control 52 is used to submit the slot information entered in the slot fill control.

[0065] Optionally, the slot information supplement page in this embodiment may only display the slot fill control 51 corresponding to the missing slot in the slot rule, or it may display the slot fill control 51 corresponding to all slots in the slot rule. Further, for the slot fill control corresponding to the missing slot, this embodiment may display a blank slot fill control, or it may display the entity available for the corresponding slot within the slot fill control or in other locations. The entity corresponding to the missing slot may be determined based on the implicit context of the dialogue request. For example, when the slot rule corresponding to the dialogue request includes the order number slot, and the order number slot information is missing in the slot information of the dialogue request, the slot information supplement page will display the slot fill control corresponding to the order number slot, and the slot fill control will display "Please enter the order number" or "Please confirm whether the order number is xxxx, the order number xxxx is determined based on the historical dialogue content".

[0066] Therefore, in this embodiment, by performing integrity checks on the slot information in the dialogue request, and by presenting an interactive page to the user to guide the user to supplement or confirm the slot information when the slot information in the dialogue request is incomplete, the slot information in the dialogue request is fully matched with the corresponding slot rules, so that the dialogue task processing is executed based on the complete dialogue request slot information, ensuring the accuracy and efficiency of the dialogue task processing.

[0067] Furthermore, when the intelligent assistant server executes dialogue processing tasks, this embodiment also provides various task interaction pages for other processing steps that require interaction between the intelligent assistant server and the client. By sending the page configuration data related to the task interaction page to the client to drive the client to render the corresponding task interaction page, the client user is guided to complete the corresponding processing, thereby completing the dialogue task processing.

[0068] To facilitate understanding, this embodiment will provide a detailed description of the task processing procedures under different dialogue intentions.

[0069] Figure 6 This is a flowchart of the application and claim task processing according to an embodiment of the present invention. Figure 6 As shown, in this embodiment, the dialogue task of requesting a claim is processed using the following method.

[0070] In step S610, the slot information that matches the slot rule is obtained from the slot information based on the prompt word corresponding to the intention to make a claim.

[0071] In step S620, it is determined whether the slot information in the dialogue request is complete.

[0072] In this embodiment, the completeness of slot information in the dialogue request is determined using the aforementioned slot information integrity detection method. If the slot information in the dialogue request is determined to be incomplete, step S630 is executed; otherwise, step S640 is executed.

[0073] In step S630, in response to incomplete slot information in the dialog request, a follow-up message is sent. The follow-up message is used to guide the user to supplement and / or confirm slot information that matches the slot rules.

[0074] In step S640, in response to the complete slot information in the dialogue request, the task interaction page configuration data is sent out. The task interaction page configuration data is used to trigger the display of the task interaction page to send a claim submission request through the task interaction page; the complete slot information indicates that the dialogue request includes slot information for all slots in the slot rules.

[0075] Optionally, to further improve the processing efficiency of claim request dialogue tasks, this embodiment also confirms the review status of the order corresponding to the dialogue request. The review status indicates whether the order can be claimed, and includes "applyable" and "unapplyable". When the review status is "applyable", the claim request is submitted; when the review status is "unapplyable", the claim request is terminated and the reason for the inability to apply is pushed to the client. Meanwhile, in this embodiment, the processing of confirming the order review status and the slot information integrity detection are performed independently, and the execution order is not restricted; it can also be performed after the slot information in the dialogue request is detected to be complete. Therefore, the above method allows users to easily understand the entire claim request processing process, improving the processing efficiency of claim request dialogue tasks while ensuring a good user experience.

[0076] Furthermore, in this embodiment, the order review status will be confirmed after the slot information in the dialogue request is detected to be complete: in response to the complete slot information in the dialogue request, the review status of the order corresponding to the dialogue request will be obtained; in response to the review status being unapplicable, the reason for the inability to apply will be pushed; or in response to the review status being applicable, the task interaction page configuration data will be issued.

[0077] Optionally, this embodiment provides different claim submission methods (such as quick claim / normal claim), each using a different task interaction page to submit the claim. Further, this embodiment can determine the corresponding claim submission method based on at least one of the attribute information of the order corresponding to the dialogue request, including user type, order type, claim initiation time, channel type, etc., and use the task interaction page corresponding to the claim submission method to submit the claim. For example, when the order type is an immediate order, a normal claim is used, i.e., the task interaction page corresponding to a normal claim; when the order type is a reservation order, a quick claim is used.

[0078] In one optional implementation, claim submission uses a quick claim mechanism, and the corresponding task interaction page includes a claim application page. When the task interaction page configuration data is distributed, the claim application page configuration data is also distributed, driving the rendering of the claim application page based on this configuration data. The claim application page includes compensation information and a claim submission control, which is used to trigger the submission of the claim.

[0079] Figure 7 This is a schematic diagram of a claim application page according to an embodiment of the present invention. Figure 7As shown, the claim application page displays compensation information 71 and a claim submission control 72. The claim submission control 72 is used to trigger the submission of a claim. Compensation information 71 displays the compensation information that the user can obtain after submitting a claim. Optionally, in this embodiment, compensation information 71 also displays compensation information usage rules to prompt the user on how to use the compensation information. The claim application page also displays claim method recommendation information 73 and correction prompt information 74. The claim method recommendation information informs the user of the reasons for recommending quick claims, and the correction prompt information 74 informs the user of prompts related to order claim processing, such as modifying the claim reason and compensation information, the review time after submission, and the method of compensation payment. Thus, in this embodiment, by driving the client to display the claim application page, the claim compensation items and related information can be displayed to the user, which is beneficial to improving the efficiency of claim dialogue task processing and enhancing user experience.

[0080] In another optional implementation, the claim submission uses a standard claim format, and the corresponding task interaction pages include a problem description page, a cause description page, and a supporting evidence upload page. Simultaneously, the configuration data for the task interaction pages includes, for example: Figure 8 The method shown includes specific processing steps.

[0081] In step S810, configuration data for the problem description page is distributed. This configuration data drives the rendering of the problem description page, which includes a problem description area and a problem submission control. The problem description area is used to input the claim question, and the problem submission control is used to trigger the upload of the claim question.

[0082] Figure 9 This is a schematic diagram of the problem description page according to an embodiment of the present invention. For example... Figure 9 As shown, the problem description page in this embodiment includes a problem description area 91 and a problem submission control 92. After the client displays the problem description page, the user can enter a claim question in the problem description area 91 and submit the claim question by triggering the problem submission control 92 after entering the question. Optionally, the problem description page in this embodiment also displays problem input prompt information 93 to guide the user to enter a claim question, thereby applying for a claim based on the user's input claim question.

[0083] In step S820, in response to submitting a claim, configuration data for the cause description page is sent. This configuration data drives the rendering of the cause description page, which includes a cause description area and a cause confirmation control. The cause description area is used to input the reason for the claim, and the cause confirmation control is used to trigger the upload of the claim reason.

[0084] Figure 10 This is a schematic diagram of the reason description page for an embodiment of the present invention. For example... Figure 10As shown, the cause description page in this embodiment includes a cause description area 101 and a cause confirmation control 102. After the client displays the cause description page, the user can enter the cause of claim in the cause description area 101, and submit the cause of claim by triggering the cause confirmation control 102 after entering the cause. Optionally, the cause description page in this embodiment also includes cause input prompt information 103, which is used to guide the user to enter the cause of claim so as to apply for a claim based on the cause of claim entered by the user.

[0085] Optionally, in this embodiment, the reason description area 101 is filled with different types of claim reason labels, allowing users to select the claim reason that matches the actual claim issue they encounter. Simultaneously, after a user submits a claim issue, this embodiment automatically determines the matching degree between the user-input claim issue and each claim reason based on the claim issue description information, and highlights claim reasons with a matching degree greater than a predetermined matching degree threshold, such as displaying the specific matching degree value in the claim reason area, so that users can quickly determine the claim reason. Furthermore, in this embodiment, if the matching degree between each claim reason and the user-input claim issue is less than the predetermined matching degree threshold, a reminder message can be sent to the user, reminding them to modify the claim issue description entered in the previous step, so as to quickly determine the claim reason and execute the subsequent processing flow.

[0086] In step S830, in response to the submission of the claim reason, configuration data for the supporting evidence upload page is issued. This configuration data drives the rendering of the supporting evidence upload page, which includes an upload area and a confirmation control. The upload area is used to input the supporting evidence for the claim, and the confirmation control is used to upload the supporting evidence.

[0087] Figure 11 This is a schematic diagram of the evidence upload page according to an embodiment of the present invention. For example... Figure 11 As shown, the supporting evidence upload page in this embodiment includes a supporting evidence upload area 111 and a supporting evidence confirmation control 112. After the client displays the supporting evidence upload page, the user can upload supporting evidence for the claim in the supporting evidence upload area 111. The supporting evidence for the claim can be in the form of images, videos, text, etc. After the supporting evidence for the claim is entered, the user completes the submission of the supporting evidence for the claim by triggering the supporting evidence confirmation control 112. Optionally, the reason description page in this embodiment also includes supporting evidence input prompt information 113. The supporting evidence input prompt information 113 is used to guide the user to enter the reason for the claim, so as to review the claim order based on the supporting evidence entered by the user, thereby completing the claim application.

[0088] In step S840, in response to the submission of supporting documentation for a claim, configuration data for the claim application page is sent. This configuration data drives the rendering of the claim application page, which includes compensation information and a claim submission control. The claim submission control is used to trigger the submission of a claim.

[0089] Figure 12 This is a schematic diagram of a claim application page according to an embodiment of the present invention. Figure 12 As shown, the claim application page in this embodiment includes compensation item information 121 and a claim submission control 122. After the client displays the claim application page, the user can view the compensation item information 121 to understand the compensation information they can obtain after submitting a claim, and submit the claim application by triggering the claim submission control 122. Optionally, the compensation item information 121 in this embodiment also displays compensation information usage rules to prompt the user on how to use the compensation information. The claim application page also displays correction prompt information 123, which is used to inform the user of prompts related to order claim processing, such as modifying the claim reason and compensation item information, the review time after submission, and the compensation item payment method.

[0090] Therefore, in this embodiment, by driving the client to display the above-mentioned problem description page, cause description page, and supporting evidence upload page, the user can be guided to submit the corresponding claim information based on the displayed interactive pages. After submitting all claim information, the claim application page can be displayed to show the user the claim payment items and related information, thereby improving the efficiency of claim dialogue task processing and enhancing user experience.

[0091] In step S650, in response to receiving a claim submission request, the claim submission tool interface is invoked to trigger the execution of the claim submission processing logic corresponding to the claim submission tool interface, and the claim submission result of the corresponding order is received.

[0092] In this embodiment, after the client user submits a claim application through the above task interaction page, a claim submission request will be sent to the intelligent assistant server. Upon receiving the claim submission request, the intelligent assistant server will automatically call the claim submission tool interface to trigger the execution of the claim submission processing logic corresponding to the claim submission tool interface, and receive the claim submission result for the corresponding order.

[0093] Therefore, in this embodiment, when the dialogue intent of the dialogue request is determined to be a claim intention, the slot information matching the slot rules in the dialogue request slot information is obtained according to the prompt words corresponding to the claim intention; when the slot information in the dialogue request is complete, the task interaction page configuration data is sent to drive the client to display the corresponding task interaction page and claim submission control; after receiving the claim submission request after the user triggers the claim submission control, the claim submission tool interface is called to trigger the execution of the claim submission processing logic corresponding to the claim submission tool interface; the claim submission result of the corresponding order is received and sent to the client, thus realizing the processing of the user's claim dialogue task.

[0094] Figure 13 This is a flowchart of the query claim task processing according to an embodiment of the present invention. Figure 13 As shown, in this embodiment, the dialog task of querying claim intent is processed through the following method.

[0095] In step S1310, the slot information that matches the slot rule is obtained from the slot information based on the prompt words corresponding to the query claim intent.

[0096] In this embodiment, the method for obtaining slot information matching the slot rules from the slot information based on the prompt words corresponding to the claim intent is similar to the aforementioned method for obtaining slot information matching the slot rules from the slot information based on the prompt words corresponding to the claim intent, and will not be elaborated here. However, it should be understood that since the task processing flow corresponding to different dialogue intents will be different, the prompt words and slot rules corresponding to the claim intent and the claim intent are different in this embodiment.

[0097] In step S1320, in response to the complete slot information in the dialogue request, the order query tool interface is called to trigger the execution of the order query processing logic corresponding to the order query tool interface, and the claim processing result of the corresponding order is received (such as under review / claimed / transferred to human agent).

[0098] In step S1330, the claim processing result is issued.

[0099] In this embodiment, after receiving the claim processing result returned by the order query tool interface, the smart assistant terminal will send the claim processing result to the client so that the user can know the claim processing result of the query order.

[0100] Therefore, in this embodiment, when the dialogue intent of the dialogue request is determined to be a claim query intent, the slot information that matches the slot rule is obtained from the slot information according to the prompt words corresponding to the claim query intent. When the slot information in the dialogue request is complete, the order query tool interface is called to trigger the execution of the order query processing logic corresponding to the order query tool interface. The claim processing result of the corresponding order is received and sent to the client, thus realizing the processing of the user's claim query dialogue task.

[0101] Figure 14 This is a flowchart illustrating the process of transferring tasks to manual intervention according to an embodiment of the present invention. For example... Figure 14 As shown, in this embodiment, the dialogue task of converting human intent is processed using the following method.

[0102] In step S1410, the session status of the corresponding order is obtained based on the prompt words corresponding to the query claim intent.

[0103] In this embodiment, when the user's dialogue request is identified as having the intention to transfer to a human agent, the session status of the order corresponding to the dialogue request and the user's emotional state in the session status are obtained based on the prompt words corresponding to the intention to transfer to a human agent.

[0104] In step S1420, it is determined whether the user's emotional state is a preset state.

[0105] In this embodiment, when the user's emotional state is not in a preset state, it indicates that the user has a lower degree of urgency regarding the task's outcome, and step S1430 is executed. However, when the user's emotional state is determined to be in a preset state, such as anger or tension, it indicates that the user has a higher degree of urgency regarding the task's outcome, and step S1440 is executed.

[0106] In step S1430, in response to the user's emotional state being non-preset, a preset message is pushed.

[0107] In this embodiment, the preset message can be determined by querying the knowledge-based Q&A messages in the solution package corresponding to the intention to transfer to a human operator. The message content can be messages such as "There is a large volume of orders being processed at this time, please wait patiently..." or "Processing in sequence, please wait!"

[0108] In step S1440, in response to the user's emotional state being a preset state, the manual processing logic corresponding to the manual processing logic corresponding to the manual processing logic corresponding to the manual processing interface is triggered by calling the manual processing interface, and the task processing result of the corresponding order is received.

[0109] In this embodiment, when the user's emotional state is in a preset state, the intelligent assistant server will call the manual tool interface to trigger the execution of the manual processing logic corresponding to the manual tool interface, so as to process the current dialogue request and the corresponding order through the manual processing method, and receive the task processing result of the corresponding order after the processing is completed, and record the task processing result in the session information in the corresponding task processing flow, so as to provide a data basis for subsequent query calls.

[0110] Therefore, in this embodiment, when the dialogue intent of the dialogue request is determined to be the intent to switch to a human agent, the session state of the corresponding order is obtained. After obtaining the user's emotional state as a preset state, the human agent tool interface is called to trigger the execution of the human agent processing logic corresponding to the human agent tool interface. The claim processing result of the corresponding order is received and sent to the client, thus realizing the processing of the user's human agent dialogue task.

[0111] It should be understood that this embodiment only provides an example of the task processing flow under some dialogue intents. For other types of dialogue intents, the task processing method provided in this application can still be implemented. For example, when the intent of the dialogue request is to terminate the dialogue, this embodiment can send a closing message to the client. Therefore, this embodiment does not limit the application scenarios of the task processing method.

[0112] Figure 15 This is a schematic diagram of the intelligent assistant server architecture according to an embodiment of the present invention. Figure 15 As shown, the intelligent assistant server in this embodiment includes an input module 151, an intent recognition module 152, a state snapshot module 153, a solution service module 154, a dialogue strategy module 155, a dialogue execution module 156, a business logic service module 157, and an output module 158. The state snapshot module 153 is configured to take snapshots of conversation information during the dialogue processing flow when the intelligent assistant server executes dialogue processing tasks. The solution service module 154 contains solution packages corresponding to different dialogue intents. The business logic service module 157 encapsulates standardized tools corresponding to various business processing logics (such as order inquiry, claim submission, and voucher verification) and provides the corresponding business logic interfaces for the intelligent assistant server to call.

[0113] Figure 16 This is a schematic diagram of the interaction process in a claim scenario according to an embodiment of the present invention. Combined with... Figure 15 He Ru Figure 16As shown, when a user sends a dialogue request to the intelligent assistant server, the input module receives the user's dialogue request, triggers a dialogue task, and processes the user's input dialogue request according to a protocol, converting the dialogue request into a preset data format. The intent recognition module performs intent recognition and slot extraction on the protocol-processed dialogue request to obtain the dialogue intent and slot information corresponding to the dialogue request. The state snapshot module records the dialogue content, dialogue intent, and slot information in the dialogue request, and recalls the solution package matching the dialogue intent from the solution service module based on the dialogue intent using retrieval enhancement technology, and takes a snapshot of the solution package. Then, the dialogue strategy module determines the prompt words to be input to the intelligent agent based on the large language model according to the slot rules and dialogue strategy in the solution package corresponding to the current dialogue intent. The dialogue execution module drives the large language model in the intelligent agent to execute the corresponding task processing flow according to the prompt words. At the same time, the dialogue execution module can trigger the business logic service module to execute the business processing logic corresponding to the business logic interface and return the business processing result by calling the business logic interface. After the dialogue execution module obtains the business processing results (such as text / images / task interaction page configuration data) from the task processing flow, it transmits the business processing results to the output module. The output module performs security processing (such as output content risk control processing) and protocol processing before outputting the results to the client, until the dialogue task processing flow ends. It should be noted that the execution process of each step in the task processing method in this embodiment has been described in detail above and will not be repeated here.

[0114] Therefore, by providing the aforementioned intelligent assistant server framework and applying it to the food delivery merchant claim process, this embodiment achieves a deep integration of large language models, retrieval enhancement technology, business logic interface calls, and page interaction. Through a structured "solution package," it drives multi-round dialogues and action execution, truly realizing a paradigm shift from "passive response" to "proactive action," and constructing an executable dialogue intelligent agent system for food delivery claim scenarios. Meanwhile, from a core technology perspective, this embodiment, in terms of dialogue state management, leverages the contextual awareness capabilities of a large language model combined with a structured dialogue task state model to dynamically track multi-turn intent evolution and slot changes, significantly improving robustness and generalization ability. Regarding knowledge injection and strategy generation, this embodiment innovatively proposes the concept of a "solution package," using retrieval enhancement technology to recall complete strategy packages matching the current intent in real time, ensuring that similar problems follow the same SOP, combining stability and iterability. In terms of tool execution integration, this embodiment encapsulates business processing logic into standardized tools and provides corresponding business logic interfaces for the intelligent assistant server to call. This allows the intelligent assistant server to complete cross-domain operations simply by calling the business logic interface, triggering the corresponding standardized tools to execute the corresponding business processing logic, and quickly obtaining processing results, achieving the engineering goals of stable main processes and horizontally scalable processing capabilities. In terms of interaction, this embodiment designs multimodal page interactions, supporting dynamic rendering of interactive pages such as "selecting the reason for the claim," "uploading supporting photos," and "claim preview confirmation," and links with the backend state machine to ensure strict synchronization between user operations and task progress, significantly improving information collection efficiency and accuracy.

[0115] Figure 17 This is a schematic diagram of a task processing device according to an embodiment of the present invention. Figure 17 As shown, the user dialogue-based task processing device in this embodiment includes an input module 171, an intent recognition module 172, a solution configuration module 173, and a task execution module 174. The input module 171 receives user dialogue requests. The intent recognition module 172 performs intent recognition and slot extraction on the dialogue request, determining the corresponding semantic parsing result. The semantic parsing result includes the dialogue intent and slot information, where the slot information includes the slot and the entity corresponding to the slot. The solution configuration module 173 retrieves a solution package matching the dialogue intent using retrieval enhancement technology. The solution package includes slot rules and dialogue strategies corresponding to the intent category. The slot rules include at least one slot, and the dialogue strategy describes the task processing flow corresponding to the dialogue intent. The task execution module 174 determines prompt words input to the large language model-based agent based on the slot rules and dialogue strategy; and drives the large language model to execute the corresponding task processing flow based on the prompt words.

[0116] Figure 18 This is a schematic diagram of an electronic device according to an embodiment of the present invention. In this embodiment, the electronic device 18 includes a server, a terminal, etc. Figure 18 As shown, the electronic device includes at least one processor 181; and a memory 182 communicatively connected to at least one processor 181; and a communication component 183 communicatively connected to a scanning device, wherein the communication component 183 receives and transmits data under the control of the processor 181; wherein the memory 182 stores instructions executable by at least one processor 181, the instructions being executed by at least one processor 181 to implement the above-described task processing method.

[0117] Specifically, the electronic device includes: one or more processors 181 and a memory 182. Figure 18 Taking a processor 181 as an example, processor 181 and memory 182 can be connected via a bus or other means. Figure 18 Taking a bus connection as an example, memory 182, as a non-volatile computer-readable storage medium, can be used to store non-volatile software programs, non-volatile computer-executable programs, and modules. Processor 181 executes various functional applications and data processing of the device by running the non-volatile software programs, instructions, and modules stored in memory 182, thereby realizing the above-mentioned task processing method.

[0118] Memory 182 may include a program storage area and a data storage area, wherein the program storage area may store the operating system and applications required for at least one function; the data storage area may store an option list, etc. Furthermore, memory 182 may include high-speed random access memory and may also include non-volatile memory, such as at least one disk storage device, flash memory device, or other non-volatile solid-state storage device. In some embodiments, memory 182 may optionally include memory remotely located relative to processor 181, and these remote memories may be connected to external devices via a network. Examples of such networks include, but are not limited to, the Internet, corporate intranets, local area networks, mobile communication networks, and combinations thereof.

[0119] One or more modules are stored in memory 182 and, when executed by one or more processors 181, perform the task processing method in any of the above method embodiments.

[0120] The above-mentioned products can perform the methods provided in the embodiments of this application, and have the corresponding functional modules and beneficial effects of performing the methods. For technical details not described in detail in this embodiment, please refer to the methods provided in the embodiments of this application.

[0121] The technical solution of this embodiment, after receiving a user dialogue request, performs intent recognition and slot extraction on the dialogue request to determine the corresponding dialogue intent and slot information; through retrieval enhancement technology, it recalls a solution package that matches the dialogue intent, and determines the prompt words input to the intelligent agent based on the large language model according to the slot rules and dialogue strategies in the solution package; and drives the large language model to execute the corresponding task processing flow according to the prompt words, thereby realizing a closed-loop link processing of understanding the request, making a decision, and executing the decision, that is, closed-loop task processing based on user dialogue, improving the intelligent agent's dialogue task processing capability, dialogue task processing efficiency, and user experience; at the same time, by introducing a solution mechanism to unify and configure the slot rules and dialogue strategies related to task processing under the dialogue intent, and by recalling a solution package that matches the dialogue intent through retrieval enhancement technology and driving the intelligent agent to execute the corresponding task processing flow based on the solution package, it can ensure that similar problems follow the same processing flow, reduce task processing failures caused by ambiguity, and further improve the intelligent agent's dialogue task processing capability.

[0122] Another embodiment of the present invention relates to a non-volatile storage medium for storing a computer-readable program for use by a computer to execute some or all of the above-described method embodiments.

[0123] That is, those skilled in the art will understand that all or part of the steps in the methods of the above embodiments can be implemented by a program instructing related hardware. This program is stored in a storage medium and includes several instructions to cause a device (which may be a microcontroller, chip, etc.) or processor to execute all or part of the steps of the methods described in the embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as a USB flash drive, a portable hard drive, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk.

[0124] The above description is merely a preferred embodiment of this application and is not intended to limit this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the protection scope of this application.

[0125] This invention discloses, A1, a user dialogue task processing method for intelligent agents, wherein the method includes: Receive user dialogue requests; The dialogue request is subjected to intent recognition and slot extraction to determine the corresponding semantic parsing result. The semantic parsing result includes dialogue intent and slot information. The slot information includes the slot and the entity corresponding to the slot. The solution package that matches the dialogue intent is recalled by retrieval enhancement technology. The solution package includes slot rules and dialogue strategies corresponding to the intent category. The slot rules include at least one slot, and the dialogue strategy is used to describe the task processing flow corresponding to the dialogue intent. The prompt words input to the agent based on the large language model are determined according to the slot rules and the dialogue strategy. The prompt words drive the large language model to execute the corresponding task processing flow.

[0126] A2. According to the method described in A1, the method further includes: A snapshot is taken of the session information in the task processing flow, the session information including session status, session content and / or session result.

[0127] A3. According to the method described in A1, wherein the dialogue intent is to apply for a claim; the step of driving the large language model to execute the corresponding task processing flow based on the prompt word includes: Based on the prompt words corresponding to the intention to file a claim, obtain the slot information that matches the slot rule from the slot information; In response to the complete slot information in the dialogue request, task interaction page configuration data is issued. The task interaction page configuration data is used to trigger the display of the task interaction page so as to send a claim submission request through the task interaction page. The complete slot information indicates that the dialogue request includes slot information of all slots in the slot rules. In response to receiving the claim submission request, the claim submission tool interface is invoked to trigger the execution of the claim submission processing logic corresponding to the claim submission tool interface, and the claim submission result of the corresponding order is received.

[0128] A4. According to the method described in A3, the step of driving the large language model to execute the corresponding task processing flow based on the prompt word further includes: In response to incomplete slot information in the dialogue request, a follow-up message is sent to guide the user to supplement and / or confirm slot information that matches the slot rules.

[0129] A5. According to the method described in A3, wherein the slot information in the response to the dialogue request is complete, and the task interaction page configuration data sent includes: If the slot information in the dialogue request is complete, the review status of the order corresponding to the dialogue request is obtained, and the review status includes whether it can be applied for or not. In response to the approval status indicating that an application is possible, the task interaction page configuration data is issued.

[0130] A6. According to the method described in A5, the step of driving the large language model to execute the corresponding task processing flow based on the prompt word further includes: In response to the stated review status being "unable to apply," a message explaining why the application could not be submitted will be sent.

[0131] A7. According to the method described in A3, the task interaction page includes a claim application page, and the configuration data for the task issuance interaction page includes: The configuration data for the claim application page is distributed to drive the rendering of the claim application page based on the configuration data. The claim application page includes compensation information and a claim submission control, which is used to trigger the submission of a claim.

[0132] A8. According to the method described in A3, the task interaction page includes a problem description page, a cause description page, and a supporting evidence upload page, and the configuration data for the task distribution interaction page includes: Issue configuration data for the problem description page is distributed. This configuration data drives the rendering of the problem description page. The problem description page includes a problem description area and a problem submission control. The problem description area is used to input the claim problem, and the problem submission control is used to trigger the upload of the claim problem. In response to a submitted claim, configuration data for the reason description page is issued. This configuration data drives the rendering of the reason description page, which includes a reason description area and a reason confirmation control. The reason description area is used to input the reason for the claim, and the reason confirmation control is used to trigger the upload of the reason for the claim. In response to the submission of the claim reason, the system issues the supporting evidence upload page configuration data. The supporting evidence upload page configuration data is used to drive the rendering of the supporting evidence upload page. The supporting evidence upload page includes a supporting evidence upload area and a supporting evidence confirmation control. The supporting evidence upload area is used to input the claim supporting evidence, and the supporting evidence confirmation control is used to upload the claim supporting evidence. In response to the submission of supporting evidence for a claim, configuration data for the claim application page is issued. This configuration data drives the rendering of the claim application page, which includes compensation information and a claim submission control. The claim submission control is used to trigger the submission of a claim.

[0133] A9. According to the method described in A1, wherein the dialogue intent is to query a claim, and the step of driving the large language model to execute the corresponding task processing flow based on the prompt words includes: Based on the prompt words corresponding to the intention to make a claim, obtain the slot information that matches the slot rule from the slot information; If the slot information in the dialogue request is complete, the order query tool interface is invoked to trigger the execution of the order query processing logic corresponding to the order query tool interface, and the claim processing result of the corresponding order is received. The claim processing results will be issued.

[0134] A10. According to the method described in A1, the step of driving the large language model to execute the corresponding task processing flow based on the prompt word includes: In response to the statement that the dialogue intent needs clarification, a clarification message is sent. The statement that the dialogue intent needs clarification includes situations where the dialogue request includes multiple dialogue intents, the confidence level of the dialogue intent is lower than a confidence level threshold, or the dialogue intent is in a state of pending confirmation.

[0135] A11. According to the method described in A2, wherein the dialogue intent is a human intervention intent, the conversation state includes the user's emotional state, and the step of driving the large language model to execute the corresponding task processing flow based on the prompt word includes: In response to the user's emotional state being a preset state, the system calls the manual processing tool interface to trigger the execution of the manual processing logic corresponding to the manual processing tool interface, and receives the task processing result of the corresponding order.

[0136] This invention also discloses B1, a user dialogue task processing system for intelligent agents, wherein the system includes: The client is configured to send conversation requests to the user; The intelligent assistant server is configured to receive the dialogue request; perform intent recognition and slot extraction on the dialogue request to determine the corresponding semantic parsing result, the semantic parsing result including dialogue intent and slot information, the slot information including a slot and the entity corresponding to the slot; recall a solution package matching the dialogue intent through retrieval enhancement technology, the solution package including slot rules and dialogue strategies corresponding to the intent category, the slot rules including at least one slot, the dialogue strategy describing the task processing flow corresponding to the dialogue intent; determine prompt words input to the agent based on the large language model according to the slot rules and the dialogue strategy; and drive the large language model to execute the corresponding task processing flow according to the prompt words.

[0137] This invention also discloses C1, a user dialogue task processing device for intelligent agents, wherein the device includes: The input module is used to receive user dialogue requests; The intent recognition module is used to perform intent recognition and slot extraction on the dialogue request, and determine the corresponding semantic parsing result. The semantic parsing result includes dialogue intent and slot information. The slot information includes the slot and the entity corresponding to the slot. The solution configuration module is used to recall a solution package that matches the dialogue intent through retrieval enhancement technology. The solution package includes slot rules and dialogue strategies corresponding to the intent category. The slot rules include at least one slot, and the dialogue strategy is used to describe the task processing flow corresponding to the dialogue intent. The task execution module is used to determine the prompt words input to the agent based on the large language model according to the slot rules and the dialogue strategy; and to drive the large language model to execute the corresponding task processing flow according to the prompt words.

[0138] The present invention also discloses D1, an electronic device including a memory and a processor, wherein the memory is used to store one or more computer program instructions, wherein the one or more computer program instructions are executed by the processor to implement the method described in any one of A1-A11.

[0139] This invention also discloses E1, a computer program product, wherein the computer program product includes a computer program / instruction, which, when executed by a processor, implements the method described in any one of A1-A11.

[0140] The present invention also discloses F1, a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program, and the computer program, when executed by a processor, implements the method described in any one of A1-A11.

Claims

1. A user dialogue task processing method for intelligent agents, characterized in that, The method includes: Receive user dialogue requests; The dialogue request is subjected to intent recognition and slot extraction to determine the corresponding semantic parsing result. The semantic parsing result includes dialogue intent and slot information. The slot information includes the slot and the entity corresponding to the slot. The solution package that matches the dialogue intent is recalled by retrieval enhancement technology. The solution package includes slot rules and dialogue strategies corresponding to the intent category. The slot rules include at least one slot, and the dialogue strategy is used to describe the task processing flow corresponding to the dialogue intent. The prompt words input to the agent based on the large language model are determined according to the slot rules and the dialogue strategy. The prompt words drive the large language model to execute the corresponding task processing flow.

2. The method according to claim 1, characterized in that, The method further includes: A snapshot is taken of the session information in the task processing flow, the session information including session status, session content and / or session result.

3. The method according to claim 1, characterized in that, The dialogue intent is to request a claim; the process of driving the large language model to execute the corresponding task processing flow based on the prompt words includes: Based on the prompt words corresponding to the intention to file a claim, obtain the slot information that matches the slot rule from the slot information; In response to the complete slot information in the dialogue request, task interaction page configuration data is issued. The task interaction page configuration data is used to trigger the display of the task interaction page so as to send a claim submission request through the task interaction page. The complete slot information indicates that the dialogue request includes slot information of all slots in the slot rules. In response to receiving the claim submission request, the claim submission tool interface is invoked to trigger the execution of the claim submission processing logic corresponding to the claim submission tool interface, and the claim submission result of the corresponding order is received.

4. The method according to claim 3, characterized in that, The process of driving the large language model to perform the corresponding task processing based on the prompt word also includes: In response to incomplete slot information in the dialogue request, a follow-up message is sent to guide the user to supplement and / or confirm slot information that matches the slot rules.

5. The method according to claim 3, characterized in that, The task interaction page includes a claim application page, and the task issuance interaction page configuration data includes: The configuration data for the claim application page is distributed to drive the rendering of the claim application page based on the configuration data. The claim application page includes compensation information and a claim submission control, which is used to trigger the submission of a claim.

6. The method according to claim 3, characterized in that, The task interaction page includes a problem description page, a cause description page, and a supporting evidence upload page. The configuration data for the task distribution interaction page includes: Issue configuration data for the problem description page is distributed. This configuration data drives the rendering of the problem description page. The problem description page includes a problem description area and a problem submission control. The problem description area is used to input the claim problem, and the problem submission control is used to trigger the upload of the claim problem. In response to a submitted claim, configuration data for the reason description page is issued. This configuration data drives the rendering of the reason description page, which includes a reason description area and a reason confirmation control. The reason description area is used to input the reason for the claim, and the reason confirmation control is used to trigger the upload of the reason for the claim. In response to the submission of the claim reason, the system issues the supporting evidence upload page configuration data. The supporting evidence upload page configuration data is used to drive the rendering of the supporting evidence upload page. The supporting evidence upload page includes a supporting evidence upload area and a supporting evidence confirmation control. The supporting evidence upload area is used to input the claim supporting evidence, and the supporting evidence confirmation control is used to upload the claim supporting evidence. In response to the submission of supporting evidence for a claim, configuration data for the claim application page is issued. This configuration data drives the rendering of the claim application page, which includes compensation information and a claim submission control. The claim submission control is used to trigger the submission of a claim.

7. The method according to claim 1, characterized in that, The dialogue intent is to query a claim, and the process of driving the large language model to execute the corresponding task processing flow based on the prompt words includes: Based on the prompt words corresponding to the intention to make a claim, obtain the slot information that matches the slot rule from the slot information; If the slot information in the dialogue request is complete, the order query tool interface is invoked to trigger the execution of the order query processing logic corresponding to the order query tool interface, and the claim processing result of the corresponding order is received. The claim processing results will be issued.

8. The method according to claim 2, characterized in that, The dialogue intent is to switch to human intervention, the conversation state includes the user's emotional state, and the process of driving the large language model to execute the corresponding task based on the prompt words includes: In response to the user's emotional state being a preset state, the system calls the manual processing tool interface to trigger the execution of the manual processing logic corresponding to the manual processing tool interface, and receives the task processing result of the corresponding order.

9. A user dialogue task processing system for intelligent agents, characterized in that, The system includes: The client is configured to send conversation requests to the user; The intelligent assistant server is configured to receive the dialogue request; perform intent recognition and slot extraction on the dialogue request to determine the corresponding semantic parsing result, the semantic parsing result including dialogue intent and slot information, the slot information including a slot and the entity corresponding to the slot; recall a solution package matching the dialogue intent through retrieval enhancement technology, the solution package including slot rules and dialogue strategies corresponding to the intent category, the slot rules including at least one slot, the dialogue strategy describing the task processing flow corresponding to the dialogue intent; determine prompt words input to the agent based on the large language model according to the slot rules and the dialogue strategy; and drive the large language model to execute the corresponding task processing flow according to the prompt words.

10. A user dialogue task processing device for intelligent agents, characterized in that, The device includes: The input module is used to receive user dialogue requests; The intent recognition module is used to perform intent recognition and slot extraction on the dialogue request, and determine the corresponding semantic parsing result. The semantic parsing result includes dialogue intent and slot information. The slot information includes the slot and the entity corresponding to the slot. The solution configuration module is used to recall a solution package that matches the dialogue intent through retrieval enhancement technology. The solution package includes slot rules and dialogue strategies corresponding to the intent category. The slot rules include at least one slot, and the dialogue strategy is used to describe the task processing flow corresponding to the dialogue intent. The task execution module is used to determine the prompt words input to the agent based on the large language model according to the slot rules and the dialogue strategy; and to drive the large language model to execute the corresponding task processing flow according to the prompt words.

Images