Information interaction method, device and equipment based on multi-agent collaboration and medium

Abstract

The invention relates to the technical field of voice semantics, and discloses an information interaction method and device based on multi-agent collaboration, equipment and a medium, and the method comprises the steps: generating a target workflow through a large language model and a current user demand; determining a target AP I according to the target workflow, and generating initial interaction information; and generating current reply evaluation information based on the initial interaction information and the criticizing agent, and determining the initial interaction information as target interaction information under the condition that the current matching degree is greater than or equal to a matching degree threshold value. Through the above mode, the user demand is split through the routing agent and the built-in large language model, the target workflow is generated, and the method adapts to complex and changeable task scenes. According to the method, the target AP I is determined through the induction agent, and the initial interaction information is generated, so that the efficiency and accuracy of calling the AP I are improved, and the performance of an automatic business process system for processing complex interaction information in the fields of financial science and technology, medical treatment, health, old-age care and the like is improved.

Description

Technical Field

[0001] The present application relates to the field of speech semantics technology, and in particular to an information interaction method, apparatus, device and medium based on multi-agent collaboration. Background Art

[0002] With the rapid development of artificial intelligence technology, automated business processes play a vital role in various fields today. Some small intelligent agent systems based on a single LLM (Large Language Model) have been introduced into different business fields. In the financial and fintech business field, especially for risk assessment, customer service, and insurance claims, traditional processes often rely on manual judgment and operation, which is not only inefficient but also prone to errors. The current introduction of a single automated business process is often limited by its reasoning ability and can only handle relatively simple tasks within a single field. When faced with more complex cross-domain tasks, such as the multi-department collaboration and integrated analysis of multiple data involved in insurance claims, errors often occur or effective solutions cannot be provided.

[0003] Similar to the fintech sector, the healthcare sector also faces numerous challenges in automating processes. Traditional information systems often rely on fixed processes and rules for patient diagnosis, treatment recommendation, and medical resource allocation, making it difficult to adapt to individual patient differences and complex and changing medical scenarios. With the introduction of artificial intelligence (AI) technology, some intelligent agent systems based on single LLMs have been used for specific tasks in the healthcare field, such as simple symptom consultation and preliminary diagnostic recommendations. However, these systems are also limited by their reasoning capabilities and struggle to handle complex cross-domain problems. Therefore, improving the performance of automated business process systems in handling complex interactive information in areas such as fintech, healthcare, and elderly care has become a pressing technical challenge. Summary of the Invention

[0004] The present application provides an information interaction method, device, equipment and medium based on multi-agent collaboration to improve the performance of automated business process systems in processing complex interactive information in the fields of financial technology, medical health and elderly care.

[0005] In a first aspect, the present application provides an information interaction method based on multi-agent collaboration, the method comprising:

[0006] Receiving current user needs through the routing agent, and splitting the current user needs through the large language model built into the routing agent to generate at least one target workflow;

[0007] Determine, by the inductive agent, a target API from a preset application programming interface (API) library according to each target workflow, and generate initial interaction information with each target workflow through the target API;

[0008] Based on the initial interaction information and the criticizing agent, generating current reply evaluation information corresponding to the initial interaction information, and determining a current matching degree between the initial interaction information and the current user demand according to the reply evaluation information;

[0009] When the current matching degree is greater than or equal to a preset matching degree threshold, the initial interaction information is determined as target interaction information.

[0010] In a second aspect, the present application further provides an information interaction device based on multi-agent collaboration, the device comprising:

[0011] a target workflow generation module, configured to receive current user requirements through the routing agent, and decompose the current user requirements using the large language model built into the routing agent to generate at least one target workflow;

[0012] an initial interaction information generating module, configured to determine a target API from a preset application programming interface (API) library according to each target workflow through the inductive agent, and generate initial interaction information with each target workflow through the target API;

[0013] a current matching degree determination module, configured to generate current reply evaluation information corresponding to the initial interaction information based on the initial interaction information and the criticizing agent, and determine a current matching degree between the initial interaction information and the current user demand based on the reply evaluation information;

[0014] The target interaction information determining module is configured to determine the initial interaction information as target interaction information when the current matching degree is greater than or equal to a preset matching degree threshold.

[0015] In a third aspect, the present application also provides a computer device comprising a memory and a processor; the memory is used to store a computer program; the processor is used to execute the computer program and implement the above-mentioned information interaction method based on multi-agent collaboration when executing the computer program.

[0016] In a fourth aspect, the present application also provides a computer-readable storage medium, which stores a computer program. When the computer program is executed by a processor, the processor implements the information interaction method based on multi-agent collaboration as described above.

[0017] The present application discloses an information interaction method, apparatus, device and medium based on multi-agent collaboration, wherein the information interaction method based on multi-agent collaboration includes receiving current user needs through the routing agent, and splitting the current user needs through the large language model built into the routing agent to generate at least one target workflow; determining the target API from a preset application programming interface API library according to each target workflow through the inductive agent, and generating initial interaction information through the target API and each target workflow; generating current reply evaluation information corresponding to the initial interaction information based on the initial interaction information and the criticizing agent, and determining the current matching degree between the initial interaction information and the current user needs based on the reply evaluation information; and determining the initial interaction information as the target interaction information when the current matching degree is greater than or equal to a preset matching degree threshold. In the above manner, the present application accurately splits user needs through the routing agent and the built-in large language model, generates target workflows, and flexibly adapts to complex and changing task scenarios. By inductively identifying the target API and generating initial interaction information, the efficiency and accuracy of API calls are improved, thereby improving the performance of automated business process systems in handling complex interaction information in areas such as financial technology, healthcare, and elderly care. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the following is a brief introduction to the drawings required for use in the description of the embodiments. Obviously, the drawings described below are some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without any creative work.

[0019] Figure 1 This is a schematic flow chart of an information interaction method based on multi-agent collaboration provided in the first embodiment of the present application;

[0020] Figure 2 is a schematic flow chart of an information interaction method based on multi-agent collaboration provided in the second embodiment of the present application;

[0021] Figure 3 A schematic block diagram of an information interaction device based on multi-agent collaboration provided in an embodiment of the present application;

[0022] Figure 4 A schematic block diagram of the structure of a computer device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0023] The following will be combined with the drawings in the embodiments of this application to clearly and completely describe the technical solutions in the embodiments of this application. Obviously, the embodiments described are part of the embodiments of this application, not all of them. Based on the embodiments in this application, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of this application.

[0024] The flowcharts shown in the accompanying drawings are for illustrative purposes only and do not necessarily include all contents and operations / steps, nor must they be executed in the order described. For example, some operations / steps may be decomposed, combined, or partially merged, so the actual execution order may vary depending on the actual situation.

[0025] It should be understood that the terms used in this specification are only for the purpose of describing specific embodiments and are not intended to limit the present application. As used in this specification and the appended claims, the singular forms "a", "an", and "the" are intended to include the plural forms unless the context clearly indicates otherwise.

[0026] It will also be understood that the term "and / or" used in this specification and the appended claims refers to and includes any and all possible combinations of one or more of the associated listed items.

[0027] The embodiments of the present application provide a method, apparatus, device, and medium for information interaction based on multi-agent collaboration. Specifically, this multi-agent collaborative information interaction method can be applied to automated business process systems. By using routing agents and a built-in large language model, it accurately breaks down user needs, generates target workflows, and flexibly adapts to complex and changing task scenarios. By using an inductive agent to determine the target API and generate initial interaction information, the efficiency and accuracy of API calls are improved, thereby enhancing the performance of automated business process systems in processing complex interaction information in fields such as financial technology, healthcare, and elderly care.

[0028] The following describes some embodiments of the present application in detail with reference to the accompanying drawings. In the absence of conflict, the following embodiments and features therein may be combined with each other.

[0029] See also Figure 1 , Figure 1 This is a schematic flow chart of a multi-agent collaborative information interaction method provided in the first embodiment of this application. This multi-agent collaborative information interaction method can be applied to automated business process systems in fields such as financial technology, healthcare, and elderly care, improving the performance of automated business process systems in processing complex interactive information.

[0030] like Figure 1As shown, the information interaction method based on multi-agent collaboration specifically includes steps S10 to S40.

[0031] Step S10: receiving current user needs through the routing agent, and splitting the current user needs through the large language model built into the routing agent to generate at least one target workflow;

[0032] Specifically, the routing agent monitors and receives real-time user requests, which are presented in natural language. These requests might include inquiries about insurance claims processes in the financial sector or initial diagnosis requests in the medical field. The routing agent then performs basic text cleanup and formatting on these requests, removing unnecessary spaces and correcting obvious text errors to ensure accuracy and completeness, preparing for subsequent processing.

[0033] Leveraging the large language model built into the routing agent, the pre-processed user needs are deeply semantically understood and analyzed, broken down into multiple subtasks or key elements, and generating at least one target workflow. For example, insurance claims in the financial sector can be broken down into workflows such as information collection, liability determination, and compensation calculation. Medical consultations can be broken down into workflows such as symptom analysis, possible disease matching, and examination recommendations.

[0034] Step S20: determining a target API from a preset application programming interface (API) library according to each target workflow through the inductive agent, and generating initial interaction information with each target workflow through the target API;

[0035] Specifically, the inductive agent obtains each target workflow generated by the routing agent, analyzes key information such as the task type, required data, and operation requirements of each workflow, and clarifies the specific requirements of this task for data and function calls.

[0036] Based on the analysis results of the workflow, the target API (Application Programming Interface) required to complete each workflow is determined by searching and matching from the preset application programming interface library. The API library stores a large number of business-related, pre-organized and defined API resources, covering various functions such as claims calculation and customer information query in the financial field, and disease data query and medical literature retrieval in the medical field.

[0037] Follow the target API's interface specifications and requirements to send a request to it, invoke the corresponding function, or obtain the required data. During the call process, ensure that the correct parameters and format are passed to ensure that the API can accurately understand and respond to the request and return valid data or execution results.

[0038] The inductive agent integrates and processes the data and execution results obtained from each target API, generating initial interaction information according to pre-set logic and format requirements. The content and format of this initial interaction information are determined based on user needs and business scenarios, providing users with more complete and intuitive feedback. For example, in the financial sector, this can include detailed descriptions of the claims process and estimated compensation amounts, while in the medical sector, it can provide analysis of possible causes of symptoms and preliminary examination recommendations.

[0039] Step S30: generating current response evaluation information corresponding to the initial interaction information based on the initial interaction information and the criticizing agent, and determining a current matching degree between the initial interaction information and the current user demand according to the response evaluation information;

[0040] Specifically, the generated initial interaction information is passed to the criticizing agent, which evaluates it from multiple perspectives based on pre-set evaluation rules. This agent assesses whether the data, facts, and other content in the initial interaction information align with reality and accurately answers the user's questions or meets their needs. For example, in the financial sector, this agent verifies whether the claim amount is calculated correctly; in the medical field, this agent determines whether the symptom analysis is consistent with medical knowledge and the patient's symptoms.

[0041] Check whether the initial interaction information covers all key aspects of user needs and whether important information or steps are omitted. For example, in the claim response in the financial field, whether the complete list of materials and process instructions required for the claim are included; in the medical field, whether each relevant symptom raised by the patient is analyzed and explained.

[0042] Analyze whether the structure and content of the initial interactive information are reasonable, whether the logical relationship between the information is clear and coherent, and whether it is easy for users to understand. For example, in the description of the claims process in the financial field, is the order of each step reasonable? In the diagnosis recommendations in the medical field, is the reasoning between symptoms and possible diseases smooth?

[0043] Step S40: When the current matching degree is greater than or equal to a preset matching degree threshold, the initial interaction information is determined as target interaction information.

[0044] Specifically, the current match degree is compared with a preset match degree threshold to determine whether the current match degree is greater than or equal to the threshold. The preset match degree threshold is a standard value set based on business requirements and user expectations. It is used to determine whether the initial interaction information is sufficient to meet user needs and can be adjusted and optimized based on actual business scenarios and experience.

[0045] If the current match degree is greater than or equal to the preset match degree threshold, the initial interaction information is considered to have reached a relatively ideal level in terms of content accuracy, completeness, logical coherence, and fit with user needs, and can provide the user with a high-quality, satisfactory response. The initial interaction information is then officially designated as the target interaction information. The target interaction information can then be returned to the user, completing the interaction process.

[0046] The present embodiment discloses an information interaction method based on multi-agent collaboration, which includes receiving current user needs through the routing agent, and splitting the current user needs through the large language model built into the routing agent to generate at least one target workflow; determining the target API from the preset application programming interface API library according to each target workflow through the inductive agent, and generating initial interaction information through the target API and each target workflow; generating current reply evaluation information corresponding to the initial interaction information based on the initial interaction information and the criticizing agent, and determining the current matching degree between the initial interaction information and the current user needs based on the reply evaluation information; and determining the initial interaction information as the target interaction information when the current matching degree is greater than or equal to the preset matching degree threshold. In the above manner, the present application accurately splits user needs through the routing agent and the built-in large language model, generates target workflows, and flexibly adapts to complex and changing task scenarios. By determining the target API and generating initial interaction information through the inductive agent, the efficiency and accuracy of calling the API are improved, thereby improving the performance of the automated business process system in processing complex interaction information in the fields of financial technology, medical health and elderly care.

[0047] based on Figure 1 In the embodiment shown, in this embodiment, before step S40, the following steps are further included:

[0048] When the current matching degree is less than a preset matching degree threshold, a precise interaction process is triggered to generate a precise reply text.

[0049] Specifically, when the current matching degree is less than a preset matching degree threshold, the precise interaction process is automatically triggered to prepare for generating a reply text that better meets the user's needs.

[0050] In a specific embodiment, when the current matching degree is less than a preset matching degree threshold, triggering a precise interaction process to generate a precise reply text includes:

[0051] When the current matching degree is less than a preset matching degree threshold, generating feedback information of the initial interaction information according to the reply evaluation information;

[0052] Specifically, according to the scores of each index (such as accuracy, completeness, logical coherence, etc.) in the reply evaluation information, the deficiencies of the initial interaction information are analyzed, for example: if the accuracy evaluation score is low, it means that the provided information deviates from the actual situation; if the completeness evaluation score is low, it means that some key content is missing.

[0053] The feedback information can be the problems in the initial interaction information, such as which parts are not accurate enough, which key points are not mentioned, and where the logic defects are, etc. At the same time, it can also include further clarification requests for user needs, asking users for more detailed information or expected modification direction about these deficiencies.

[0054] For example, the feedback information can be: "We noticed that in the insurance claim process you consulted, the explanation of the compensation ratio under special circumstances may not be accurate enough. Could you please explain the special circumstances you are concerned about in detail so that we can provide you with more accurate information?"

[0055] The routing agent generates a current supplementary demand of the current user demand according to the current user demand and the feedback information;

[0056] Specifically, the routing agent receives the feedback information generated by the criticism agent, carefully analyzes the problems pointed out and the user's further clarification requests, combines the feedback information with the original current user demand, and reorganizes and deeply understands the user demand, for example, the original demand is that the user wants to understand the general process of insurance claim, and the feedback information points out that the compensation under special circumstances needs to be supplemented.

[0057] According to the reanalyzed user demand, a current supplementary demand of the current user demand is generated, which needs to clearly supplement the specific content and key points, for example, the current supplementary demand is to explain in detail the calculation method of the loss compensation ratio caused by natural disasters in insurance claim, the required additional materials, and the claim time limit, etc.

[0058] According to the current user demand and the current supplementary demand, an accurate workflow is determined, and the accurate reply text is generated through the target API and the accurate workflow.

[0059] Specifically, according to the current user demand and the current supplementary demand, the accurate workflow is re-determined, which may include the interpretation of special compensation clauses, historical case query, API data acquisition with relevant departments, etc. According to the accurate workflow, the target API is called again, and the purpose of calling the API is more clear and the parameter setting is more accurate.

[0060] Integrate the data and information obtained from the target API and generate a precise response text in a clear logical structure and easy-to-understand language style. For example, the precise response text begins by expressing understanding of the user's questions about special compensation situations, then elaborates on the calculation method of the compensation ratio (such as including specific formulas and parameter descriptions), and then lists one or two similar historical claims cases for auxiliary explanation. Finally, inform the user of key information such as the required additional materials and claim time limit to ensure that the user can get a complete, accurate and satisfactory response.

[0061] See also Figure 2 , Figure 2 This is a schematic flow chart of an information interaction method based on multi-agent collaboration provided in the second embodiment of the present application. This information interaction method based on multi-agent collaboration can be applied to an automated business process system. It is used to perform a multi-dimensional evaluation of the initial interaction information through the preset evaluation rules built into the criticism agent, generate single-indicator evaluation scores such as accuracy, completeness, and logical coherence, calculate the comprehensive evaluation score of the initial interaction information based on the weight of each evaluation indicator, and generate the current response evaluation information based on this, reflecting the difference in importance of different evaluation indicators and improving the performance of the automated business process system in processing complex interaction information.

[0062] based on Figure 1 The embodiment shown, this embodiment Figure 2 As shown, step S30 includes steps S301 to S304.

[0063] Step S301: Perform a multi-dimensional evaluation on the initial interaction information using the preset evaluation rules built into the critic agent to generate individual indicator evaluation scores, wherein the individual indicator evaluation scores include an accuracy evaluation score, a completeness evaluation score, and a logical coherence evaluation score;

[0064] Specifically, the critic agent calls built-in preset evaluation rules, which define how to evaluate the initial interaction information from different dimensions.

[0065] Compare the data and facts in the initial interactive information with known accurate information sources (such as insurance policy terms in the financial field and authoritative medical knowledge bases in the medical field). Calculate the accuracy evaluation score based on the degree of accuracy match. For example, if the information is completely accurate, the score is 100 points. If there are some deviations, points will be deducted based on the degree of deviation, such as 40-60 points for large deviations and 10-30 points for small deviations.

[0066] analyzing whether the initial interaction information covers all key aspects related to the user demand, checking whether there is missing information or steps, calculating a completeness evaluation score according to the degree of completeness, for example, if all points are covered completely, the score is 100 points, if an important point is missed, 30-50 points are deducted, if a secondary point is missed, 10-20 points are deducted, and so on;

[0067] checking whether the structure and content organization of the initial interaction information is reasonable, whether the logical relationship between the information is clear, coherent, and easy for the user to understand, calculating a logical coherence evaluation score according to the performance of logical coherence, if the logical structure is rigorous and clear, the score is 100 points, if the logic is chaotic and difficult to understand, 40-60 points are deducted, if the logic has some flaws, 10-30 points are deducted, and so on

[0068] In step S302, a comprehensive evaluation score of the initial interaction information is calculated according to the evaluation index weight corresponding to the accuracy evaluation score, the completeness evaluation score and the logical coherence evaluation score, and the current reply evaluation information is generated according to the comprehensive evaluation score;

[0069] Specifically, the weights of the three evaluation indexes of accuracy, completeness and logical coherence are obtained from the preset parameters, for example, the accuracy weight is 0.4, the completeness weight is 0.3, and the logical coherence weight is 0.3. Each single-index evaluation score (accuracy evaluation score, completeness evaluation score, logical coherence evaluation score) is multiplied by the corresponding evaluation index weight, and then the three results are added to obtain the comprehensive evaluation score of the initial interaction information.

[0070] In step S303, a historical reply text corresponding to a related user demand is obtained, wherein the related user demand has an association degree greater than a preset association degree threshold with the current user demand.

[0071] Specifically, in the historical database, the user demand records related to the current user demand are searched, and the search range covers various business scenarios in the financial field and the medical field.

[0072] For the searched related user demand, the association degree thereof with the current user demand is calculated. The calculation of the association degree can be based on text similarity (comparing the similarity of keywords, semantics, etc. in the demand text), business scenario relevance (judging whether the business field and process to which the demand belongs are the same or similar), and other factors to make a comprehensive judgment. For example, in the financial field, if the current demand is about insurance claim process consultation, and the searched historical demand is about the requirement of insurance claim material submission, because both demands involve insurance claim business and some keywords are similar, it is preliminarily judged that there is a certain association degree. According to the association degree calculation result, the historical reply text with an association degree greater than a preset association degree threshold with the current user demand is selected.

[0073] Step S304, calculate the similarity between the initial interaction information and the historical reply text, and determine the current matching degree of the initial interaction information and the current user demand according to the similarity.

[0074] Specifically, a suitable text similarity algorithm, such as a cosine similarity algorithm, a Jaccard similarity algorithm, etc., is selected to calculate the similarity between the initial interaction information and the historical reply text. Text feature extraction is performed on the initial interaction information and the historical reply text respectively to extract feature information such as keywords, phrases, semantic vectors, etc. in the text for similarity calculation.

[0075] According to the comprehensive evaluation score of the initial interaction information and the similarity with the historical reply text, the current matching degree of the initial interaction information and the current user demand is determined comprehensively. The calculation method of the current matching degree can be designed according to specific business rules. A basic matching degree value is set, and the basic matching degree is adjusted according to the comprehensive evaluation score. The higher the comprehensive evaluation score, the greater the adjustment range of the basic matching degree. At the same time, the adjusted basic matching degree is further corrected according to the similarity with the historical reply text. The higher the similarity, the higher the finally determined current matching degree.

[0076] The embodiment discloses an information interaction method based on multi-agent cooperation, which comprises the following steps: performing multi-dimensional evaluation on the initial interaction information through the preset evaluation rule built in the criticism agent to generate single-index evaluation scores, wherein the single-index evaluation scores include accuracy evaluation scores, integrity evaluation scores, and logical coherence evaluation scores; calculating a comprehensive evaluation score of the initial interaction information according to the evaluation index weights corresponding to the accuracy evaluation scores, the integrity evaluation scores, and the logical coherence evaluation scores, and generating the current reply evaluation information according to the comprehensive evaluation score; obtaining historical reply texts corresponding to related user demands, wherein the correlation degree between the related user demands and the current user demand is greater than a preset correlation degree threshold; calculating the similarity between the initial interaction information and the historical reply texts, and determining the current matching degree of the initial interaction information and the current user demand according to the similarity. In the above manner, the initial interaction information is evaluated in multiple dimensions through the preset evaluation rule built in the criticism agent to generate single-index evaluation scores such as accuracy, integrity, and logical coherence. The comprehensive evaluation score of the initial interaction information is calculated according to the evaluation index weights, and the current reply evaluation information is generated accordingly, which reflects the importance difference of different evaluation indexes and improves the performance of the automatic business process system in processing complex interaction information.

[0077] Based on Figure 1 In the embodiment shown in the figure, step S20 comprises:

[0078] Analyzing each of the target workflows by the inductive agent to determine target task parameters of each process node in each of the target workflows;

[0079] Matching the preset task parameters of each preset API in the preset API library with the target task parameters;

[0080] The preset API corresponding to the preset task parameters matching the target task parameters is determined as the target API.

[0081] Specifically, the inductive agent receives the target workflows to be processed from the routing agent, carefully analyzes each target workflow, and identifies the various process nodes therein. A process node refers to an independent part of the workflow with a specific task or function. For example, in an insurance claims workflow, process nodes can be nodes such as "information collection", "liability determination", and "compensation calculation"; in a medical diagnosis workflow, process nodes can be nodes such as "symptom analysis", "disease matching", and "examination recommendations".

[0082] For each process node, extract its target task parameters. The target task parameters refer to the key information and conditions required to complete the node task, including but not limited to data type (such as text, number, date, etc.), data format, parameter name, and parameter value range.

[0083] The preset API library covers various functions and services in the financial and medical fields, such as insurance claim calculation API, customer credit assessment API, etc. in the financial field, and disease data query API, medical literature retrieval API, etc. in the medical field.

[0084] For each preset API in the preset API library, extract its preset task parameters. Preset task parameters refer to the input parameters required for the tasks that the API can handle, and also include information such as data type, data format, parameter name, parameter value range, etc. For example, the preset task parameters of an insurance claim calculation API may include "claim type" (data type is text, value range is "accident insurance", "health insurance", "auto insurance", etc.), "loss amount" (data type is numeric, value range is numbers greater than 0), etc.

[0085] In a specific embodiment, step S20 further includes:

[0086] Analyze the target task parameters to determine the operation instructions and data requirements of each process node in the target workflow;

[0087] According to the operation instruction and the data requirement, the corresponding target information is called through the target API;

[0088] The initial interaction information is generated using the target API and the target information.

[0089] Specifically, each target workflow is analyzed to identify the various process nodes therein, such as the "information collection", "liability determination", "compensation calculation" and other nodes in the insurance claims workflow. For each process node, operation instructions are extracted. Based on the extracted operation instructions and data requirements, combined with the preset API library, the target API that can provide the required functions and data is determined. The extracted data requirements are formatted and parameterized according to the requirements of the target API to ensure that they meet the input requirements of the API. The target API is called through network requests or other communication methods, and operation instructions and data requirements are sent to obtain target information (such as claim calculation results, disease diagnosis recommendations, etc.).

[0090] Based on any of the above embodiments, in this embodiment, step S10 includes:

[0091] Extracting at least one demand keyword of the current user demand through the routing agent;

[0092] Determining the semantic logical relationship between the demand keywords based on the demand keywords using a preset semantic database and the large language model;

[0093] Based on the semantic logical relationships, at least one target workflow corresponding to the current user requirement is generated.

[0094] Specifically, the semantic analysis capability of the large language model built into the routing agent is used to conduct in-depth semantic understanding of the extracted demand keywords. Based on its massive training data and powerful language generation capabilities, the large language model can capture the subtle semantic differences and potential logical associations between keywords, and comprehensively analyze the results of the preset semantic database and the large language model to determine the semantic logical relationship between each demand keyword.

[0095] Based on the semantic logical relationship between each demand keyword, matching and mapping are performed in the system's preset workflow template library. The workflow template library stores a variety of workflow templates for different business scenarios and semantic logic combinations, covering various common business processes such as claims processing and customer service in the financial field, diagnostic processes and treatment plan recommendations in the medical field.

[0096] Each target workflow includes a series of process nodes and operational steps required to fulfill user needs. It clarifies the task parameters, execution sequence, and logical relationships of each node, providing clear process guidance for subsequent business processing and information exchange. For example, the insurance claim query and compensation calculation workflow generated in the financial field will guide the system to complete tasks such as progress query, information verification, and amount calculation. The disease diagnosis and examination recommendation workflow generated in the medical field will guide the system to perform operations such as symptom analysis, disease matching, and examination recommendations to meet user consultation needs.

[0097] See also Figure 3 , Figure 3 The embodiment of the present application provides a schematic block diagram of an information interaction device based on multi-agent collaboration, which is used to execute the aforementioned information interaction method based on multi-agent collaboration. The information interaction device based on multi-agent collaboration can be configured on a server.

[0098] like Figure 3 As shown, the information interaction device 400 based on multi-agent collaboration includes:

[0099] A target workflow generation module 410 is configured to receive current user requirements through the routing agent, and to decompose the current user requirements using the large language model built into the routing agent to generate at least one target workflow;

[0100] An initial interaction information generating module 420 is configured to determine a target API from a preset application programming interface (API) library according to each target workflow through the inductive agent, and generate initial interaction information with each target workflow through the target API;

[0101] A current matching degree determination module 430 is configured to generate current response evaluation information corresponding to the initial interaction information based on the initial interaction information and the criticizing agent, and determine a current matching degree between the initial interaction information and the current user demand based on the response evaluation information;

[0102] The target interaction information determining module 440 is configured to determine the initial interaction information as target interaction information if the current matching degree is greater than or equal to a preset matching degree threshold.

[0103] Furthermore, the information interaction device 400 based on multi-agent collaboration further includes:

[0104] The precise reply text generation module is used to trigger the precise interaction process and generate the precise reply text when the current matching degree is less than a preset matching degree threshold.

[0105] Furthermore, the accurate reply text generation module includes:

[0106] A feedback information generating unit, configured to generate feedback information of the initial interaction information according to the reply evaluation information when the current matching degree is less than a preset matching degree threshold;

[0107] a current supplementary demand generating unit, configured to generate, through the routing agent, a current supplementary demand for the current user demand based on the current user demand and the feedback information;

[0108] The precise reply text generating unit is used to determine a precise workflow according to the current user demand and the current supplementary demand, and to generate the precise reply text through the target API and the precise workflow.

[0109] Furthermore, the current matching degree determination module 430 includes:

[0110] a single-indicator evaluation score generating unit, configured to perform a multi-dimensional evaluation of the initial interaction information using preset evaluation rules built into the critic agent, and generate individual single-indicator evaluation scores, wherein the individual single-indicator evaluation scores include an accuracy evaluation score, a completeness evaluation score, and a logical coherence evaluation score;

[0111] a previous reply evaluation information generating unit, configured to calculate a comprehensive evaluation score of the initial interactive information according to the evaluation indicator weights corresponding to the accuracy evaluation score, the completeness evaluation score, and the logical coherence evaluation score, and generate the current reply evaluation information according to the comprehensive evaluation score;

[0112] A historical reply text acquisition unit, configured to acquire historical reply texts corresponding to relevant user needs, wherein the relevance between the relevant user needs and the current user needs is greater than a preset relevance threshold;

[0113] A current matching degree determination unit is used to calculate the similarity between the initial interaction information and the historical reply text, and determine the current matching degree between the initial interaction information and the current user demand based on the similarity.

[0114] Furthermore, the initial interaction information generating module 420 includes:

[0115] a target task parameter determination unit, configured to parse each of the target workflows through the inductive agent and determine the target task parameters of each process node in each of the target workflows;

[0116] A task parameter matching unit, configured to match the preset task parameters of each preset API in the preset API library with the target task parameters;

[0117] The target API determining unit is configured to determine the preset API corresponding to the preset task parameters that match the target task parameters as the target API.

[0118] Furthermore, the initial interaction information generating module 420 includes:

[0119] An operation instruction and data requirement determination unit, configured to parse the target task parameters and determine the operation instructions and data requirements of each process node in the target workflow;

[0120] A target information calling unit, configured to call corresponding target information through the target API according to the operation instruction and the data requirement;

[0121] The initial interaction information generating unit is configured to generate the initial interaction information through the target API and the target information.

[0122] Furthermore, the target workflow generation module 410 includes:

[0123] A demand keyword extraction unit, configured to extract at least one demand keyword of the current user demand through the routing agent;

[0124] a semantic logical relationship determining unit, configured to determine the semantic logical relationship between the demand keywords based on the demand keywords using a preset semantic database and the large language model;

[0125] The target workflow generating unit is configured to generate at least one target workflow corresponding to the current user requirement based on each of the semantic logical relationships.

[0126] It should be noted that those skilled in the art will clearly understand that, for the convenience and brevity of description, the specific working processes of the above-described devices and modules can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0127] The above-mentioned device can be realized in the form of a computer program. The computer program can be used in Figure 4 Runs on the computer equipment shown.

[0128] See also Figure 4 , Figure 4 1 is a schematic block diagram of a computer device provided in an embodiment of the present application. The computer device may be a server.

[0129] See Figure 4 The computer device includes a processor, a memory, and a network interface connected through a system bus, wherein the memory may include a non-volatile storage medium and an internal memory.

[0130] The non-volatile storage medium can store an operating system and a computer program. The computer program includes program instructions, which, when executed, can enable the processor to execute any information interaction method based on multi-agent collaboration.

[0131] The processor is used to provide computing and control capabilities and support the operation of the entire computer equipment.

[0132] The internal memory provides an environment for the operation of computer programs in non-volatile storage media. When the computer program is executed by the processor, the processor can execute any information interaction method based on multi-agent collaboration.

[0133] The network interface is used for network communication, such as sending assigned tasks, etc. Those skilled in the art will understand that Figure 4 The structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than shown in the figure, or combine certain components, or have a different component arrangement.

[0134] It should be understood that the processor may be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor or any conventional processor, etc.

[0135] In one embodiment, the processor is configured to execute a computer program stored in the memory to implement the following steps:

[0136] Receiving current user needs through the routing agent, and splitting the current user needs through the large language model built into the routing agent to generate at least one target workflow;

[0137] Determine, by the inductive agent, a target API from a preset application programming interface (API) library according to each target workflow, and generate initial interaction information with each target workflow through the target API;

[0138] Based on the initial interaction information and the criticizing agent, generating current reply evaluation information corresponding to the initial interaction information, and determining a current matching degree between the initial interaction information and the current user demand according to the reply evaluation information;

[0139] When the current matching degree is greater than or equal to a preset matching degree threshold, the initial interaction information is determined as target interaction information.

[0140] In one embodiment, when the current matching degree is greater than or equal to a preset matching degree threshold, before determining the initial interaction information as target interaction information, it is used to implement:

[0141] When the current matching degree is less than a preset matching degree threshold, a precise interaction process is triggered to generate a precise reply text.

[0142] In one embodiment, when the current matching degree is less than a preset matching degree threshold, a precise interaction process is triggered to generate a precise reply text, so as to achieve:

[0143] When the current matching degree is less than a preset matching degree threshold, generating feedback information of the initial interaction information according to the reply evaluation information;

[0144] generating, by the routing agent, a current supplementary demand for the current user demand according to the current user demand and the feedback information;

[0145] According to the current user demand and the current supplementary demand, a precise workflow is determined, and the precise reply text is generated through the target API and the precise workflow.

[0146] In one embodiment, based on the initial interaction information and the critic agent, current response evaluation information corresponding to the initial interaction information is generated, and the current matching degree between the initial interaction information and the current user demand is determined based on the response evaluation information, so as to achieve:

[0147] Performing a multi-dimensional evaluation of the initial interaction information using the preset evaluation rules built into the criticizing agent to generate individual indicator evaluation scores, wherein the individual indicator evaluation scores include an accuracy evaluation score, a completeness evaluation score, and a logical coherence evaluation score;

[0148] Calculating a comprehensive evaluation score of the initial interactive information according to the evaluation indicator weights corresponding to the accuracy evaluation score, the completeness evaluation score, and the logical coherence evaluation score, and generating the current reply evaluation information according to the comprehensive evaluation score;

[0149] Obtaining historical reply texts corresponding to relevant user needs, wherein the relevance between the relevant user needs and the current user needs is greater than a preset relevance threshold;

[0150] The similarity between the initial interaction information and the historical reply text is calculated, and the current matching degree between the initial interaction information and the current user demand is determined based on the similarity.

[0151] In one embodiment, the inductive agent determines a target API from a preset application programming interface (API) library according to each target workflow, for implementing:

[0152] Analyzing each of the target workflows by the inductive agent to determine target task parameters of each process node in each of the target workflows;

[0153] Matching the preset task parameters of each preset API in the preset API library with the target task parameters;

[0154] The preset API corresponding to the preset task parameters matching the target task parameters is determined as the target API.

[0155] In one embodiment, initial interaction information is generated through the target API and each target workflow to achieve:

[0156] Analyze the target task parameters to determine the operation instructions and data requirements of each process node in the target workflow;

[0157] According to the operation instruction and the data requirement, the corresponding target information is called through the target API;

[0158] The initial interaction information is generated using the target API and the target information.

[0159] In one embodiment, the routing agent receives current user requirements and uses a large language model built into the routing agent to decompose the current user requirements to generate at least one target workflow for achieving:

[0160] Extracting at least one demand keyword of the current user demand through the routing agent;

[0161] Determining the semantic logical relationship between the demand keywords based on the demand keywords using a preset semantic database and the large language model;

[0162] Based on the semantic logical relationships, at least one target workflow corresponding to the current user requirement is generated.

[0163] A computer-readable storage medium is also provided in an embodiment of the present application, wherein the computer-readable storage medium stores a computer program, wherein the computer program includes program instructions, and the processor executes the program instructions to implement any one of the information interaction methods based on multi-agent collaboration provided in the embodiments of the present application.

[0164] The computer-readable storage medium may be an internal storage unit of the computer device described in the aforementioned embodiment, such as a hard disk or memory of the computer device. The computer-readable storage medium may also be an external storage device of the computer device, such as a plug-in hard disk, a SmartMedia Card (SMC), a Secure Digital (SD) card, a flash memory card, etc., equipped on the computer device.

[0165] The above description is merely a specific embodiment of the present application, but the scope of protection of the present application is not limited thereto. Any person skilled in the art can easily conceive of various equivalent modifications or substitutions within the technical scope disclosed in the present application, and such modifications or substitutions should be included in the scope of protection of the present application. Therefore, the scope of protection of the present application should be based on the scope of protection of the claims.

Claims

1. An information interaction method based on multi-agent collaboration, characterized in that: The information interaction method based on multi-agent collaboration is applied to an information interaction system based on multi-agent collaboration, wherein the information interaction system based on multi-agent collaboration includes a routing agent, an induction agent, and a criticism agent. The information interaction method based on multi-agent collaboration includes: Receiving current user needs through the routing agent, and splitting the current user needs through the large language model built into the routing agent to generate at least one target workflow; Determine a target API from a preset application programming interface API library according to each target workflow through the inductive agent, and generate initial interaction information with each target workflow through the target API; Based on the initial interaction information and the criticizing agent, generating current reply evaluation information corresponding to the initial interaction information, and determining a current matching degree between the initial interaction information and the current user demand according to the reply evaluation information; When the current matching degree is greater than or equal to a preset matching degree threshold, the initial interaction information is determined as target interaction information.

2. The information interaction method based on multi-agent collaboration according to claim 1 is characterized in that: Before determining the initial interaction information as target interaction information when the current matching degree is greater than or equal to a preset matching degree threshold, the method includes: When the current matching degree is less than a preset matching degree threshold, a precise interaction process is triggered to generate a precise reply text.

3. The information interaction method based on multi-agent collaboration according to claim 2 is characterized in that: When the current matching degree is less than a preset matching degree threshold, triggering a precise interaction process and generating a precise reply text includes: When the current matching degree is less than a preset matching degree threshold, generating feedback information of the initial interaction information according to the reply evaluation information; generating, by the routing agent, a current supplementary demand for the current user demand according to the current user demand and the feedback information; According to the current user demand and the current supplementary demand, a precise workflow is determined, and the precise reply text is generated through the target API and the precise workflow.

4. The information interaction method based on multi-agent collaboration according to claim 1 is characterized in that: The step of generating current reply evaluation information corresponding to the initial interaction information based on the initial interaction information and the criticizing agent, and determining a current matching degree between the initial interaction information and the current user demand according to the reply evaluation information, includes: Performing a multi-dimensional evaluation of the initial interaction information using the preset evaluation rules built into the criticizing agent to generate individual indicator evaluation scores, wherein the individual indicator evaluation scores include an accuracy evaluation score, a completeness evaluation score, and a logical coherence evaluation score; Calculating a comprehensive evaluation score of the initial interactive information according to the evaluation indicator weights corresponding to the accuracy evaluation score, the completeness evaluation score, and the logical coherence evaluation score, and generating the current reply evaluation information according to the comprehensive evaluation score; Obtaining historical reply texts corresponding to relevant user needs, wherein the relevance between the relevant user needs and the current user needs is greater than a preset relevance threshold; The similarity between the initial interaction information and the historical reply text is calculated, and the current matching degree between the initial interaction information and the current user demand is determined based on the similarity.

5. The information interaction method based on multi-agent collaboration according to claim 1 is characterized in that: The step of determining a target API from a preset application programming interface (API) library according to each target workflow by the inductive agent includes: Analyzing each of the target workflows by the inductive agent to determine target task parameters of each process node in each of the target workflows; Matching the preset task parameters of each preset API in the preset API library with the target task parameters; The preset API corresponding to the preset task parameters matching the target task parameters is determined as the target API.

6. The information interaction method based on multi-agent collaboration according to claim 5 is characterized in that: Generating initial interaction information with each target workflow through the target API includes: Analyze the target task parameters to determine the operation instructions and data requirements of each process node in the target workflow; According to the operation instruction and the data requirement, the corresponding target information is called through the target API; The initial interaction information is generated using the target API and the target information.

7. The information interaction method based on multi-agent collaboration according to any one of claims 1 to 6, characterized in that: The receiving of current user requirements by the routing agent and splitting the current user requirements by the large language model built into the routing agent to generate at least one target workflow includes: Extracting at least one demand keyword of the current user demand through the routing agent; Determining the semantic logical relationship between the demand keywords based on the demand keywords using a preset semantic database and the large language model; Based on the semantic logical relationships, at least one target workflow corresponding to the current user requirement is generated.

8. An information interaction device based on multi-agent collaboration, characterized in that: include: a target workflow generation module, configured to receive current user requirements through the routing agent, and decompose the current user requirements using the large language model built into the routing agent to generate at least one target workflow; an initial interaction information generating module, configured to determine a target API from a preset application programming interface (API) library according to each target workflow through the inductive agent, and generate initial interaction information with each target workflow through the target API; a current matching degree determination module, configured to generate current reply evaluation information corresponding to the initial interaction information based on the initial interaction information and the criticizing agent, and determine a current matching degree between the initial interaction information and the current user demand based on the reply evaluation information; The target interaction information determining module is configured to determine the initial interaction information as target interaction information when the current matching degree is greater than or equal to a preset matching degree threshold.

9. A computer device, characterized in that: The computer device includes a memory and a processor; The memory is used to store computer programs; The processor is used to execute the computer program and implement the information interaction method based on multi-agent collaboration as described in any one of claims 1 to 7 when executing the computer program.

10. A computer-readable storage medium, characterized in that The computer-readable storage medium stores a computer program, which, when executed by a processor, enables the processor to implement the information interaction method based on multi-agent collaboration as described in any one of claims 1 to 7.

Images