Multi-agent customer service processing method and system, electronic device and storage medium

Abstract

The application provides a multi-agent customer service processing method and system, an electronic device and a storage medium. The method is applied to the field of customer service and includes the following steps: obtaining a query text corresponding to a scenic spot service query request; performing semantic analysis on the query text based on a question disassembly agent to obtain a first query set corresponding to the query text; obtaining user portrait data corresponding to the scenic spot service query request and pre-constructed scenic spot knowledge structure data; performing reasoning on the first query set, the scenic spot knowledge structure data and the user portrait data based on a knowledge reasoning agent to obtain a second query set; querying a scenic spot knowledge base based on the first query set and the second query set to obtain a query result; checking the query result based on a reflection evaluation agent to obtain a check result; and outputting the query result in response to the check result meeting a preset check condition. In this way, the accuracy of multi-intent recognition and the reliability of output answers are improved.

Description

Technical Field

[0001] This application relates to the field of customer service, and more particularly to a multi-agent customer service processing method, system, electronic device, and storage medium. Background Technology

[0002] With the development of smart tourism and digital services, the demand for online consultation services among tourists in scenic areas is constantly increasing. Current intelligent customer service systems are purely response-oriented, lacking the ability to structurally analyze multi-intent questions and struggling to effectively reason using the relationships between knowledge. For example, most current customer service systems based on traditional machine learning are classification models, requiring large amounts of labeled data for model training and validation, and they struggle to accurately identify multiple user intents. Furthermore, long model iteration cycles make them prone to outputting erroneous information. Another example is that current customer service systems based on knowledge bases and large models exhibit randomness in handling multiple intents. They cannot build user profiles and can only provide passive question-and-answer services. They cannot sense users' specific needs, lack effective recommendations of relevant knowledge, and suffer from significant illusions.

[0003] Therefore, current multi-agent customer service processing methods suffer from low accuracy in multi-intent recognition and low reliability of output responses. Summary of the Invention

[0004] This application provides a multi-agent customer service processing method, system, electronic device, and storage medium to solve one or more problems existing in the related technology.

[0005] This application provides a multi-agent customer service processing method, the method comprising: acquiring query text corresponding to a scenic area service query request; performing semantic parsing on the query text based on a problem decomposition agent to obtain a first query set corresponding to the query text; the first query set including at least one query instruction corresponding to a first intent of the query text; acquiring user profile data and pre-constructed scenic area knowledge structure data corresponding to the scenic area service query request; performing reasoning on the first query set, the scenic area knowledge structure data, and the user profile data based on a knowledge reasoning agent to obtain a second query set; the second query set including at least one query instruction corresponding to a second intent of the query text; querying a scenic area knowledge base based on the first query set and the second query set to obtain query results; verifying the query results based on a reflection and evaluation agent to obtain verification results; and outputting the query results in response to the verification results satisfying preset verification conditions.

[0006] According to one embodiment of this application, the method further includes: obtaining historical query data corresponding to the scenic area service query request; the historical query data includes at least historical dialogue content, user behavior data, and user characteristic information; performing feature analysis on the historical query data based on a user profile intelligent agent to obtain user tag data; the user tag data includes at least user interest tags, user characteristic tags, and demand tags; after each round of dialogue, updating the user profile data based on the user tag data to obtain updated user profile data.

[0007] According to one embodiment of this application, the method further includes: in response to the initialization of the scenic area knowledge base or the knowledge update of the scenic area knowledge base, acquiring text data included in the scenic area knowledge base; performing entity extraction and relation extraction on the text data based on a knowledge modeling agent to obtain extraction results; and performing structured modeling on the extraction results based on the knowledge modeling agent to obtain the scenic area knowledge structure data.

[0008] According to one embodiment of this application, the knowledge-based reasoning agent infers the first query set, the scenic area knowledge structure data, and the user profile data to obtain a second query set, including: the scenic area knowledge structure data includes at least the relationships between scenic area entities; performing association reasoning based on the relationships and the query instructions corresponding to the first intent to obtain query directions associated with the first intent; filtering each query direction based on the user profile data to obtain a target query direction; performing user demand reasoning based on the target query direction to determine the user's second intent; and generating a query instruction corresponding to the second intent of the query text based on the target query direction and the second intent.

[0009] According to one embodiment of this application, the step of verifying the query results based on a reflective evaluation agent to obtain a verification result includes: obtaining the source identifier and knowledge text of each knowledge entry in the query results; performing semantic consistency verification between the query instructions included in the first query set and the knowledge text based on the reflective evaluation agent to obtain a first verification result; performing semantic consistency verification between the query instructions included in the second query set and the knowledge text based on the reflective evaluation agent to obtain a second verification result; and determining that the verification result meets a preset verification condition in response to the source identifier satisfying a set knowledge source condition, and the first verification result and the second verification result satisfying a set semantic consistency condition.

[0010] According to one embodiment of this application, the step of performing semantic parsing on the query text based on the problem decomposition agent to obtain a first query set corresponding to the query text includes: performing intent recognition on the query text based on the problem decomposition agent to obtain at least one first intent; using the problem decomposition agent, calling a large language model to generate query instructions corresponding to each first intent; and adding the query instructions corresponding to each first intent to the first query set.

[0011] According to one embodiment of this application, the method further includes: in response to the verification result indicating that the query result does not meet the preset verification conditions, regenerating the query instruction through a problem decomposition agent, and querying the scenic area knowledge base based on the regenerated query instruction.

[0012] This application also provides a multi-agent customer service processing system, comprising: a first acquisition module for acquiring query text corresponding to a scenic area service query request; a parsing module for performing semantic parsing on the query text based on a problem decomposition agent to obtain a first query set corresponding to the query text; the first query set includes at least one query instruction corresponding to a first intent of the query text; a second acquisition module for acquiring user profile data and pre-built scenic area knowledge structure data corresponding to the scenic area service query request; a reasoning module for reasoning on the first query set, the scenic area knowledge structure data, and the user profile data based on a knowledge reasoning agent to obtain a second query set; the second query set includes at least one query instruction corresponding to a second intent of the query text; a query module for querying a scenic area knowledge base based on the first query set and the second query set to obtain query results; a verification module for verifying the query results based on a reflection and evaluation agent to obtain verification results; and an output module for outputting the query results in response to the verification results meeting preset verification conditions.

[0013] This application also provides an electronic device, including: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to enable the at least one processor to perform the method of the above-described embodiments.

[0014] This application also provides a non-transitory computer-readable storage medium storing computer instructions for causing a computer to perform the methods described above.

[0015] This application provides a multi-agent customer service processing method, system, electronic device, and storage medium. The method includes: acquiring query text corresponding to a scenic area service query request; performing semantic parsing on the query text based on a problem decomposition agent to obtain a first query set corresponding to the query text; the first query set includes at least one query instruction corresponding to a first intent of the query text; acquiring user profile data and pre-constructed scenic area knowledge structure data corresponding to the scenic area service query request; performing reasoning on the first query set, the scenic area knowledge structure data, and the user profile data based on a knowledge reasoning agent to obtain a second query set; the second query set includes at least one query instruction corresponding to a second intent of the query text; querying a scenic area knowledge base based on the first query set and the second query set to obtain query results; verifying the query results based on a reflection and evaluation agent to obtain verification results; and outputting the query results in response to the verification results meeting preset verification conditions. This improves the accuracy of multi-intent recognition and the reliability of the output response.

[0016] It should be understood that the teachings of this application are not required to achieve all the beneficial effects described above, but rather that a specific technical solution can achieve a specific technical effect, and other embodiments of this application can also achieve beneficial effects not mentioned above. Attached Figure Description

[0017] The above and other objects, features, and advantages of exemplary embodiments of this application will become readily apparent from the following detailed description taken in conjunction with the accompanying drawings. Several embodiments of this application are illustrated in the drawings by way of example and not limitation, in which: In the accompanying drawings, the same or corresponding reference numerals indicate the same or corresponding parts.

[0018] Figure 1 A schematic diagram of the processing flow of the multi-agent customer service processing method provided in the embodiments of this application is shown; Figure 2 This application illustrates an application scenario of the multi-agent customer service processing method provided in this embodiment. Figure 1 ; Figure 3 This application illustrates an application scenario of the multi-agent customer service processing method provided in this embodiment. Figure 2 ; Figure 4 This illustration shows an optional schematic diagram of a multi-agent customer service processing system provided in an embodiment of this application; Figure 5 A schematic diagram of the composition structure of the electronic device provided in the embodiments of this application is shown. Detailed Implementation

[0019] To make the objectives, features, and advantages of this application more apparent and understandable, the technical solutions in the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0020] In the following description, references are made to “some embodiments,” which describe a subset of all possible embodiments. However, it is understood that “some embodiments” may be the same subset or different subsets of all possible embodiments and may be combined with each other without conflict.

[0021] In the following description, the terms "first" and "second" are used merely to distinguish similar objects and do not represent a specific ordering of objects. It is understood that "first" and "second" may be interchanged in a specific order or sequence where permitted, so that the embodiments of this application described herein can be implemented in an order other than that illustrated or described herein.

[0022] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs. The terminology used herein is for the purpose of describing embodiments of this application only and is not intended to limit this application.

[0023] The processing flow of the multi-agent customer service processing method provided in the embodiments of this application is described below. See also Figure 1 , Figure 1 This is a schematic diagram of the processing flow of the multi-agent customer service processing method provided in the embodiments of this application, which will be combined with Figure 1 Steps S101-S107 are explained below.

[0024] Step S101: Obtain the query text corresponding to the scenic area service query request.

[0025] In some embodiments, a scenic area service query request may include a query request entered by a user through an interactive device such as a client. The query text may include questions entered by the user regarding scenic area services. For example, "When do the magnolias bloom?", "Are guide dogs allowed in the scenic area?", and "How much are the tickets?".

[0026] Step S102: Based on the problem decomposition agent, perform semantic parsing on the query text to obtain the first query set corresponding to the query text; the first query set includes at least one query instruction corresponding to the first intent of the query text.

[0027] In some embodiments, the problem-decomposition agent can be software including a memory storage module and a tool invocation module, capable of calling the large language model multiple times and executing corresponding code when processing a problem. The problem-decomposition agent can be used to: perform semantic parsing on user-input query text, identify the core and potential intents in the user's question, and decompose complex questions into structured query instructions. Semantic parsing can be used to: convert unstructured query text into a structured text representation. The problem-decomposition agent can parse and obtain at least one query instruction corresponding to a first intent. The first intent can be the query purpose directly expressed by the user through the query text. The query instruction can be used to retrieve corresponding knowledge entries from the scenic area knowledge base.

[0028] Step S103: Obtain user profile data and pre-built scenic area knowledge structure data corresponding to the scenic area service query request.

[0029] In some embodiments, the user profiling agent can generate user tag data based on historical dialogue content, user behavior data, and user feature information, and then determine the corresponding user profile data based on the user tag data. The knowledge modeling agent can extract entities and relations from the scenic area knowledge base, obtain extraction results, and then construct or update the scenic area knowledge structure data based on the extraction results. The scenic area knowledge structure data can serve as a query index for the scenic area knowledge base. The knowledge reasoning agent and the user profiling agent can be software containing a memory storage module and a tool invocation module, which can call the large language model multiple times and execute the corresponding code when processing problems. The knowledge reasoning agent can be used to fuse the first query set, the scenic area knowledge structure data, and the user profile data to perform potential demand reasoning and generate extended query instructions. The user profile agent can be used to analyze user behavior during multi-turn dialogues, construct user interest tags, user feature tags, and demand tags, generate user profile data, and continuously update it during the dialogue process.

[0030] Step S104: Based on the knowledge reasoning agent, reason about the first query set, scenic area knowledge structure data and user profile data to obtain the second query set; the second query set includes a query instruction corresponding to the second intent of at least one query text.

[0031] In some embodiments, the knowledge-reasoning agent can infer and generate at least one query instruction corresponding to a second intent. The second intent may be a potential user need intent derived from the association between user profiles and knowledge structures.

[0032] Step S105: Based on the first query set and the second query set, query the scenic area knowledge base to obtain the query results.

[0033] In some embodiments, the scenic area knowledge base may be a local database storing text data and knowledge entries related to the scenic area. Each knowledge entry includes a source identifier and knowledge text. Query results may include: knowledge entries retrieved from the scenic area knowledge base based on query instructions in a first query set and a second query set.

[0034] Step S106: Verify the query results based on the reflective evaluation agent to obtain the verification results.

[0035] Step S107: In response to the verification result meeting the preset verification conditions, output the query result.

[0036] In some embodiments, the reflective evaluation agent can be software including a memory storage module and a tool invocation module, capable of repeatedly invoking the large language model and executing corresponding code when processing problems. The reflective evaluation agent can be used to perform source verification and semantic consistency verification on query results. Source verification and semantic consistency verification can identify and eliminate erroneous and illusory knowledge that does not meet preset conditions. Preset verification conditions may include: pre-defined rules for the validity of knowledge sources and semantic matching thresholds.

[0037] As an example, a user sends a scenic area service query request through the client. The query text is "When do magnolias bloom?". The problem decomposition agent performs semantic analysis on the query text, identifying the user's first intent as "querying the magnolia blooming time" and generating a corresponding first query set. The user profile data, derived from the user profile agent, is "the user likes magnolias." Simultaneously, the scenic area knowledge structure data, pre-constructed by the knowledge modeling agent, is also retrieved. The knowledge reasoning agent, based on the first query set, the scenic area knowledge structure data, and the user profile data, infers the user's second intent, including "1. Which park has the most magnolias? 2. What are the opening and closing times of this park? 3. Are there any restrictions or requirements in this park?", thus generating a second query set corresponding to the second intent. The scenic area knowledge base is queried based on the first and second query sets, retrieving the corresponding knowledge entries and obtaining the query results. The reflection and evaluation agent performs semantic consistency checks between the query instructions in the first and second query sets and the corresponding knowledge text of each knowledge entry, determining whether the source identifier of each knowledge entry meets the knowledge source conditions, and obtaining the corresponding verification results. In response to the verification results meeting the set verification conditions, the query results are output to the client. The specific search results could be: "Hello, magnolias bloom from late March to mid-April. We recommend visiting Hall A and Hall B to see them. Hall A and Hall B are attractions within the park, open from 8:00 AM to 6:00 PM from April to October, and from 8:30 AM to 5:00 PM from November to March. Don't miss the opening hours! Additionally, there are magnolias to see at Gate C; if convenient, we suggest entering the park through Gate C."

[0038] The technical effects of the method in this application embodiment include: By constructing a multi-agent collaborative architecture comprising a problem-decomposition agent, a knowledge reasoning agent, a reflection and evaluation agent, a user profiling agent, and a knowledge modeling agent, the problem-decomposition agent performs semantic parsing of user input, enabling it to identify and decompose multiple intents within complex problems and generate structured query instructions, thus improving the ability to analyze complex problems in a structured manner. The knowledge modeling agent extracts entities and relationships from knowledge base information, constructing a knowledge network structure containing entity relationships, enabling effective reasoning using the connections between knowledge points and overcoming the difficulty of utilizing discrete storage of knowledge base information. The user profiling agent analyzes user behavior and dynamically constructs and updates user profiles during multi-turn dialogues, continuously building user profiles and achieving a sustained understanding of users. Furthermore, the knowledge reasoning agent integrates user questions, user profiles, and knowledge structures to infer potential needs, proactively generating extended query instructions, thus proactively predicting user needs and providing relevant information. Simultaneously, the reflection and evaluation agent verifies the source of search results and evaluates semantic consistency, effectively reducing the risk of illusion in large language models and improving the reliability and accuracy of the output answers.

[0039] In some embodiments, the semantic parsing of the query text by the problem decomposition agent in step S102 to obtain the first query set corresponding to the query text may include: performing intent recognition on the query text based on the problem decomposition agent to obtain at least one first intent; calling a large language model based on the problem decomposition agent to generate query instructions corresponding to each first intent; and adding the query instructions corresponding to each first intent to the first query set.

[0040] In this embodiment, semantic parsing may include: word segmentation, intent recognition, and structuring of the query text. Intent recognition involves semantic analysis of the query text to determine the user's direct query purpose expressed through the query text. The large language model can be a pre-trained natural language processing model, which can perform structuring processing on the identified intent to generate structured query instructions. The problem decomposition agent, knowledge reasoning agent, user profiling agent, knowledge modeling agent, and reflection and evaluation agent can call the large language model multiple times during the processing and coordinate tasks through a unified scheduling mechanism.

[0041] As an example, the user's query text could be "Can guide dogs enter the scenic area? How to buy tickets?" The problem decomposition agent is invoked to perform intent recognition on the query text, revealing two primary intents: "Inquire about guide dog admission policies" and "Inquire about ticket purchase methods." The problem decomposition agent then uses a large language model to structure the two primary intents, generating two corresponding query instructions. These two query instructions are then added to the first query set.

[0042] In some embodiments, the method further includes: obtaining historical query data corresponding to scenic area service query requests; performing feature analysis on the historical query data based on the user profile intelligent agent to obtain user tag data; and updating the user profile data based on the user tag data after each round of dialogue to obtain updated user profile data.

[0043] In this embodiment, historical query data may include historical dialogue content, user behavior data, and user characteristic information. User tag data may include user interest tags, user characteristic tags, and demand tags. Historical dialogue content may include: text content generated during multiple rounds of dialogue between the user and the customer service system. User behavior data may include: user operation records and feedback records on the interactive interface. User characteristic information may include: user's age, gender, region, and identity identifier. The user profile agent can call a large language model to perform feature analysis on historical dialogue content to obtain user tag data. User interest tags may be classification identifiers representing the user's areas of interest. User characteristic tags may be classification identifiers representing the user's attribute characteristics. Demand tags may be classification identifiers representing the user's service needs. After each round of dialogue, the user profile data is incrementally updated based on the latest user tag data to obtain updated user profile data.

[0044] As an example, the system retrieves historical query data corresponding to the current scenic area service query request. Specifically, this historical query data includes the user's past three days of dialogue content, user behavior data, and user characteristic information. The user profiling agent uses a large language model to perform feature analysis on the historical query data, generating user tag data including "fond of flowers," "photography enthusiast," and "elderly tourist." After the current round of dialogue ends, the user profiling agent incrementally updates the original user profile data based on the user tag data, obtaining the updated user profile data.

[0045] In some embodiments, the method further includes: in response to the initialization of the scenic area knowledge base or the knowledge update of the scenic area knowledge base, acquiring the text data included in the scenic area knowledge base; performing entity extraction and relation extraction on the text data based on the knowledge modeling agent to obtain extraction results; and performing structured modeling on the extraction results based on the knowledge modeling agent to obtain scenic area knowledge structure data.

[0046] In this embodiment, the initialization of the scenic area knowledge base can be the initial deployment of the scenic area knowledge base. Knowledge updates can involve adding, modifying, or deleting data in the scenic area knowledge base. Text data can be used as input for entity extraction and relation extraction. Entity extraction can be used to construct nodes in the knowledge structure data. Relation extraction can be used to construct edges in the knowledge structure data. The extraction results can include: the set of entities and the corresponding set of relationships obtained after entity extraction and relation extraction. Entities can serve as nodes in the knowledge graph, and relationships can serve as edges in the knowledge graph. The scenic area knowledge structure data can be a knowledge network structure, specifically, a knowledge graph. Structured modeling can involve constructing a knowledge graph from the extracted node data and edge data. The knowledge graph can serve as an index for the knowledge stored in the scenic area knowledge base.

[0047] As an example, when a scenic area's knowledge base adds content about its flowers, the first step is to retrieve the text data already included in the knowledge base, including the newly added flower content. Specifically, this could be "Magnolia flowers are located in Area A of the park." The knowledge modeling agent extracts entities from the text data, obtaining entities such as "Magnolia flowers" and "located in Area A of the park," and then extracts relationships, revealing a "distribution location" relationship between "Magnolia flowers" and "located in Area A of the park." The knowledge modeling agent then performs structured modeling on these entities and relationships, using entities as nodes in a knowledge graph and relationships as edges, resulting in the scenic area's knowledge structure data.

[0048] In some embodiments, the knowledge-based reasoning agent in step S104 reasoning on the first query set, scenic area knowledge structure data, and user profile data to obtain the second query set may include: the scenic area knowledge structure data including at least the relationships between scenic area entities; performing association reasoning based on the relationships and query instructions corresponding to the first intent to obtain query directions associated with the first intent; filtering each query direction based on the user profile data to obtain the target query direction; performing user demand reasoning based on the target query direction to determine the user's second intent; and generating query instructions corresponding to the second intent of the query text based on the target query direction and the second intent.

[0049] In this embodiment, the scenic area knowledge structure data specifically refers to structured knowledge network data of scenic area entities and their corresponding semantic associations. Association relationships characterize the semantic connections between different scenic area entities within the scenic area knowledge structure data. Scenic area entities may include: attractions, plants, buildings, service facilities, opening hours, ticketing policies, etc. This application embodiment does not limit specific scenic area entities. Association reasoning can determine associated scenic area entities and knowledge structures based on known first intentions, using association relationships. Association reasoning can determine candidate query directions that are associated with the first intention. Query directions are specifically extended topics related to the first intention. Filtering can evaluate the relevance of candidate query directions based on user interest tags, user feature tags, and demand tags in user profile data to determine the target query direction that meets the user's potential needs from multiple candidate query directions. User demand reasoning can infer the user's potential demand intentions based on the target query direction and user profile data. The knowledge reasoning agent can generate at least one query instruction corresponding to at least one second intention based on the determined target query direction and second intention.

[0050] As an example, the first query set may include the query instruction corresponding to the first intent "query the magnolia blooming time". The first query set, pre-constructed scenic area knowledge structure data, and user profile data are input into the knowledge reasoning agent. The knowledge reasoning agent parses the scenic area knowledge structure data to obtain the association relationships between scenic area entities such as "magnolia" and "viewing location", and performs association reasoning based on these relationships and the query instruction corresponding to the first intent, resulting in three candidate query directions: "magnolia distribution area", "flower type", and "spring blooming time of the area". Based on the user interest tag "magnolia enthusiast" in the user profile data, the above three query directions are filtered, and it is determined that "magnolia distribution area" and "spring blooming time of the area" are more consistent with the user's interest tags. Therefore, the target query direction is determined to be "magnolia distribution area" and "spring blooming time of the area". Next, the knowledge reasoning agent performs user demand reasoning based on the target query direction, determining that the user's second intent is "query the best magnolia viewing area and the opening time of that area during the flowering season". Based on the target query direction and the second intent, a corresponding structured query instruction is generated and added to the second query set.

[0051] In some embodiments, step S106, which verifies the query results based on a reflective evaluation agent to obtain a verification result, may include: obtaining the source identifier and knowledge text of each knowledge entry in the query results; performing semantic consistency verification between the query instructions included in the first query set and the knowledge text based on the reflective evaluation agent to obtain a first verification result; performing semantic consistency verification between the query instructions included in the second query set and the knowledge text based on the reflective evaluation agent to obtain a second verification result; and determining that the verification result meets the preset verification conditions in response to the source identifier satisfying the set knowledge source conditions and the first and second verification results satisfying the set semantic consistency conditions.

[0052] In this embodiment, a knowledge entry may include: specific knowledge content retrieved from a scenic area knowledge base and its corresponding source identifier. The source identifier can represent the original source of the knowledge entry in the scenic area knowledge base. The knowledge text can be the specific text content used to answer a query instruction. Semantic consistency verification can be performed by matching the question content corresponding to the query instruction with the knowledge text. Semantic consistency verification can be used to verify whether the knowledge text can answer the question content corresponding to the query instruction. A first verification result can represent whether the knowledge text can answer the question content corresponding to the query instructions included in a first query set. A second verification result can represent whether the knowledge text can answer the question content corresponding to the query instructions included in a second query set. The first verification result and the second verification result can specifically be the semantic matching degree between the query instruction and the knowledge text. The knowledge source condition may include: a pre-set credible source identifier. The semantic consistency condition may include: a pre-set semantic matching degree threshold between the query instruction and the knowledge text.

[0053] As an example, the query results contain three knowledge entries: Knowledge entry A states that the magnolia blooming period is from early March to mid-April each year (source identifier 01); Knowledge entry B states that Park A is the main viewing area for magnolias (source identifier 02); and Knowledge entry C states that Park A's spring opening hours are from 7:00 AM to 6:00 PM (source identifier 03). The reflective evaluation agent performs semantic consistency checks on the query command corresponding to the first intent "querying magnolia blooming time" in the first query set with knowledge entry A, confirming that the first check result represents the set semantic consistency condition. Simultaneously, it performs semantic consistency checks on the query command corresponding to the second intent "querying the best viewing park and opening time" in the second query set with knowledge entries B and C respectively, confirming that the second check result represents the set semantic consistency condition. Since the source identifiers 01, 02, and 03 of the three knowledge entries are all reliable, they satisfy the set knowledge source conditions. Finally, the reflective evaluation agent determines that the check results satisfy the preset check conditions.

[0054] In some embodiments, the method further includes: in response to the verification result indicating that the query result does not meet the preset verification conditions, regenerating the query instruction through the problem decomposition agent, and querying the scenic area knowledge base based on the regenerated query instruction.

[0055] As an example, after the system queries the scenic area knowledge base in parallel based on the first query set and the second query set, it obtains the query results. When the intelligent agent verifies the query results, it determines that the knowledge entry "plum blossom viewing location" is inconsistent with the first intention "query the magnolia blossom opening time", or determines that the source of the knowledge entry is unreliable, and determines that the query results do not meet the preset verification conditions.

[0056] In response to the verification result, the erroneous knowledge entry for "plum blossom viewing location" is deleted. The problem decomposition agent is then invoked to re-parse the semantically inconsistent information as input, generating a new query instruction: "Last query was incorrect; query again for magnolia blossom blooming time." Based on the regenerated query instruction, the scenic area knowledge base is queried again, retrieving the correct knowledge entry for the magnolia blossom blooming time.

[0057] refer to Figure 2 Application scenarios of the multi-agent customer service processing method provided in this application embodiment Figure 1 It is applied to a proactive intelligent customer service system based on local knowledge base and multi-agent collaborative reasoning.

[0058] The proactive intelligent customer service system includes a security layer, a processing layer, a user profile intelligent agent, a knowledge modeling intelligent agent, and a local knowledge base.

[0059] The proactive intelligent customer service system adopts a multi-agent collaborative architecture (Agentic AI Architecture). In this architecture, different agents are responsible for different tasks and work together through a unified scheduling mechanism to achieve efficient processing of complex consultation tasks.

[0060] When a user asks "When do magnolias bloom?", the system can not only answer with information about the blooming period, but also proactively infer relevant information that the user might need, such as the main areas where magnolias are distributed, the best viewing areas, and the opening time of those areas during the magnolia blooming season, thus providing the user with more complete service information.

[0061] When a user asks, "Can guide dogs enter the scenic area?", the system can further infer related needs, such as ticket policies for blind tourists, the location of accessible facilities, and suitable accessible tour routes, thereby providing users with more comprehensive services and humanistic care.

[0062] The system's intelligent agents include: The problem decomposition agent is used to perform semantic parsing on user-input questions, identify the intent in the user's questions, and decompose complex questions into multiple structured query instructions.

[0063] Knowledge modeling agents are used to perform structured modeling of information in a local knowledge base. By extracting knowledge entities and constructing relationships between entities, a knowledge network structure is formed, enabling the system to perform reasoning using the relationships between different knowledge.

[0064] User profiling intelligence agents are used to analyze user behavior and build dynamic user profiles, such as user interests, user characteristics, and potential needs, during multi-turn conversations, and continuously update them during the conversation.

[0065] The knowledge reasoning agent is used to combine user questions, user profile data, and knowledge structures to reason about potential intentions and generate extended query instructions, thereby enabling proactive services.

[0066] A reflective evaluation agent is used to assess the credibility of generated answers, reducing the risk of illusion in large language models by analyzing knowledge sources and semantic consistency.

[0067] The security layer is used to filter user input and generated system responses, detecting harmful or toxic input or output to determine whether the query text or system response contains sensitive or illegal content.

[0068] refer to Figure 3 Application scenarios of the multi-agent customer service processing method provided in this application embodiment Figure 2 It is applied to the customer service process of a proactive intelligent customer service system based on local knowledge base and multi-agent collaborative reasoning.

[0069] Users input the query text "When do magnolias bloom?" through the client. The system first calls the question decomposition agent to perform semantic analysis on the query text, obtaining the user's first intent as "querying when magnolias bloom," and generates a corresponding first query set. The knowledge reasoning agent receives the first query set and performs association reasoning based on the user profile data "the user likes magnolias" and the scenic area's knowledge structure data to determine that the user has a potential need to visit the park. It then predicts that the user may also want to obtain information such as "which park has the most magnolias," "the opening and closing times of the park," and "whether the park has any restrictions or requirements," and generates a corresponding second query set. The system then queries the scenic area's knowledge base in parallel based on the first and second query sets to obtain query results including flowering period, viewing locations, opening hours, and ticket policies. The evaluation agent determined that the knowledge entries in the query results could answer the user's original question and the park's opening and closing times. However, it was found that the information "a separate ticket is required to enter Hall A" was not verified in the fee knowledge base and was identified as potentially incorrect, so it was removed. Finally, the system generated a response based on the verified knowledge entries, outputting to the user's client: "Hello, magnolias bloom from late March to mid-April. We recommend visiting Hall A and Hall B. Hall A and Hall B are park attractions, open from 8:00 AM to 6:00 PM from April to October, and from 8:30 AM to 5:00 PM from November to March. Don't miss the opening times! Additionally, there are magnolias to see at Gate C; if convenient, we recommend entering the park through Gate C." The following continues to describe the exemplary structure of the software modules included in the multi-agent customer service processing system 90 provided in the embodiments of this application. In some embodiments, such as Figure 4 As shown, the multi-agent customer service processing system 90 may include: The first acquisition module 901 is used to acquire the query text corresponding to the scenic area service query request; Parsing module 902 is used to perform semantic parsing on the query text based on the problem decomposition agent to obtain a first query set corresponding to the query text; the first query set includes at least one query instruction corresponding to the first intent of the query text; The second acquisition module 903 is used to acquire user profile data and pre-built scenic area knowledge structure data corresponding to the scenic area service query request; The reasoning module 904 is used to reason about the first query set, scenic area knowledge structure data and user profile data based on the knowledge reasoning agent to obtain the second query set; the second query set includes a query instruction corresponding to the second intent of at least one query text. The query module 905 is used to query the scenic area knowledge base based on the first query set and the second query set to obtain the query results. The verification module 906 is used to verify the query results based on the reflective evaluation agent and obtain the verification results. Output module 907 is used to output query results in response to the verification result meeting the preset verification conditions.

[0070] In some embodiments, the multi-agent customer service processing system 90 further includes an update module, which is used to: obtain historical query data corresponding to scenic area service query requests; the historical query data includes at least historical dialogue content, user behavior data, and user characteristic information; perform feature analysis on the historical query data based on user profile agents to obtain user tag data; the user tag data includes at least user interest tags, user characteristic tags, and demand tags; and update the user profile data based on the user tag data after each round of dialogue to obtain updated user profile data.

[0071] In some embodiments, the multi-agent customer service processing system 90 further includes a construction module, which is configured to: in response to the initialization of the scenic area knowledge base or the knowledge update of the scenic area knowledge base, acquire text data included in the scenic area knowledge base; perform entity extraction and relation extraction on the text data based on the knowledge modeling agent to obtain extraction results; and perform structured modeling on the extraction results based on the knowledge modeling agent to obtain the scenic area knowledge structure data.

[0072] In some embodiments, the reasoning module 904 is configured to: the scenic area knowledge structure data include at least the relationships between scenic area entities; perform association reasoning based on the relationships and the first intent to obtain a query direction associated with the first intent; filter each query direction based on the user profile data to obtain a target query direction; perform user demand reasoning based on the target query direction to determine the user's second intent; and generate a query instruction corresponding to the second intent of the query text based on the target query direction and the second intent.

[0073] In some embodiments, the verification module 906 is configured to: obtain the source identifier and knowledge text of each knowledge entry in the query results; perform semantic consistency verification between the query instructions included in the first query set and the knowledge text based on the reflective evaluation agent to obtain a first verification result; perform semantic consistency verification between the query instructions included in the second query set and the knowledge text based on the reflective evaluation agent to obtain a second verification result; and determine that the verification result meets the preset verification conditions in response to the source identifier satisfying the set knowledge source conditions and the first verification result and the second verification result satisfying the set semantic consistency conditions.

[0074] In some embodiments, the parsing module 902 is configured to: perform intent recognition on the query text based on the problem decomposition agent to obtain at least one first intent; based on the problem decomposition agent, call a large language model to generate query instructions corresponding to each first intent; and add the query instructions corresponding to each first intent to the first query set.

[0075] In some embodiments, the output module 907 is further configured to: in response to the verification result indicating that the query result does not meet the preset verification conditions, regenerate the query instruction through the problem decomposition agent, and query the scenic area knowledge base based on the regenerated query instruction.

[0076] It should be noted that the description of the apparatus in this application embodiment is similar to the description of the method embodiment above, and has similar beneficial effects as the method embodiment, therefore it will not be repeated. For any technical details not covered in the multi-agent customer service processing system provided in this application embodiment, please refer to... Figures 1 to 3 The meaning is understood in accordance with the description of any of the accompanying drawings.

[0077] According to embodiments of this application, this application also provides an electronic device and a non-transitory computer-readable storage medium.

[0078] Figure 5 A schematic block diagram of an example electronic device 800 that can be used to implement embodiments of this application is shown. The electronic device is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device may also represent various forms of mobile devices, such as personal digital processors, cellular phones, smartphones, wearable devices, and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely illustrative and are not intended to limit the implementation of the application described and / or claimed herein.

[0079] like Figure 5 As shown, the electronic device 800 includes a computing unit 801, which can perform various appropriate actions and processes based on a computer program stored in a read-only memory (ROM) 802 or a computer program loaded from a storage unit 808 into a random access memory (RAM) 803. The RAM 803 may also store various programs and data required for the operation of the electronic device 800. The computing unit 801, ROM 802, and RAM 803 are interconnected via a bus 804. An input / output (I / O) interface 805 is also connected to the bus 804.

[0080] Multiple components in electronic device 800 are connected to I / O interface 805, including: input unit 806, such as keyboard, mouse, etc.; output unit 807, such as various types of displays, speakers, etc.; storage unit 808, such as disk, optical disk, etc.; and communication unit 809, such as network card, modem, wireless transceiver, etc. Communication unit 809 allows electronic device 800 to exchange information / data with other devices through computer networks such as the Internet and / or various telecommunications networks.

[0081] The computing unit 801 can be a variety of general-purpose and / or special-purpose processing components with processing and computing capabilities. Some examples of the computing unit 801 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various special-purpose artificial intelligence (AI) computing chips, various computing units running machine learning model algorithms, a digital signal processor (DSP), and any suitable processor, controller, microcontroller, etc. The computing unit 801 performs the various methods and processes described above, such as multi-agent client processing methods. For example, in some embodiments, the multi-agent client processing method can be implemented as a computer software program tangibly contained in a machine-readable medium, such as storage unit 808. In some embodiments, part or all of the computer program can be loaded and / or installed on the electronic device 800 via ROM 802 and / or communication unit 809. When the computer program is loaded into RAM 803 and executed by the computing unit 801, one or more steps of the multi-agent client processing method described above can be performed. Alternatively, in other embodiments, the computing unit 801 can be configured to perform multi-agent client processing methods by any other suitable means (e.g., by means of firmware).

[0082] Various embodiments of the systems and techniques described above herein can be implemented in digital electronic circuit systems, integrated circuit systems, field-programmable gate arrays (FPGAs), application-specific integrated circuits (ASICs), application-specific standard products (ASSPs), systems-on-a-chip (SoCs), payload-programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various embodiments may include implementations in one or more computer programs that can be executed and / or interpreted on a programmable system including at least one programmable processor, which may be a dedicated or general-purpose programmable processor, capable of receiving data and instructions from a storage system, at least one input device, and at least one output device, and transmitting data and instructions to the storage system, the at least one input device, and the at least one output device.

[0083] The program code used to implement the methods of this application may be written in any combination of one or more programming languages. This program code may be provided to a processor or controller of a general-purpose computer, a special-purpose computer, or other programmable multi-agent client processing system, such that when executed by the processor or controller, the functions / operations specified in the flowcharts and / or block diagrams are implemented. The program code may be executed entirely on a machine, partially on a machine, as a standalone software package partially on a machine and partially on a remote machine, or entirely on a remote machine or server.

[0084] In the context of this application, a machine-readable medium can be a tangible medium that may contain or store a program for use by or in conjunction with an instruction execution system, apparatus, or device. A machine-readable medium can be a machine-readable signal medium or a machine-readable storage medium. Machine-readable media can be, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination of the foregoing. More specific examples of machine-readable storage media include electrical connections based on one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fibers, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination of the foregoing.

[0085] To provide interaction with a user, the systems and techniques described herein can be implemented on a computer having: a display device for displaying information to the user (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor); and a keyboard and pointing device (e.g., a mouse or trackball) through which the user provides input to the computer. Other types of devices can also be used to provide interaction with the user; for example, feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form (including sound input, voice input, or tactile input).

[0086] The systems and technologies described herein can be implemented in computing systems that include backend components (e.g., as a data server), or computing systems that include middleware components (e.g., an application server), or computing systems that include frontend components (e.g., a user computer with a graphical user interface or web browser through which a user can interact with implementations of the systems and technologies described herein), or any combination of such backend, middleware, or frontend components. The components of the system can be interconnected via digital data communication of any form or medium (e.g., a communication network). Examples of communication networks include local area networks (LANs), wide area networks (WANs), and the Internet.

[0087] Computer systems can include clients and servers. Clients and servers are generally located far apart and typically interact via communication networks. Client-server relationships are created by computer programs running on the respective computers and having a client-server relationship with each other. Servers can be cloud servers, servers in distributed systems, or servers incorporating blockchain technology.

[0088] It should be understood that the various forms of processes shown above can be used to rearrange, add, or delete steps. For example, the steps described in this application can be executed in parallel, sequentially, or in different orders, as long as the desired result of the technical solution disclosed in this application can be achieved, and this is not limited herein.

[0089] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A multi-agent customer service processing method, characterized in that, The method includes: Retrieve the query text corresponding to the scenic area service query request; Based on the problem decomposition intelligent agent, the query text is semantically parsed to obtain a first query set corresponding to the query text; the first query set includes at least one query instruction corresponding to the first intent of the query text. Obtain the user profile data and pre-constructed scenic area knowledge structure data corresponding to the scenic area service query request; Based on the knowledge reasoning agent, a second query set is obtained by reasoning about the first query set, the scenic area knowledge structure data, and the user profile data; the second query set includes at least one query instruction corresponding to the second intent of the query text. Based on the first query set and the second query set, the scenic area knowledge base is queried to obtain the query results; The query results are verified by a reflective evaluation agent to obtain the verification results. In response to the verification result satisfying the preset verification conditions, the query result is output.

2. The method according to claim 1, characterized in that, The method further includes: Obtain historical query data corresponding to scenic area service query requests; the historical query data includes at least historical dialogue content, user behavior data, and user characteristic information. Based on the user profile intelligent agent, feature analysis is performed on the historical query data to obtain user tag data; the user tag data includes at least user interest tags, user feature tags, and demand tags. After each round of dialogue, the user profile data is updated based on the user tag data to obtain the updated user profile data.

3. The method according to claim 1, characterized in that, The method further includes: In response to the initialization of the scenic area knowledge base or the knowledge update of the scenic area knowledge base, the text data included in the scenic area knowledge base is obtained. Based on the knowledge modeling intelligent agent, entity extraction and relation extraction are performed on text data to obtain the extraction results; The knowledge modeling agent performs structured modeling on the extraction results to obtain the knowledge structure data of the scenic area.

4. The method according to claim 1, characterized in that, The knowledge-based reasoning agent performs reasoning on the first query set, the scenic area knowledge structure data, and the user profile data to obtain a second query set, including: The scenic area knowledge structure data includes at least the relationships between scenic area entities; Based on the association relationship and the query instruction corresponding to the first intent, association reasoning is performed to obtain the query direction associated with the first intent; Based on the user profile data, each of the query directions is filtered to obtain the target query direction; Based on the target query direction, user needs are inferred to determine the user's secondary intent; Based on the target query direction and the second intent, a query instruction corresponding to the second intent of the query text is generated.

5. The method according to claim 1, characterized in that, The validation of the query results by the reflective evaluation agent yields validation results, including: Obtain the source identifier and knowledge text of each knowledge entry in the query results; Based on the reflective evaluation agent, the query instructions included in the first query set are semantically consistent with the knowledge text to obtain the first verification result; Based on the reflective evaluation agent, the query instructions included in the second query set are semantically consistent with the knowledge text to obtain the second verification result; In response to the source identifier satisfying the set knowledge source condition, and the first verification result and the second verification result satisfying the set semantic consistency condition, it is determined that the verification result satisfies the preset verification condition.

6. The method according to claim 1, characterized in that, The problem-decomposition-based intelligent agent performs semantic parsing on the query text to obtain a first query set corresponding to the query text, including: Based on the problem decomposition intelligent agent, the query text is subjected to intent recognition to obtain at least one first intent; Based on the problem decomposition intelligent agent, the large language model is called to generate query instructions corresponding to each first intent; Add the query instructions corresponding to each first intent to the first query set.

7. The method according to claim 1, characterized in that, The method further includes: In response to the verification result indicating that the query result does not meet the preset verification conditions, the query instruction is regenerated by the problem decomposition agent, and the scenic area knowledge base is queried based on the regenerated query instruction.

8. A multi-agent customer service processing system, characterized in that, include: The first acquisition module is used to acquire the query text corresponding to the scenic area service query request; The parsing module is used to perform semantic parsing on the query text based on the problem decomposition agent to obtain a first query set corresponding to the query text; the first query set includes at least one query instruction corresponding to the first intent of the query text; The second acquisition module is used to acquire user profile data and pre-built scenic area knowledge structure data corresponding to the scenic area service query request. The reasoning module is used to reason about the first query set, the scenic area knowledge structure data, and the user profile data based on the knowledge reasoning agent to obtain a second query set; the second query set includes at least one query instruction corresponding to the second intent of the query text; The query module is used to query the scenic area knowledge base based on the first query set and the second query set to obtain query results; The verification module is used to verify the query results based on the reflective evaluation agent and obtain the verification results. The output module is used to output the query result in response to the verification result meeting the preset verification conditions.

9. An electronic device, characterized in that, include: At least one processor; And a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to enable the at least one processor to perform the method of any one of claims 1-7.

10. A non-transitory computer-readable storage medium storing computer instructions, characterized in that, The computer instructions are used to cause the computer to perform the method according to any one of claims 1-7.

Images