Asynchronous quotation method, device, storage medium and electronic equipment

Abstract

The application discloses an asynchronous pricing method and device, a storage medium and an electronic equipment, and relates to the technical field of artificial intelligence. The method comprises the following steps: identifying the demand scheme intention of a user based on the user message and historical dialogue carried by a first dialogue request, and synchronously determining the target processing action for the user in the next step; if the target processing action is a normal dialogue, a dialogue reply is generated; if the dialogue reply contains an adjustment pricing signal, an asynchronous pricing instruction is generated based on the demand scheme intention and sent to a second server, and the current session stage is switched to a session stage waiting for pricing update, wherein the second server is used for sending a pricing request to a pricing system according to the pricing request parameter carried by the asynchronous pricing instruction; a pricing calculation result fed back by the pricing system is received, an engineering event callback is taken as a trigger type identifier, and a first server is called back so as to generate a pricing display reply and feed back to the user. The application can decouple the AI service and the pricing calculation process.

Description

Technical Field

[0001] This application relates to the field of artificial intelligence technology, and in particular to an asynchronous quotation method, apparatus, storage medium, and electronic device. Background Technology

[0002] With the widespread application of instant messaging platforms in financial sales scenarios such as insurance, users can use these platforms to propose personalized solution requirements and expect to receive quotes instantly.

[0003] Currently, a synchronous blocking pricing method is commonly used when providing quotes to users. However, this method tightly couples the AI ​​(Artificial Intelligence) service with the pricing calculation process. During the pricing calculation, the AI ​​service is always in a waiting state and cannot continue the dialogue, resulting in a fragmented user experience, which is particularly noticeable when the pricing calculation takes a long time. Summary of the Invention

[0004] In view of this, this application provides an asynchronous quotation method, apparatus, storage medium, and electronic device, the main purpose of which is to decouple the AI ​​service from the quotation calculation process, so that the AI ​​service does not block and wait during the quotation calculation process, thereby improving the user experience.

[0005] According to a first aspect of this application, an asynchronous quoting method is provided, applied to a first server, the method comprising: Receive a first dialogue request sent by the second server, the first dialogue request carrying user messages and historical dialogues; Based on the user messages and the historical conversations, identify the user's desired solution intent and simultaneously determine the next target processing action for the user; If the target processing action is a normal dialogue, then a dialogue response is generated; If the dialogue response contains a price adjustment signal, then based on the intent of the demand solution, an asynchronous price quotation instruction is generated and sent to the second server, and the current session stage is switched to a session stage waiting for price updates, ending the response. The second server is used to send a price quotation request to the quotation system according to the price quotation request parameters carried by the asynchronous price quotation instruction; receive the price calculation result fed back by the quotation system; and send a second dialogue request to the first server with the engineering event callback as the trigger type identifier. Based on the second dialogue request, a quote display response is generated and fed back to the user through the second server.

[0006] According to a second aspect of this application, an asynchronous quoting method is provided, applied to a second server, the method comprising: Based on user messages and historical conversations, a first conversation request is sent to a first server. The first server is used to identify the user's desired solution intent based on the user messages and historical conversations, and simultaneously determine the next target processing action for the user. If the target processing action is a normal conversation, a conversation response is generated. If the conversation response contains a price adjustment signal, an asynchronous price instruction is generated based on the desired solution intent and sent to the second server. The current session stage is then switched to a session stage waiting for price updates, and the current response ends. Based on the quotation request parameters carried in the asynchronous quotation instruction, a quotation request is sent to the quotation system; The system receives the quotation calculation result from the quotation system, uses the engineering event callback as the trigger type identifier, and sends a second dialogue request to the first server so that the first server can generate a quotation display response based on the second dialogue request. The system receives the price quote display reply sent by the first server and sends it back to the user.

[0007] According to a third aspect of this application, an asynchronous quotation apparatus is provided, the apparatus comprising: The first receiving unit is used to receive a first dialogue request sent by the second server, wherein the first dialogue request carries user messages and historical dialogues. The identification unit is used to identify the user's desired solution intent based on the user message and the historical dialogue, and simultaneously determine the next target processing action for the user; A generation unit is used to generate a dialogue response if the target processing action is a normal dialogue. The generation unit is further configured to, if the dialogue response contains a price adjustment signal, generate an asynchronous price instruction based on the intent of the demand solution and send it to the second server, switch the current session phase to a session phase waiting for price updates, and end the current response. The second server is configured to send a price request to the price system according to the price request parameters carried by the asynchronous price instruction; receive the price calculation result fed back by the price system; and send a second dialogue request to the first server using an engineering event callback as the trigger type identifier. The first sending unit is used to generate a quote display reply based on the second dialogue request and to send it back to the user through the second server.

[0008] According to a fourth aspect of this application, an asynchronous quotation apparatus is provided, the apparatus comprising: The second sending unit is used to send a first dialogue request to the first server based on user messages and historical dialogues. The first server is used to identify the user's desired solution intent based on the user messages and historical dialogues, and simultaneously determine the next target processing action for the user. If the target processing action is a normal dialogue, a dialogue response is generated. If the dialogue response contains a price adjustment signal, an asynchronous price instruction is generated based on the desired solution intent and sent to the second server. The current session stage is then switched to a session stage waiting for price updates, and the current response ends. The second sending unit is further configured to send a quotation request to the quotation system according to the quotation request parameters carried by the asynchronous quotation instruction; The second sending unit is also used to receive the quotation calculation result fed back by the quotation system, and send a second dialogue request to the first server with the engineering event callback as the trigger type identifier, so that the first server can generate a quotation display reply based on the second dialogue request; The second receiving unit is used to receive the quotation display reply sent by the first server and to provide feedback to the user.

[0009] According to a fifth aspect of this application, a storage medium is provided that stores a computer program thereon, which, when executed by a processor, implements the asynchronous quoting method applied to a first server as described above.

[0010] According to a sixth aspect of this application, an electronic device is provided, including a storage medium, a processor, and a computer program stored on the storage medium and executable on the processor, wherein the processor executes the program to implement the asynchronous quoting method applied to a first server as described above.

[0011] According to a seventh aspect of this application, a storage medium is provided having a computer program stored thereon, which, when executed by a processor, implements the asynchronous quoting method applied to a second server as described above.

[0012] According to an eighth aspect of this application, an electronic device is provided, including a storage medium, a processor, and a computer program stored on the storage medium and executable on the processor, wherein the processor executes the program to implement the asynchronous quoting method applied to a second server as described above.

[0013] By employing the above technical solutions, this application provides an asynchronous pricing method, apparatus, storage medium, and electronic device. Compared to existing synchronous blocking pricing methods, the AI ​​service returns immediately after issuing the asynchronous pricing instruction, ending the current response and switching the current session phase to a session phase waiting for pricing updates. This avoids blocking while waiting for pricing calculation results, allowing pricing calculations to occur asynchronously downstream. This decouples the AI ​​service from the pricing calculation process. Since the AI ​​service's response time is unaffected by the pricing calculation time, it improves user experience and architectural scalability. Furthermore, because this application has a clearly defined phase—the session phase waiting for pricing updates—the AI ​​service will not falsify pricing calculation results, further enhancing the user experience.

[0014] The above description is only an overview of the technical solution of this application. In order to better understand the technical means of this application and to implement it in accordance with the contents of the specification, and to make the above and other objects, features and advantages of this application more obvious and understandable, the following are specific embodiments of this application. Attached Figure Description

[0015] The accompanying drawings, which are included to provide a further understanding of this application and form part of this application, illustrate exemplary embodiments and are used to explain this application, but do not constitute an undue limitation of this application. In the drawings: Figure 1 A flowchart illustrating an asynchronous quoting method provided in an embodiment of this application is shown; Figure 2 A schematic diagram of the system architecture provided in an embodiment of this application is shown; Figure 3 This paper illustrates a flowchart of the intent extraction and constraint verification method provided in an embodiment of this application. Figure 4 This paper illustrates a flowchart of a feature hash deduplication method provided in an embodiment of this application. Figure 5 A schematic diagram illustrating the interaction flow between the first server and the second server provided in an embodiment of this application is shown. Figure 6 A flowchart illustrating another asynchronous quoting method provided in an embodiment of this application is shown; Figure 7 This paper shows a schematic diagram of the structure of an asynchronous quotation device provided in an embodiment of this application; Figure 8 A schematic diagram of another asynchronous quotation device provided in an embodiment of this application is shown. Detailed Implementation

[0016] The present application will be described in detail below with reference to the accompanying drawings and embodiments. It should be noted that, unless otherwise specified, the embodiments and features described in the embodiments of the present application can be combined with each other.

[0017] In existing technologies, AI services are tightly coupled with the quotation calculation process. During quotation calculation, the AI ​​service is always in a waiting state and cannot continue the dialogue, resulting in a fragmented user experience.

[0018] To address the aforementioned issues, embodiments of the present invention provide an asynchronous quotation method, applied to a first server, namely an AI server, such as... Figure 1 As shown, the method includes: Step 10: Receive the first dialogue request sent by the second server. The first dialogue request carries user messages and historical dialogues.

[0019] Among them, user messages are messages currently sent by the user through an instant messaging platform (such as WeChat), and historical conversations are the context of conversations between the user and the platform.

[0020] In this embodiment of the invention, when a user sends a message to a second server (the calling server) through an instant messaging platform, the second server obtains the user message and, based on the user message and previous historical conversations with the user, generates a first dialogue request and sends it to the first server (the AI ​​server). The first dialogue request carries the user message and historical conversations. Thus, the first server (the AI ​​server) can receive the first dialogue request.

[0021] Step 20: Based on the user messages and the historical dialogues, identify the user's desired solution intent, and simultaneously determine the next target processing action for the user.

[0022] The target processing action can be any one of the following: defeat, transfer to human agent, or normal dialogue. Furthermore, the meaning of the intent of the demand solution differs in different business scenarios. In an insurance sales scenario, the intent of the demand solution is the user's intent regarding the type of insurance; in a financial product recommendation scenario, it is the user's intent regarding the financial product; in an e-commerce customized product scenario, it is the user's intent regarding product configuration; and in a corporate procurement scenario, it is the user's procurement needs. It should be noted that the business scenarios involved in the embodiments of this invention are not limited to the above-listed examples and can also include other business scenarios.

[0023] In this embodiment of the invention, after receiving the first dialogue request, the first server (AI server) simultaneously starts two sets of concurrent tasks, such as... Figure 2As shown, these two sets of concurrent tasks are intent recognition and action determination, respectively. For the intent recognition and action determination process, step 20 specifically includes: using an intent extraction agent to extract the product type field and its corresponding values ​​from the user message and the historical dialogue; determining the user's desired solution intent based on the product type field and its corresponding values; simultaneously using an action determination agent group to determine the processing actions involved by the user, wherein the action determination agent group includes a user state determination agent, a defeat detection agent, a conversion agent, and a knowledge base retrieval agent executed in parallel; and determining the target processing action according to the processing actions and their corresponding priorities.

[0024] Specifically, when performing intent recognition, it is started asynchronously in advance using asyncio.create_task(). The intent extraction agent is used to identify the user's intent for the solution from the multi-turn dialogue history (user messages and historical dialogues). Specifically, the product type field and its corresponding value are output. In the insurance sales scenario, the structured fields of insurance type and coverage level are output. The above intent recognition process runs independently in the background and does not block the main process.

[0025] like Figure 3 As shown, the intent extraction agent, centered on a large language model, extracts the user's insurance intent from multi-turn dialogue contexts in a car insurance sales scenario. The output is a structured JSON containing 10 insurance type fields, specifically including third-party liability insurance, out-of-pocket medical insurance, additional holiday double coverage, vehicle damage insurance, seat insurance, seat out-of-pocket medical insurance, driver and passenger insurance (Safe Travel Insurance), home delivery of medicine, mental distress compensation, and external power grid failure loss insurance (new energy only). Each field has a predefined enumeration value (e.g., third-party liability insurance values: 1 million / 1.5 million / 2 million / 3 million / not required). Incremental updates based on historical baseline schemes are supported, meaning only the values ​​of fields explicitly mentioned by the user are modified, while the values ​​of other fields remain unchanged.

[0026] When making action decisions, the user status judgment agent, the defeat detection agent, the transfer to artificial intelligence agent, and the knowledge base retrieval agent execute in parallel using the run_parallel_agent_tasks() method to jointly determine the next action to be taken for the user. These actions include normal dialogue, defeat, and transfer to human agent. Defeat specifically refers to situations such as receiving a complaint from the user or the user explicitly refusing.

[0027] The two sets of tasks mentioned above run synchronously and concurrently. After the action judgment agent group completes its task, it determines the next target processing action for the user based on the priority routing rules. For example, the priority of defeat is higher than the priority of transferring to human agent, the priority of transferring to human agent is higher than the priority of no response, and the priority of no response is higher than the priority of normal conversation.

[0028] It should be noted that the intent to extract the demand solution determined by the agent is only read when the dialogue agent determines that the price needs to be adjusted; otherwise, the task is canceled to save computing resources.

[0029] In this embodiment of the invention, the intent extraction agent and the action judgment agent group are executed concurrently. The total execution time depends on the slower group, rather than the sum of the serial execution times of the two groups. In actual deployment, the execution time of the intent extraction agent (LLM call) and the action judgment agent group (multiple LLM calls in parallel) is similar. Concurrent execution can reduce the response time by about 40%-60%, thereby improving the user dialogue experience.

[0030] Step 30: If the target processing action is a normal dialogue, then generate a dialogue response.

[0031] For embodiments of the present invention, such as Figure 1 As shown, if the target processing action is a normal dialogue, the dialogue response is generated using the dialogue agent based on the retrieval results of the knowledge base retrieval agent, the user message, and the historical dialogue; if the target processing action is to switch to human intervention or to fail, the requirement scheme intent recognition task is canceled, and a corresponding engineering event instruction is generated and added to the engineering event list, and the engineering event list is fed back to the second server.

[0032] Step 40: If the dialogue response contains a quote adjustment signal, then based on the intent of the demand solution, generate an asynchronous quote instruction and send it to the second server, switch the current session stage to the session stage of waiting for quote updates, and end this response.

[0033] The second server is used to send a quotation request to the quotation system according to the quotation request parameters carried by the asynchronous quotation instruction; receive the quotation calculation result fed back by the quotation system; and send a second dialogue request to the first server with the project event callback as the trigger type identifier.

[0034] In this embodiment of the invention, if the dialogue response contains a target structured instruction (such as a structured tool_call instruction), then it is determined that the dialogue response contains a price adjustment signal; using product rule constraints, the intent of the demand solution is mapped to compliant price request parameters; based on the price request parameters, an asynchronous price instruction is generated and sent to the second server.

[0035] Specifically, after confirming that the dialogue response contains a signal to adjust the price, such as Figure 3As shown, product rule constraint validation is performed on the intent of the demand solution, that is, the natural language intent is mapped to compliant quotation request parameters. In insurance sales scenarios, product rules and constraints specifically include: mandatory attachment of supplementary insurance to main insurance, binding rules between coverage tiers and supporting services, linkage rules for complimentary insurance, and consistency rules for seat insurance. Specifically, in the mandatory attachment of supplementary insurance to main insurance, out-of-pocket medical expenses insurance, supplementary holiday doubling insurance, and mental distress compensation are all supplementary insurance to third-party liability insurance, while external power grid failure loss insurance is a supplementary insurance to vehicle damage insurance (only for new energy vehicles). In the binding rules between coverage tiers and supporting services, when the driver and passenger insurance coverage is 1.427 million or 1.736 million per seat, home delivery of medicines is mandatory; when the driver and passenger insurance coverage is 844,000 per seat, home delivery of medicines is optional according to user needs; if driver and passenger insurance is not purchased, home delivery of medicines is not purchased at all. In the linkage rules for complimentary insurance, when third-party liability insurance is purchased, supplementary services such as roadside assistance, valet driving, and vehicle inspection are provided free of charge; if third-party liability insurance is not purchased, the above complimentary services are automatically cancelled. In the consistency rules for seat insurance, the coverage of out-of-pocket medical expenses insurance for seats is always consistent with the coverage of seat insurance.

[0036] The validated demand plan is converted into a standardized quotation request parameter format (commercial_insurances list + non_car_insurance_info) for use by downstream quotation APIs. The commercial_insurances list and non_car_insurance_info can correspond to two different product types. Both the commercial_insurances list and non_car_insurance_info include the identifier, value, and selection information for each product type. The selection information specifically refers to whether the product has been selected by the user.

[0037] It should be noted that the format of the quotation request parameters can be set according to actual business needs and is not limited to the above format.

[0038] Based on the combination of intent extraction agents and product rule constraints, users can describe product requirements in natural language. The first server (AI server) can automatically convert natural language into compliant quotation calculation parameters without requiring users to operate structured forms. Product rule constraints can be dynamically maintained through prompts, and changes do not require modification of the engineering code.

[0039] When generating an asynchronous quote instruction, a list of triples is generated based on the quote request parameters. This list contains the identifier, value, and selection information for each product type. The triple list is serialized into a standard format string, which is then used as the scheme feature hash key. The scheme feature hash key is compared with the set of scheme feature hash keys submitted in the current session, stored in the cache. If the scheme feature hash key exists in the set of submitted scheme feature hash keys, a prompt text is returned directly. If the scheme feature hash key does not exist in the set of submitted scheme feature hash keys, the scheme feature hash key is written into the cache, and the quote request parameters are encapsulated into an asynchronous quote instruction and sent to the second server.

[0040] Specifically, such as Figure 4 As shown, commercial_insurances is sorted by insurance type name, and a list of triples {insurance type name, coverage amount, whether selected} is extracted. Non-auto insurance is sorted by product code. The above information is serialized into a standardized JSON string and used as the scheme feature hash key (quote_key). Then, the scheme feature hash key (quote_key) is compared with the set of scheme feature hash keys submitted in the current session stored in the cache (Redis). If the scheme feature hash key exists in the set, a prompt text is returned directly, such as "The scheme is the same as last time, no change, what else needs to be modified?". If the scheme feature hash key does not exist in the set, it means that the scheme is a new scheme. At this time, the scheme feature hash key (quote_key) needs to be written to the cache, and the quotation request parameters are encapsulated into an asynchronous quotation instruction (OFFLINE_GEN_INSURANCE_PRICE) and sent to the second server (calling the server). At the same time, the current session stage is switched to the session stage of waiting for quotation update (customer_new_quote), the current response ends, and the waiting for quotation calculation results is not blocked.

[0041] After the downstream quotation system asynchronously completes the quotation and sends it back to the second server (the calling server), the second server (the calling server) calculates the result based on the quotation and sends a callback to the first server (the AI ​​server) using an engineering event callback as the trigger type identifier. This means it sends a second dialogue request to the first server. The overall interaction flow is as follows: Figure 5 As shown.

[0042] This invention calculates a standardized feature hash key for each price adjustment request and compares it with the historical set to intercept duplicate schemes. This avoids repeatedly triggering the pricing system for the same scheme, thereby reducing system load and providing users with clear prompts about duplicate schemes.

[0043] In addition, the first server (AI server) returns immediately after issuing the asynchronous quotation instruction, without blocking and waiting for the quotation calculation result. The quotation calculation is performed asynchronously downstream. Compared with the synchronous blocking quotation method, the response time of the AI ​​service is not affected by the quotation calculation time, and the architecture has better scalability.

[0044] Step 50: Based on the second dialogue request, generate a quote display response and send it back to the user through the second server.

[0045] For embodiments of the present invention, such as Figure 5 As shown, the first server (AI server) parses the quote calculation results (such as insurance coverage, premium details, marketing gifts, etc.) carried in the second dialogue request, injects them into the dialogue context, generates a quote display response, and sends it to the second server (the calling server). The second server then returns this generated quote display response to the user. Figure 5 Throughout the entire interaction process, AI dialogue and price calculation are completely decoupled.

[0046] In some embodiments, a preset text splitting model is used to split the dialogue reply and / or the quote display reply into several candidate short messages according to semantic boundaries; the several candidate short messages are standardized to achieve an optimal balance between message length and expression completeness; the standardized short messages are sent to the second server so that the second server can send them to the user in sequence through a serial delivery action.

[0047] Specifically, after the dialogue agent generates the dialogue response text, it needs to be processed by the message intelligent segmentation module before being sent to the user. This module adopts a two-stage processing approach that combines a preset large text segmentation model (LLM model) with rules. In the first stage, the preset large text segmentation model splits the entire response into several candidate short messages according to semantic boundaries. In the second stage, the rule post-processing engine performs standardization processing. For example, short messages with a length of less than 50 characters are merged into the next short message, more than 3 short messages are merged in pairs, and the last short message with less than 30 characters is merged into the second to last short message, thereby ensuring that the length and expression completeness of each message are optimally balanced.

[0048] The split short messages (several short messages after standardization) are sent to the second server so that the second server can send them to the user in sequence through a serial delivery action (QX_SEND_LIST, ordered: serial), thereby ensuring the sequential arrival of multiple messages in the instant messaging scenario.

[0049] In some embodiments, if the user state determination agent detects that the user is in a waiting-to-speak state, it embeds several standardized short messages into a delay field to obtain a message-embedded delay field; based on the message-embedded delay field, it generates a message delay delivery instruction; and feeds the message delay delivery instruction back to the second server, wherein the second server is used to start a countdown based on the waiting time in the delay field, while simultaneously listening for user messages; if no user message is received before the countdown time expires, it uses a scheduled delivery callback as the trigger type identifier and sends a third dialogue request to the first server based on the delivery message text list in the delay field; based on the third dialogue request, it generates a delivery reply and sends it to the user through the second server. The delay field includes a delivery message text list and a waiting time.

[0050] Specifically, when the user status judgment agent detects that the user is in a waiting state to speak, it deeply integrates the message splitting result with the delayed delivery mechanism, that is, it directly embeds the split short messages into the pre-stored content (delay_msg[0]['msg']) of the delay field. When the delayed delivery is triggered, the pre-stored multiple message sequences are sent in order, thereby realizing intelligent segmented follow-up in the delayed scenario.

[0051] This invention employs a two-stage processing approach combining a pre-defined large text segmentation model (LLM model) with rules. This approach automatically transforms long AI replies into multiple short messages that conform to the reading habits of instant messaging users, and ensures message order through serial delivery. Furthermore, by deeply integrating with a delayed delivery mechanism, it enables the output of segmented messages with a natural rhythm even in delayed follow-up scenarios, thereby improving the overall conversational experience of instant messaging channels.

[0052] To make the technical implementation of the embodiments of the present invention clearer, the specific implementation process of this solution will be described in detail using the following specific scenario as an example.

[0053] Scenario 1: User requests adjustment to insurance plan Step S1: The user sends the message "Change the third-party liability insurance to 3 million and remove the seat insurance".

[0054] Step S2: After the first server (AI server) obtains the user message, it simultaneously starts the intent extraction agent (background asynchronous task) and the action judgment agent group (executed in parallel). If the user status judgment agent in the action judgment agent group recognizes that the user status is a normal conversation and no special path such as defeat or transfer to human agent is triggered, then the conversational agent (ChatAgent) processing flow is entered.

[0055] Step S3: ChatAgent generates a dialogue response based on the dialogue context. If it determines that the price needs to be adjusted, the response will include a signal to adjust the price. I'll adjust your third-party liability insurance to 3 million yuan and remove the seat insurance. I'll also recalculate your premium.<tool_call> {"name": "get_offer"}< / tool_call> Step S4: The first server (AI server) detects the price adjustment signal, waits for the intent extraction agent to complete (at this point, concurrent execution has been completed), and obtains the extraction result, i.e., the intent of the demand solution: {"Third-Party Liability Insurance": 3 million RMB, "Out-of-Holiday Drug Insurance": Shared with Third-Party Liability Insurance, "Double Coverage During Holidays": Not Required, "Vehicle Damage Insurance": Required, "Seat Insurance": Not Required, "Seat Out-of-Holiday Drug Insurance": Not Required, "Driver and Passenger Insurance": 1.736 million RMB per seat, "Medicine Delivery": Required, "Mental Distress Compensation": Not Required, "External Power Grid Failure Loss Insurance": Not Required} Step S5: The product constraint verification module performs the verification: Seat insurance is "not required", seat out-of-pocket medication insurance automatically remains consistent (not required); driver and passenger insurance tier 1.736 million is forcibly bound to home delivery medication (already satisfied); no third-party liability insurance removal scenario, complimentary insurance linkage remains. Verification passes, convert to commercial_insurances list (e.g. Figure 3 (As shown).

[0056] Step S6: The feature hash deduplication module calculates the feature hash key `quote_key` for this scheme and compares it with the set of submitted schemes in Redis. If no match is found, a new scheme is proposed (e.g., ...). Figure 4 (As shown).

[0057] Step S7: The first server writes the quote_key to Redis, encapsulates the quote request parameters into an asynchronous quote command OFFLINE_GEN_INSURANCE_PRICE and sends it out, and at the same time switches the current session stage to customer_new_quote, thus ending this AI response.

[0058] Step S8: The downstream engineering system (quotation system) asynchronously completes the quotation calculation. The second server (calling server) calls back the first server (AI server) with caller=action, carrying the quotation calculation results (including premium details for each type of insurance and the URL of the quotation image).

[0059] Step S9: The first server (AI server) parses the quote calculation result, injects the insurance type, coverage amount, premium details, and marketing gift information into the dialogue context, and generates a quote display response (e.g., Figure 5 As shown in the figure, it is sent to the user through the second server (AI server).

[0060] Scenario 2: Users repeatedly submit the same solution Step S1: The user sends the message "Let's stick to the previous plan, with 2 million in third-party liability insurance" (consistent with the previously submitted plan).

[0061] Step S2: The first server (AI server) concurrently starts the intent extraction Agent and action judgment Agent group. The ChatAgent determines that the price needs to be adjusted and triggers the price adjustment process.

[0062] Step S3: The Agent extracts the extraction results. After product constraint verification, a commercial_insurances list is generated, which is exactly the same as the previously submitted scheme.

[0063] Step S4: The feature hash deduplication module calculates the feature hash key `quote_key` for this scheme and compares it with the set of submitted scheme keys in Redis. It detects duplicates (e.g., ...). Figure 4 (As shown).

[0064] Step S5: Skip the quotation calculation. The first server (AI server) directly returns the prompt text: "Your solution is the same as the last quotation. There is no change. Do you need to modify anything?", thereby avoiding repeated triggering of downstream quotation calculation.

[0065] This invention provides an asynchronous pricing method where the AI ​​service returns immediately after issuing an asynchronous pricing instruction, ending the current response and switching the current session to a session waiting for pricing updates. This avoids blocking the waiting for pricing calculation results, allowing pricing calculation to occur asynchronously downstream. This decouples the AI ​​service from the pricing calculation process. Since the AI ​​service's response time is unaffected by the pricing calculation time, it improves user experience and architectural scalability. Furthermore, because this invention has a clearly defined stage—the session waiting for pricing updates—the AI ​​service cannot fabricate pricing calculation results, further enhancing the user experience.

[0066] Furthermore, embodiments of the present invention provide an asynchronous quotation method, applied to a second server, i.e., a server-side caller, such as... Figure 6 As shown, the method includes: Step 60: Based on user messages and historical conversations, send the first conversation request to the first server.

[0067] The first server is used to identify the user's desired solution intent based on the user message and the historical dialogue, and simultaneously determine the next target processing action for the user; if the target processing action is a normal dialogue, a dialogue response is generated; if the dialogue response contains a price adjustment signal, an asynchronous price instruction is generated based on the desired solution intent and sent to the second server, and the current session stage is switched to a session stage waiting for price updates, ending the current response.

[0068] In this embodiment of the invention, when a user sends a message to a second server (the calling server) through an instant messaging platform, the second server obtains the user message and, based on the user message and previous historical conversations with the user, generates a first dialogue request and sends it to the first server (the AI ​​server). The first dialogue request carries the user message and historical conversations. After receiving the first dialogue request, the first server identifies the user's intended solution based on the user message and historical conversations, and simultaneously determines the next target processing action for the user. If the target processing action is a normal dialogue, the first server generates a dialogue response. If the dialogue response contains a price adjustment signal, it uses product rule constraints to map the intended solution to compliant price request parameters, encapsulates the price request parameters into an asynchronous price instruction (OFFLINE_GEN_INSURANCE_PRICE), sends it to the second server (the calling server), and simultaneously switches the current dialogue state to the session stage of waiting for price updates, ending the current response.

[0069] Step 70: Send a quotation request to the quotation system according to the quotation request parameters carried by the asynchronous quotation instruction.

[0070] In this embodiment of the invention, after receiving the asynchronous quotation instruction, the second server (calling server) sends a quotation request to the quotation system based on the quotation request parameters carried by the asynchronous quotation instruction.

[0071] Step 80: Receive the quotation calculation result fed back by the quotation system, and send a second dialogue request to the first server with the engineering event callback as the trigger type identifier, so that the first server can generate a quotation display response based on the second dialogue request.

[0072] In this embodiment of the invention, after the quotation system completes the calculation, it feeds back the quotation calculation result to the second server (calling server). The second server uses the engineering event callback as the trigger type identifier, and based on the quotation calculation result, sends a second dialogue request to the first server (AI server). The first server parses the quotation calculation result carried in the second dialogue request, injects it into the dialogue context, generates a quotation display response, and feeds it back to the second server (calling server). Step 90: Receive the price quote display reply sent by the first server and send it back to the user.

[0073] In some embodiments, the system receives a number of standardized short messages sent by the first server. The first server is configured to use a preset text splitting model to split the dialogue reply and / or the quote display reply into a number of candidate short messages according to semantic boundaries; to standardize the number of candidate short messages to achieve an optimal balance between message length and expression completeness; and the second server sends the number of standardized short messages to the user in sequence through a serial delivery action.

[0074] Specifically, after the dialogue agent on the first server (AI server) generates the dialogue response text, it needs to be processed by the message intelligent segmentation module before being sent to the user. This module adopts a two-stage processing approach that combines a preset large text segmentation model (LLM model) with rules. In the first stage, the preset large text segmentation model splits the entire response into several candidate short messages according to semantic boundaries. In the second stage, the rule post-processing engine performs standardization processing. For example, short messages with a length of less than 50 characters are merged into the next short message, more than 3 short messages are merged in pairs, and the last short message with less than 30 characters is merged into the second to last short message, thereby ensuring that the length and expression completeness of each message are optimally balanced.

[0075] The split short messages (several short messages after standardization) are sent to the second server. The second server then sends them to the user in sequence via a serial delivery action (QX_SEND_LIST, ordered: serial), thereby ensuring the sequential arrival of multiple messages in the instant messaging scenario.

[0076] In some embodiments, the system receives a delayed message delivery instruction sent by the first server. The first server, if a user status determination agent detects that the user is in a waiting-to-speak state, embeds several standardized short messages into a delay field and generates a delayed message delivery instruction based on the embedded delay field. The second server starts a countdown based on the waiting time in the delay field and simultaneously listens for user messages. If no user message is received by the end of the countdown, a scheduled delivery callback is used as the trigger type identifier, and a third dialogue request is sent to the first server based on the delivery message text list in the delay field, so that the first server generates a delivery response based on the third object request. The system receives the delivery response sent by the first server and sends it to the user.

[0077] Specifically, when the user status judgment agent of the first server detects that the user is in a waiting state, it deeply integrates the message splitting result with the delayed delivery mechanism, that is, it directly embeds the split short messages into the pre-stored content (delay_msg[0]['msg']) of the delay field. When the delayed delivery is triggered, the pre-stored multiple message sequences are sent in order, thereby realizing intelligent segmented follow-up in the delayed scenario.

[0078] To make the technical solutions of the embodiments of the present invention clearer, the asynchronous quotation closed-loop interaction process between the user, the first server (AI server), and the second server (calling server) in the embodiments of the present invention will now be described in detail.

[0079] Step S1: Receive user messages. The first server asynchronously starts the intent extraction Agent using `asyncio.create_task()`, and simultaneously starts the user status judgment Agent, defeat detection Agent, manual detection Agent, and knowledge base retrieval Agent in parallel using `run_parallel_agent_tasks()`. These two sets of tasks are executed concurrently. Figure 2 As shown.

[0080] Step S2: After the Action Judgment Agent group completes its execution, the subsequent processing path is determined according to the priority routing rules: if a defeat or switch to manual control signal is detected, the corresponding action is executed directly, the intention to extract Agent task is canceled, and the process ends; otherwise, continue to execute Step S3.

[0081] Step S3: The ChatAgent generates a response based on the knowledge base retrieval results and the dialogue context. If the response does not contain a price adjustment signal, the Intent Extraction Agent task is canceled, and the dialogue response is sent directly, ending the process. If the response contains a price adjustment signal (<tool_call> If + get_offer), then proceed to step S4.

[0082] Step S4: Wait for the intent extraction agent to complete, and obtain the structured extraction result (demand solution intent) containing 10 insurance type fields, such as... Figure 3 As shown, the current session's base insurance information is read from Redis as the initial state for extraction.

[0083] Step S5: Perform product constraint validation on the extracted results (intended requirements). Based on the relationship between supplementary insurance and main insurance, the rules for binding coverage tiers and supporting services, the rules for linking complimentary insurance, and the rules for consistency of seat insurance, generate a compliant list of commercial_insurances and non-motor insurance parameters, i.e., quote request parameters, such as... Figure 3 As shown.

[0084] Step S6: Calculate the feature hash key for the verified scheme and compare it with the set of committed scheme keys in Redis, such as... Figure 4 As shown. If the result is a duplicate, a prompt text is returned and the process ends; if it is a new solution, the key is written to the Redis cache and step S7 is executed.

[0085] Step S7: Encapsulate the quote request parameters into an asynchronous action command OFFLINE_GEN_INSURANCE_PRICE, and send it along with the dialogue response text to the second server (the calling server). Simultaneously, switch the session phase in Redis to customer_new_quote. The AI ​​service completes this processing and does not wait for the quote calculation result. Figure 5 As shown.

[0086] Step S8: After the pricing system asynchronously completes the pricing calculation, the second server (the calling server) calls back to the first server (the AI ​​server) using an engineering event callback as the trigger type identifier, carrying the insurance pricing calculation result. The first server (the AI ​​server) calls the pricing parsing module to extract information such as the insurance coverage amount, premium details, and marketing value-added services, injects it into the dialogue context, generates a pricing display reply, and sends it to the user, completing the pricing loop. Figure 5 As shown.

[0087] Step S9: The text replies generated by the dialogue agent (including the ordinary dialogue replies in step S3 and the price display replies in step S8) are all processed by the message intelligent splitting module before being sent out: LLM splits the messages according to semantic boundaries, the rule engine standardizes the number and length of messages, and the output message list is serially delivered to the user via QX_SEND_LIST (ordered: serial). If there is a "wait to speak" delay instruction at the same time, the split message list is embedded in the delay_msg[0]['msg'] field and sent out together with the delay instruction, and sent out in order when the delay is triggered.

[0088] This invention provides an asynchronous pricing method where the AI ​​service returns immediately after issuing an asynchronous pricing instruction, ending the current response and switching the current session to a session waiting for pricing updates. This avoids blocking the waiting for pricing calculation results, allowing pricing calculation to occur asynchronously downstream. This decouples the AI ​​service from the pricing calculation process. Since the AI ​​service's response time is unaffected by the pricing calculation time, it improves user experience and architectural scalability. Furthermore, because this invention has a clearly defined stage—the session waiting for pricing updates—the AI ​​service cannot fabricate pricing calculation results, further enhancing the user experience.

[0089] Furthermore, as Figure 1 The specific implementation of the method shown in this embodiment provides an asynchronous quotation device, such as... Figure 7 As shown, the device includes: a first receiving unit 101, an identification unit 102, a generating unit 103, and a first transmitting unit 104.

[0090] The first receiving unit 101 can be used to receive a first dialogue request sent by the second server, the first dialogue request carrying user messages and historical dialogues.

[0091] The identification unit 102 can be used to identify the user's desired solution intent based on the user message and the historical dialogue, and simultaneously determine the next target processing action for the user.

[0092] The generation unit 103 can be used to generate a dialogue response if the target processing action is a normal dialogue.

[0093] The generation unit 103 can also be used to generate an asynchronous quotation instruction and send it to the second server based on the intent of the demand scheme if the dialogue response contains a quotation adjustment signal, and switch the current session stage to a session stage waiting for quotation updates, thereby ending the current response. The second server is used to send a quotation request to the quotation system according to the quotation request parameters carried by the asynchronous quotation instruction; receive the quotation calculation result fed back by the quotation system; and send a second dialogue request to the first server with the engineering event callback as the trigger type identifier.

[0094] The first sending unit 104 can be used to generate a quote display reply based on the second dialogue request and feed it back to the user through the second server.

[0095] In some embodiments, the identification unit 102 may be specifically configured to: utilize an intent extraction agent to extract a product type field and its corresponding value from the user message and the historical dialogue; determine the user's desired solution intent based on the product type field and its corresponding value; simultaneously utilize an action judgment agent group to determine the processing action involved by the user, wherein the action judgment agent group includes a user state judgment agent, a defeat detection agent, a conversion to artificial intelligence agent, and a knowledge base retrieval agent that are executed in parallel; and determine the target processing action according to the processing action and its corresponding priority.

[0096] In some embodiments, the generation unit 103 may be specifically used to generate the dialogue response using a dialogue agent based on the retrieval results of the knowledge base retrieval agent, the user message, and the historical dialogue if the target processing action is a normal dialogue; if the target processing action is to switch to human intervention or to fail, cancel the requirement scheme intent recognition task, generate corresponding engineering event instructions and add them to the engineering event list, and feed the engineering event list back to the second server.

[0097] In some embodiments, the generation unit 103 includes: a determination module, a mapping module, and a generation module.

[0098] The determining module can be used to determine that the dialogue response contains an adjustment quote signal if the dialogue response contains a target structured instruction.

[0099] The mapping module can be used to map the intent of the demand solution into compliant quotation request parameters by utilizing product rule constraints.

[0100] The generation module can be used to generate an asynchronous quotation instruction based on the quotation request parameters and send it to the second server.

[0101] In some embodiments, the generation module may be specifically configured to generate a list of triples based on the quotation request parameters, wherein the list of triples contains the identifier, value, and selection information of each type of product; serialize the list of triples into a standard format string, and determine the standard format string as the scheme feature hash key; compare the scheme feature hash key with the set of scheme feature hash keys submitted in the current session stored in the cache; if the scheme feature hash key exists in the set of submitted scheme feature hash keys, a prompt text is returned directly; if the scheme feature hash key does not exist in the set of submitted scheme feature hash keys, the scheme feature hash key is written into the cache, and the quotation request parameters are encapsulated into an asynchronous quotation instruction and sent to the second server.

[0102] In some embodiments, the first sending unit 104 may be specifically used to parse the quote calculation result carried in the second dialogue request, inject it into the dialogue context, and generate a quote display reply.

[0103] In some embodiments, the apparatus further includes a splitting unit and a processing unit.

[0104] The splitting unit can be used to split the dialogue response and / or the quote display response into several candidate short messages according to semantic boundaries using a preset text splitting model.

[0105] The processing unit can be used to standardize the candidate short messages to achieve an optimal balance between message length and completeness of expression.

[0106] The first sending unit 104 can also be used to send a number of standardized short messages to the second server so that the second server can send them to the user in sequence through a serial delivery action.

[0107] In some embodiments, the first sending unit 104 may be specifically configured to, if the user state judgment agent detects that the user is in a waiting state, embed a number of standardized short messages into a delay field to obtain a short message embedded delay field; generate a message delay delivery instruction based on the short message embedded delay field; and feed back the message delay delivery instruction to the second server, wherein the second server is configured to start a countdown based on the waiting time in the delay field and listen for user messages; if no user message is received before the countdown time expires, a scheduled delivery callback is used as the trigger type identifier, and a third dialogue request is sent to the first server based on the delivery message text list in the delay field; and a delivery reply is generated based on the third dialogue request and sent to the user through the second server.

[0108] It should be noted that other corresponding descriptions of the functional units involved in the asynchronous quotation device provided in this embodiment of the invention can be found in [reference]. Figure 1 The corresponding description in [the document] will not be repeated here.

[0109] Based on the above, Figure 1 Accordingly, this embodiment also provides a storage medium storing a computer program that, when executed by a processor, implements the above-described method. Figure 1 The asynchronous quoting method shown.

[0110] Based on this understanding, the technical solution of this application can be embodied in the form of a software product, which can be stored in a non-volatile storage medium (such as CD-ROM, USB flash drive, mobile hard drive, etc.) and includes several instructions to cause an electronic device (such as personal computer, server, or network device, etc.) to execute the methods of various implementation scenarios of this application.

[0111] Based on the above, Figure 1 The method shown, and Figure 7 To achieve the above objectives, the present application also provides an electronic device, specifically a personal computer, tablet computer, server, or other network device, as shown in the virtual device embodiment. This device includes a storage medium and a processor; the storage medium stores a computer program; the processor executes the computer program to achieve the above-described objectives. Figure 1 The asynchronous quoting method shown.

[0112] Optionally, the aforementioned physical devices may also include a user interface, a network interface, a camera, radio frequency (RF) circuitry, sensors, audio circuitry, a Wi-Fi module, etc. The user interface may include a display screen, input units such as a keyboard, etc., and optional user interfaces may also include USB interfaces, card reader interfaces, etc. The network interface may optionally include standard wired interfaces, wireless interfaces (such as Wi-Fi interfaces), etc.

[0113] Those skilled in the art will understand that the physical device structure provided in this embodiment does not constitute a limitation on the physical device, and may include more or fewer components, or combine certain components, or have different component arrangements.

[0114] The storage medium may also include an operating system and a network communication module. The operating system is a program that manages the hardware and software resources of the aforementioned physical device, supporting the operation of information processing programs and other software and / or programs. The network communication module is used to enable communication between the various components within the storage medium, as well as communication with other hardware and software in the information processing physical device.

[0115] Through the above description of the embodiments, those skilled in the art can clearly understand that this application can be implemented by means of software plus necessary general-purpose hardware platform, or it can be implemented by hardware.

[0116] Furthermore, as Figure 6 The specific implementation of the method shown in this embodiment provides an asynchronous quotation device, such as... Figure 8 As shown, the device includes a second transmitting unit 105 and a second receiving unit 106.

[0117] The second sending unit 105 can be used to send a first dialogue request to a first server based on user messages and historical dialogues. The first server is used to identify the user's demand plan intent based on the user messages and historical dialogues, and simultaneously determine the next target processing action for the user. If the target processing action is a normal dialogue, a dialogue response is generated. If the dialogue response contains a price adjustment signal, an asynchronous price instruction is generated based on the demand plan intent and sent to the second server. The current session stage is then switched to a session stage waiting for price updates, and the current response ends.

[0118] The second sending unit 105 can also be used to send a quotation request to the quotation system according to the quotation request parameters carried by the asynchronous quotation instruction.

[0119] The second sending unit 105 can also be used to receive the quotation calculation result fed back by the quotation system, and send a second dialogue request to the first server with the engineering event callback as the trigger type identifier, so that the first server can generate a quotation display reply based on the second dialogue request; The second receiving unit 106 can be used to receive the quotation display reply sent by the first server and provide feedback to the user.

[0120] In some embodiments, the second receiving unit 106 can also be used to receive a number of standardized short messages sent by the first server, wherein the first server is used to use a preset text splitting model to split the dialogue reply and / or the quotation display reply into a number of candidate short messages according to semantic boundaries; and to perform standardization processing on the number of candidate short messages to achieve an optimal balance between message length and expression completeness.

[0121] The second sending unit 105 can also be used to send the standardized short messages sequentially to the user through a serial delivery action.

[0122] In some embodiments, the device further includes a monitoring unit.

[0123] The second receiving unit 106 can also be used to receive a message delay delivery instruction sent by the first server, wherein the first server is used to embed a number of standardized short messages into a delay field if the user status judgment agent detects that the user is in a waiting state, and to generate a message delay delivery instruction based on the delay field after the short messages are embedded.

[0124] The monitoring unit can be used to start a countdown based on the waiting time in the delay field, while simultaneously monitoring user messages.

[0125] The second sending unit 105 can also be used to send a third dialogue request to the first server if no user message is received by the countdown time, using the scheduled outreach callback as the trigger type identifier and based on the outreach message text list in the delay field, so that the first server can generate an outreach response based on the third object request.

[0126] The second receiving unit 106 can also be used to receive the reach response sent by the first server and send it to the user.

[0127] It should be noted that other corresponding descriptions of the functional units involved in the asynchronous quotation device provided in this embodiment of the invention can be found in [reference]. Figure 6 The corresponding description in [the document] will not be repeated here.

[0128] Based on the above, Figure 6 Accordingly, this embodiment also provides a storage medium storing a computer program that, when executed by a processor, implements the above-described method. Figure 6 The asynchronous quoting method shown.

[0129] Based on this understanding, the technical solution of this application can be embodied in the form of a software product, which can be stored in a non-volatile storage medium (such as CD-ROM, USB flash drive, mobile hard drive, etc.) and includes several instructions to cause an electronic device (such as personal computer, server, or network device, etc.) to execute the methods of various implementation scenarios of this application.

[0130] Based on the above, Figure 6 The method shown, and Figure 8 To achieve the above objectives, the present application also provides an electronic device, specifically a personal computer, tablet computer, server, or other network device, as shown in the virtual device embodiment. This device includes a storage medium and a processor; the storage medium stores a computer program; the processor executes the computer program to achieve the above-described objectives. Figure 6 The asynchronous quoting method shown.

[0131] Optionally, the aforementioned physical devices may also include a user interface, a network interface, a camera, radio frequency (RF) circuitry, sensors, audio circuitry, a Wi-Fi module, etc. The user interface may include a display screen, input units such as a keyboard, etc., and optional user interfaces may also include USB interfaces, card reader interfaces, etc. The network interface may optionally include standard wired interfaces, wireless interfaces (such as Wi-Fi interfaces), etc.

[0132] Those skilled in the art will understand that the physical device structure provided in this embodiment does not constitute a limitation on the physical device, and may include more or fewer components, or combine certain components, or have different component arrangements.

[0133] The storage medium may also include an operating system and a network communication module. The operating system is a program that manages the hardware and software resources of the aforementioned physical device, supporting the operation of information processing programs and other software and / or programs. The network communication module is used to enable communication between the various components within the storage medium, as well as communication with other hardware and software in the information processing physical device.

[0134] Through the above description of the embodiments, those skilled in the art can clearly understand that this application can be implemented by means of software plus necessary general-purpose hardware platform, or it can be implemented by hardware.

[0135] In this embodiment of the invention, the AI ​​service returns immediately after issuing the asynchronous quotation command, ending the current response and switching the current session phase to a session phase waiting for quotation updates. This avoids blocking the waiting for quotation calculation results, allowing quotation calculation to occur asynchronously downstream. This decouples the AI ​​service from the quotation calculation process. Since the AI ​​service's response time is unaffected by the quotation calculation time, it improves user experience and architectural scalability. Furthermore, because this embodiment of the invention has a clearly defined phase—the session phase waiting for quotation updates—the AI ​​service cannot fabricate quotation calculation results, further enhancing the user experience.

[0136] Those skilled in the art will understand that the accompanying drawings are merely schematic diagrams of a preferred embodiment, and the modules or processes shown in the drawings are not necessarily essential for implementing this application. Those skilled in the art will understand that the modules in the apparatus of the embodiment can be distributed within the apparatus of the embodiment as described, or can be modified to be located in one or more apparatuses different from this embodiment. The modules of the above-described embodiment can be combined into one module, or further divided into multiple sub-modules.

[0137] The serial numbers in this application are for descriptive purposes only and do not represent the superiority or inferiority of any particular implementation scenario. The above disclosures are merely a few specific implementation scenarios of this application; however, this application is not limited thereto, and any variations conceived by those skilled in the art should fall within the protection scope of this application.

Claims

1. An asynchronous quotation method, characterized in that, Applied to the first server, including: Receive a first dialogue request sent by the second server, the first dialogue request carrying user messages and historical dialogues; Based on the user messages and the historical conversations, identify the user's desired solution intent and simultaneously determine the next target processing action for the user; If the target processing action is a normal dialogue, then a dialogue response is generated; If the dialogue response contains a price adjustment signal, then based on the intent of the demand solution, an asynchronous price quotation instruction is generated and sent to the second server, and the current session stage is switched to a session stage waiting for price updates, ending the response. The second server is used to send a price quotation request to the quotation system according to the price quotation request parameters carried by the asynchronous price quotation instruction; receive the price calculation result fed back by the quotation system; and send a second dialogue request to the first server with the engineering event callback as the trigger type identifier. Based on the second dialogue request, a quote display response is generated and fed back to the user through the second server.

2. The method according to claim 1, characterized in that, The process of identifying the user's desired solution intent based on the user messages and historical dialogues, and simultaneously determining the next target processing action for the user, includes: Using an intent extraction agent, the product type field and its corresponding values ​​are extracted from the user messages and the historical dialogues. Based on the product type field and its corresponding values, the user's intended solution is determined; The action judgment intelligent agent group is used synchronously to determine the processing actions involved by the user. The action judgment intelligent agent group includes a user state judgment intelligent agent, a defeat detection intelligent agent, a conversion intelligent agent, and a knowledge base retrieval intelligent agent that are executed in parallel. The target processing action is determined based on the processing action and its corresponding priority.

3. The method according to claim 2, characterized in that, If the target processing action is a normal dialogue, then generating a dialogue response includes: If the target processing action is a normal dialogue, then based on the retrieval results of the knowledge base retrieval agent, as well as the user message and the historical dialogue, the dialogue agent generates the dialogue response. If the target processing action is to switch to manual processing or to fail, then the requirement scheme intent recognition task is canceled, and a corresponding engineering event instruction is generated and added to the engineering event list. The engineering event list is then fed back to the second server.

4. The method according to claim 1, characterized in that, If the dialogue response contains a signal to adjust the price, then based on the intent of the demand solution, an asynchronous price quotation instruction is generated and sent to the second server, including: If the dialogue response contains a target structured instruction, then it is determined that the dialogue response contains an adjustment quote signal; By utilizing product rule constraints, the intent of the proposed solution is mapped to compliant quotation request parameters; Based on the quoted price request parameters, an asynchronous quoted price instruction is generated and sent to the second server.

5. An asynchronous quotation method, characterized in that, Applications to the second server include: Based on user messages and historical conversations, a first conversation request is sent to a first server. The first server is used to identify the user's desired solution intent based on the user messages and historical conversations, and simultaneously determine the next target processing action for the user. If the target processing action is a normal conversation, a conversation response is generated. If the conversation response contains a price adjustment signal, an asynchronous price instruction is generated based on the desired solution intent and sent to the second server. The current session stage is then switched to a session stage waiting for price updates, and the current response ends. Based on the quotation request parameters carried in the asynchronous quotation instruction, a quotation request is sent to the quotation system; The system receives the quotation calculation result from the quotation system, uses the engineering event callback as the trigger type identifier, and sends a second dialogue request to the first server so that the first server can generate a quotation display response based on the second dialogue request. The system receives the price quote display reply sent by the first server and sends it back to the user.

6. The method according to claim 5, characterized in that, The method further includes: The system receives a number of standardized short messages sent by the first server. The first server is used to use a preset text splitting model to split the dialogue reply and / or the quotation display reply into a number of candidate short messages according to semantic boundaries. The system then performs standardization processing on the candidate short messages to achieve an optimal balance between message length and expression completeness. The standardized short messages are sent to the user sequentially via a serial delivery action.

7. An asynchronous quotation device, characterized in that, Applied to the first server, including: The first receiving unit is used to receive a first dialogue request sent by the second server, wherein the first dialogue request carries user messages and historical dialogues. The identification unit is used to identify the user's desired solution intent based on the user message and the historical dialogue, and simultaneously determine the next target processing action for the user; A generation unit is used to generate a dialogue response if the target processing action is a normal dialogue. The generation unit is further configured to, if the dialogue response contains a price adjustment signal, generate an asynchronous price instruction based on the intent of the demand solution and send it to the second server, switch the current session phase to a session phase waiting for price updates, and end the current response. The second server is configured to send a price request to the price system according to the price request parameters carried by the asynchronous price instruction; receive the price calculation result fed back by the price system; and send a second dialogue request to the first server using an engineering event callback as the trigger type identifier. The first sending unit is used to generate a quote display reply based on the second dialogue request and to send it back to the user through the second server.

8. An asynchronous quotation device, characterized in that, Applications to the second server include: The second sending unit is used to send a first dialogue request to the first server based on user messages and historical dialogues. The first server is used to identify the user's desired solution intent based on the user messages and historical dialogues, and simultaneously determine the next target processing action for the user. If the target processing action is a normal dialogue, a dialogue response is generated. If the dialogue response contains a price adjustment signal, an asynchronous price instruction is generated based on the desired solution intent and sent to the second server. The current session stage is then switched to a session stage waiting for price updates, and the current response ends. The second sending unit is further configured to send a quotation request to the quotation system according to the quotation request parameters carried by the asynchronous quotation instruction; The second sending unit is also used to receive the quotation calculation result fed back by the quotation system, and send a second dialogue request to the first server with the engineering event callback as the trigger type identifier, so that the first server can generate a quotation display reply based on the second dialogue request; The second receiving unit is used to receive the quotation display reply sent by the first server and to provide feedback to the user.

9. A storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the method of any one of claims 1 to 4 and / or 5 to 6.

10. An electronic device, comprising a storage medium, a processor, and a computer program stored on the storage medium and executable on the processor, characterized in that, When the processor executes the computer program, it implements the method of any one of claims 1 to 4 and / or 5 to 6.

Images