Data processing method based on multi-agent cooperation and related device

Abstract

The application provides a data processing method and related device based on multi-agent cooperation, the method comprises the following steps: a parent agent calculates the decision weight of each child agent according to user question data, allocates data processing tasks and task execution rules to each child agent according to the decision weight, the child agent executes the data processing task according to the task execution rule, and returns the data processing result to the parent agent, the parent agent fuses the data processing result of each child agent to obtain target reply content. In this way, the matching degree of the child agent and the user demand is quantified by the decision weight, breaking through the limitation of only allocating tasks according to functions in the prior art, improving the matching accuracy of tasks and agent capabilities, and at the same time, the task execution rules are configured to the child agent based on the decision weight, improving the overall processing efficiency and reliability of multi-agent cooperation.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of artificial intelligence, in particular to a data processing method based on multi-agent collaboration and related devices. BACKGROUND

[0002] An agent is an intelligent entity capable of autonomously perceiving the environment, making decisions and executing tasks. Limited by the functional boundaries and knowledge reserves of a single agent, a data processing mechanism based on multi-agent collaboration has become the mainstream technical route in complex scenarios, especially in the intelligent dialogue scenarios in the financial field. This mechanism can to some extent solve the field problems caused by massive heterogeneous data and complex terminology.

[0003] However, under the existing mechanism, the task allocation of the parent agent to the child agent depends mainly on simple function matching, i.e., assigning tasks according to the functions of the child agent, and lacks quantitative evaluation of the matching degree of the child agent and the user demand, resulting in inaccurate task allocation and affecting the overall processing efficiency and reliability of multi-agent collaboration. SUMMARY

[0004] The embodiments of the present application provide a data processing method based on multi-agent collaboration and related devices, in order to quantitatively evaluate the matching degree of the child agent and the user demand, and based on this, to realize accurate configuration of data processing tasks and task execution rules, and to improve the overall processing efficiency and reliability of multi-agent collaboration.

[0005] In a first aspect, the embodiments of the present application provide a data processing method based on multi-agent collaboration, applied to an electronic device, the electronic device comprising a parent agent and a plurality of child agents, the method comprising:

[0006] The parent agent calculates a decision weight of each child agent according to user question data, the decision weight being used to represent the matching degree of the child agent and the user demand;

[0007] The parent agent allocates data processing tasks and task execution rules to each child agent according to the decision weight, the task execution rule being used to indicate the constraint condition of the child agent when executing the data processing task;

[0008] The plurality of child agents execute the data processing tasks according to the task execution rules, and return data processing results to the parent agent;

[0009] The parent agent fuses the data processing results of each child agent to obtain target reply content.

[0010] In a second aspect, an embodiment of the present application provides a data processing apparatus based on multi-agent cooperation, applied to an electronic device, the electronic device comprising a parent agent and a plurality of child agents, the apparatus comprising:

[0011] a calculation unit configured to cause the parent agent to calculate a decision weight of each child agent according to user question data, the decision weight being used to represent a matching degree of the child agent and user demand;

[0012] an allocation unit configured to cause the parent agent to allocate a data processing task and a task execution rule to each child agent according to the decision weight, the task execution rule being used to indicate a constraint condition of the child agent when executing the data processing task;

[0013] an execution unit configured to cause the plurality of child agents to execute the data processing task according to the task execution rule, and return a data processing result to the parent agent;

[0014] a fusion unit configured to cause the parent agent to fuse the data processing result of each child agent to obtain target reply content.

[0015] In a third aspect, an embodiment of the present application provides an electronic device, comprising a processor, a memory, and one or more programs, the one or more programs being stored in the memory and configured to be executed by the processor, the program comprising instructions for performing steps in the method of the first aspect of the present application.

[0016] In a fourth aspect, an embodiment of the present application provides a computer readable storage medium, having stored thereon a computer program or instructions, the computer program or instructions being executed by a processor to implement steps of the method of the first aspect of the present application.

[0017] It can be seen that, in the embodiment of the present application, the parent agent calculates the decision weight of each child agent according to user question data, to quantify the matching degree of the child agent and user demand, and then allocates a data processing task and a task execution rule to each child agent according to the decision weight, the task execution rule being used to indicate a constraint condition of the child agent when executing the data processing task, the child agent executing the data processing task according to the task execution rule and returning a data processing result to the parent agent, and the parent agent fusing the data processing result of each child agent to obtain target reply content. In this way, the matching degree of the child agent and user demand is quantified by the decision weight, breaking through the limitation of allocating tasks according to functions in the prior art, improving the matching accuracy of tasks and agent capabilities, and at the same time, configuring differentiated task execution rules to the child agents based on the decision weight, improving the overall processing efficiency and reliability of multi-agent cooperation. BRIEF DESCRIPTION OF DRAWINGS

[0018] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the accompanying drawings needed to be used in the description of the embodiments or the prior art will be briefly introduced. Obviously, the accompanying drawings in the following description only represent some embodiments of the present application, and for those skilled in the art, other drawings can also be obtained from these accompanying drawings without any creative effort.

[0019] Figure 1 is an architecture block diagram of an intelligent dialogue system provided by an embodiment of the present application;

[0020] Figure 2 is another architecture block diagram of an intelligent dialogue system provided by an embodiment of the present application;

[0021] Figure 3 is an agent deployment block diagram of an electronic device provided by an embodiment of the present application;

[0022] Figure 4 is a flow schematic diagram of a data processing method based on multi-agent cooperation provided by an embodiment of the present application;

[0023] Figure 5 is a structural block diagram of a data processing device based on multi-agent cooperation provided by an embodiment of the present application;

[0024] Figure 6 is another structural block diagram of a data processing device based on multi-agent cooperation provided by an embodiment of the present application;

[0025] Figure 7 is a structural schematic diagram of an electronic device provided by an embodiment of the present application. DETAILED DESCRIPTION

[0026] In order to make the person skilled in the art better understand the present application, the technical solutions in the embodiments of the present application will be described clearly and completely with reference to the accompanying drawings in the embodiments of the present application. Obviously, the described embodiments only represent some of the embodiments of the present application, but not all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without any creative effort are within the scope of protection of the present application.

[0027] The terms "first", "second", and the like in the description and in the claims of the present application and above drawings are used for distinguishing between similar objects talking about the illustrations and not necessarily for describing a specific sequential or chronological order. Moreover, the terms "comprises", "comprising", "includes", "including" and the like are meant to be interpreted open-ended. For example, a process, method, system, product, or apparatus that comprises a list of steps or elements is not necessarily limited to the listed steps or elements, but can include additional steps or elements not expressly listed or inherent to such process, method, system, product, or apparatus.

[0028] Reference herein to "an embodiment" means that a particular feature, structure, or characteristic described in connection with an embodiment is included in at least one embodiment of the application. The appearances of the phrase in various places in the specification are not necessarily all referring to the same embodiment, nor are they necessarily mutually exclusive of one another. As will be apparent to those of ordinary skill in the art, embodiments described herein can be combined with other embodiments.

[0029] In the related art, the data processing mechanism based on multi-agent collaboration generally is that the parent agent (or main agent) performs semantic understanding and intent recognition, and then distributes data processing tasks to each child agent according to the core function of each child agent, for example, instructing the data retrieval agent to retrieve target data from the target data source, and / or instructing the literature analysis agent to read and analyze the literature content. However, when facing complex requirements, the parent agent only distributes data processing tasks according to the function, without considering the matching degree of user requirements and each child agent. Under the limitation of computing resources, the child agent with high matching degree may not be able to explore deeply, and the child agent with low matching degree undertakes redundant work, which affects the overall processing efficiency and reliability of multi-agent collaboration. Especially in the intelligent dialogue scene in the financial field, the user requirements implied in the user's question are usually extremely demanding on the accuracy, reliability and response efficiency of the data, and the existing data processing mechanism based on multi-agent collaboration cannot effectively support the intelligent dialogue scene in the financial field.

[0030] To solve the above problems, the embodiments of the present application provide a data processing method based on multi-agent collaboration and related devices.

[0031] The following introduces the system architecture related to the embodiments of the present application.

[0032] In one embodiment, as Figure 1As shown, the intelligent dialogue system 10 includes a first electronic device 11 and a second electronic device 12, which are communicatively connected. Among them, the first electronic device 11 is a client, and the second electronic device 12 is a server. The user performs a questioning operation on the first electronic device 11. The first electronic device 11 converts the questioning operation into user questioning data and sends the user questioning data to the second electronic device 12. The second electronic device 12 performs the data processing method based on multi-agent collaboration as involved in the embodiments of the present application.

[0033] In another embodiment, as shown in Figure 2 , the intelligent dialogue system 10 only includes a first electronic device 11. The user performs a questioning operation on the first electronic device 11. The first electronic device 11 converts the questioning operation into user questioning data and performs the data processing method based on multi-agent collaboration as involved in the embodiments of the present application.

[0034] Further, as shown in Figure 3 , the electronic device 30 includes a parent agent 31 and a plurality of child agents 32. The parent agent 31 and the plurality of child agents 32 are both functional modules or computing entities running on the electronic device 30, and together constitute a multi-agent collaborative processing architecture. The parent agent 31 is the core control unit of the collaborative architecture, responsible for overall planning of the whole process data processing, including receiving user questioning data, calculating the decision weight of each child agent 32, assigning data processing tasks and task execution rules, and fusing the processing results returned by the child agent 32; the plurality of child agents 32 are functional execution units, each having a specific data processing capability, and need to be dispatched by the parent agent 31 to complete the corresponding data processing work according to the assigned tasks and execution rules, and feed back the results to the parent agent 31. In the embodiments of the present application, the child agent 32 can be implemented as a data retrieval agent, a literature analysis agent and a knowledge question and answer agent. The data retrieval agent is used to retrieve target fields from target data sources, the literature analysis agent is used to read and analyze the content of target literature, extract viewpoints or conclusions, and the knowledge question and answer agent is used to search for explanations or associations of target terms from a knowledge graph.

[0035] Among them, the electronic device 30 can be the first electronic device 11 or the second electronic device 12 as described above.

[0036] Next, a data processing method based on multi-agent collaboration provided by the embodiments of the present application will be introduced.

[0037] Please refer to Figure 4 , Figure 4 is a flowchart of a data processing method based on multi-agent collaboration provided by the embodiments of the present application. As shown in Figure 4 , the method comprises:

[0038] S401, the parent agent calculates a decision weight of each child agent according to user question data.

[0039] The user question data includes text data and behavior data, the text data refers to the text content input by the user, and the behavior data refers to other operation behavior data of the user in addition to the text input operation, such as document upload behavior, picture upload behavior, function selection behavior, etc., which is not limited here.

[0040] The decision weight is used to represent the matching degree of the child agent and the user demand, that is, the parent agent calculates the decision weight of each child agent by quantifying the matching degree of each child agent and the user demand. For example, the user performs the question operation including uploading a document and inputting the text content "summarize the main points of the document", and then the decision weight of each child agent is calculated based on the quantitative evaluation method after intent recognition. It can be understood that, in this example, the matching degree of the document analysis agent and the user demand is higher than that of other child agents, that is, the corresponding decision weight is higher.

[0041] S402, the parent agent allocates a data processing task and a task execution rule to each child agent according to the decision weight.

[0042] The data processing task refers to the specific work allocated by the parent agent to the child agent based on the decision weight and the user demand, which is adapted to the ability of the child agent. For example, the instruction data retrieval agent retrieves the target field from the target data source, and limits the depth and breadth of its retrieval through the decision weight, so as to balance the efficiency and resources of a single child agent when executing the data processing task.

[0043] The task execution rule is used to indicate the constraint condition of the child agent when executing the data processing task. For example, the constraint of the system resources that can be scheduled by the child agent when executing the data processing task depends on the decision weight, so as to configure different task execution rules for each child agent, and improve the overall processing efficiency and reliability of the multi-agent collaboration.

[0044] S403, the plurality of child agents execute the data processing task according to the task execution rule, and return the data processing result to the parent agent.

[0045] S404, the parent agent fuses the data processing result of each child agent to obtain the target reply content.

[0046] The parent agent performs correlation analysis on the data processing results of different child agents, correlates data processing results with logical complementarity, de-duplicates or corrects repeated or conflicting results, organizes the integrated information into text content conforming to natural language expression habits and meeting user needs, and obtains target reply content. On this basis, the parent agent can also adaptively adjust the presentation of the data processing results in the target reply content based on the decision weight, to accurately adapt to the direct needs of the user.

[0047] As can be seen, in the embodiments of the present application, the parent agent calculates the decision weight of each child agent according to the user question data, to quantify the matching degree of the child agent and the user needs, and then allocates data processing tasks and task execution rules to each child agent according to the decision weight, the task execution rules being used to indicate the constraint conditions of the child agent when executing the data processing tasks, the child agent executing the data processing tasks according to the task execution rules and returning the data processing results to the parent agent, and the parent agent fusing the data processing results of each child agent to obtain the target reply content. In this way, the matching degree of the child agent and the user needs is quantified by the decision weight, breaking through the limitation of allocating tasks according to functions in the prior art, improving the matching accuracy of tasks and agent capabilities, and at the same time, configuring differentiated task execution rules to the child agent based on the decision weight, improving the overall processing efficiency and reliability of multi-agent collaboration.

[0048] In one possible example, the parent agent calculates the decision weight of each child agent according to the user question data, including: the parent agent performs word segmentation processing on the user question data to obtain a plurality of keywords; the parent agent performs feature extraction on the user question data to obtain a plurality of demand features; the parent agent determines the corresponding child agent and the score value of each demand feature based on a preset feature matching rule, the feature matching rule including the correspondence relationship of demand features, associated child agents and score values; for each child agent, the plurality of keywords and the keyword library corresponding to the child agent are matched to calculate a first matching score of the child agent; and the second matching score of the child agent is calculated according to the sum of the base score value and the score value of the child agent; and the decision weight of the child agent is calculated according to the first matching score and the second matching score.

[0049] In this example, the parent agent calculates the first matching score of the child agent through keyword matching, calculates the second matching score of the child agent through feature matching, and then quantifies the matching degree of the child agent and the user needs to calculate the decision weight.

[0050] Specifically, in terms of keyword matching, the parent agent calls a pre-trained word segmentation model to perform word segmentation processing on the user question data, to obtain a plurality of keywords with independent semantics. The word segmentation model is trained based on financial corpus, which can avoid splitting special terms in the financial field, and improve the accuracy of word segmentation. Then, keyword matching is performed for each sub-agent, that is, the plurality of keywords are matched with the keyword library corresponding to the sub-agent, to calculate a first matching score. Each sub-agent is configured with a keyword library corresponding to its function, for example, the keyword library corresponding to the data retrieval agent includes index field class keywords such as “net profit, profit margin, asset-liability ratio”, the keyword library corresponding to the literature analysis agent includes literature and analysis class keywords such as “literature, document, annual report, announcement, interpretation, chapter”, and the keyword library corresponding to the knowledge Q&A agent includes knowledge query class keywords such as “definition, meaning, term, explanation”. Further, in calculating the first matching score, firstly, the word frequency of the keywords that are exactly matched with the keyword library of the sub-agent is counted, and for the keywords that are not exactly matched, the semantic similarity between each keyword and the keyword library of the sub-agent is calculated through a pre-trained word vector model, and finally the word frequency and the semantic similarity are weighted and summed and then normalized to obtain the first matching score corresponding to the sub-agent.

[0051] Specifically, in terms of feature matching, the parent agent calls a pre-trained feature extraction model to perform feature extraction on the user question data, to obtain a plurality of demand features. The feature extraction model is trained based on a pre-set rule template, so that the output demand features include the following four categories: interactive behavior features, time element features, and term density features. The interactive behavior features are used to represent the operation behavior of the user, such as literature upload behavior, picture upload behavior, click behavior, etc.; the time element features are used to represent the time identifiers contained in the user input text, such as “latest”, “current”, “2024Q3”, etc.; and the term density features are used to represent the frequency of financial professional terms in the user input text content. Then, based on the pre-set feature matching rules, the sub-agent corresponding to each demand feature and the score value are determined. For example, if there is a literature upload behavior in the interactive behavior features, it is determined that the associated sub-agent is the literature analysis agent, and the score value is a first pre-set value, which can be further distinguished based on the type of uploaded literature; for another example, if the term density in the text content is not less than three, it is determined that the associated sub-agent is the knowledge Q&A agent, and the score value is a second pre-set value, which can be further distinguished based on the actual term density. Further, the base score value of each sub-agent is set, and the sum of the base score value and the score value is normalized to obtain a second matching score.

[0052] Finally, the first matching score and the second matching score are dynamically weighted and fused to obtain a decision weight. Specifically, the decision weight = a x the first matching score + (1-a) x the second matching score, where a is a fusion coefficient that can be dynamically adjusted based on the task type complexity of the user question data. For example, if it is a simple task, such as a pure data query, then a = 0.7, that is, the decision weight is determined based on keyword matching first; if it is a complex task, such as a mixed query analysis combining literature and data, then a = 0.3, that is, the decision weight is determined based on feature matching first, further improving the accuracy.

[0053] As can be seen, in this example, the parent agent calculates the first matching score of each sub-agent through keyword matching, calculates the second matching score of each sub-agent through demand feature matching, and then calculates the decision weight of each sub-agent, comprehensively evaluates the matching degree of each sub-agent with the user demand, provides data support for subsequent task allocation and rule configuration, and improves the overall processing efficiency and result reliability of multi-agent collaboration.

[0054] In one possible example, the task execution rule includes an abnormal situation coping rule and a normal execution rule, the abnormal situation coping rule is used to indicate the constraint condition of the sub-agent when returning the data processing result in the abnormal situation, and the normal execution rule is used to indicate the constraint condition of the sub-agent when not executing the data processing task in the abnormal situation, the abnormal situation refers to a situation that affects the normal execution of the data processing task, and the parent agent allocates the data processing task and the task execution rule to each sub-agent according to the decision weight, including: the parent agent allocates a first data processing task and a second data processing task to the first sub-agent, and allocates the first data processing task to the second sub-agent, the first sub-agent refers to a sub-agent whose decision weight is not less than a first threshold, the second sub-agent refers to a sub-agent whose decision weight is less than the first threshold, the first data processing task refers to a necessary processing step that meets the user demand, and the second data processing task refers to an associated processing step that assists in understanding the reply content; the parent agent allocates a first normal execution rule to the first sub-agent, and allocates a second normal execution rule to the second sub-agent, the constraint strength of the first normal execution rule is higher than that of the second normal execution rule; and the parent agent allocates the abnormal situation coping rule to all sub-agents.

[0055] The parent agent allocates data processing tasks and task execution rules to each child agent according to a decision weight, so as to achieve more accurate demand matching and improve the overall processing efficiency of multi-agent collaboration. In this example, a child agent with a decision weight not less than a first threshold is a first child agent, and a child agent with a decision weight less than the first threshold is a second child agent. The first child agent can perform a deeper exploration when executing a data processing task and is subject to stricter rules to ensure the reliability of the output result compared to the second child agent.

[0056] Specifically, in terms of allocating data processing tasks, the first child agent is allocated a first data processing task and a second data processing task, i.e., the first child agent is instructed to not only perform necessary processing steps to meet user demand, but also perform associated processing steps to assist in understanding the reply content; for example, when the data retrieval agent is the first child agent, it not only needs to retrieve the required index field from the target data source to answer the user's question, but also can further retrieve other data of the same type or at the same time period for comparative analysis of the index field. In addition, the second child agent is allocated a first data processing task, i.e., the second child agent is instructed to only perform necessary processing steps to meet user demand, without performing redundant work unrelated to the necessary processing steps, achieving resource balancing based on demand matching.

[0057] In this example, the task execution rules include an exception handling rule and a normal execution rule to restrict the processing manner of the child agent in the abnormal scenario and the normal scenario, respectively. The abnormal scenario refers to a scenario that affects the normal execution of the data processing task, such as data loss, data error, and other data abnormal scenarios, or system abnormal scenarios such as maintenance failure. The exception handling rule is used to indicate the constraint condition of the child agent when returning the data processing result in the abnormal scenario, so as to ensure the reliability and traceability of the output result as much as possible while avoiding output interruption. The normal execution rule is used to indicate the constraint condition of the child agent when not executing the data processing task in the abnormal scenario, so as to ensure that the reliability of the output result of the child agent is equivalent to the importance of the task it undertakes.

[0058] Specifically, in terms of assigning task execution rules, the first sub-agent is assigned a first normal execution rule, and the second sub-agent is assigned a second normal execution rule, and the constraint strength of the first normal execution rule is higher than that of the second normal execution rule. Wherein, the constraint strength can be embodied in multiple aspects, for example, in terms of resource quota, compared with the second sub-agent, the first sub-agent can have a higher proportion of computing power quota to adapt to the importance of the data processing task it undertakes; for example, in terms of verification strategy, compared with the second sub-agent, the first sub-agent can have a more complex verification strategy to ensure that the output content has higher reliability. And, all sub-agents are assigned abnormal coping rules to give all sub-agents the ability to cope with abnormal scenarios, further improving the overall reliability of the output results.

[0059] It can be seen that in this example, the parent agent assigns different data processing tasks and differentiated task execution rules to each sub-agent based on the decision weight, to adapt the matching degree of the sub-agent to the user demand, so that the sub-agent with functions more suitable for user demand can have more resources under the differentiated rule constraints to perform more in-depth exploration when performing data processing tasks, improving the efficiency and reliability of data processing, while giving all sub-agents the ability to cope with abnormal scenarios, further improving the overall reliability of the output results.

[0060] In one possible example, the normal execution rule includes a verification rule, the verification rule refers to the verification strategy of the sub-agent for its data processing result, and the parent agent assigns the first sub-agent a first normal execution rule, and assigns the second sub-agent a second normal execution rule, including: the parent agent assigns the first sub-agent a first verification rule, the first verification rule is used to instruct the first sub-agent to verify its data processing result from multiple verification dimensions; the parent agent assigns the second sub-agent a second verification rule, the second verification rule is used to instruct the second sub-agent to verify its data processing result from a single verification dimension; wherein, the number of verification dimensions is used to represent the coverage of the sub-agent's risk prevention and control of its data processing result.

[0061] Wherein, in the absence of an abnormal scene, the parent agent assigns differentiated normal execution rules to the first and second sub-agents based on the decision weight to adapt to the importance level of the tasks undertaken by different sub-agents, which includes a verification rule, the verification rule refers to the verification strategy of the sub-agent to its data processing result to improve the reliability of its output result, the more complex the verification strategy is, the higher the reliability of its output result is, therefore, the first sub-agent with a higher decision weight needs to be assigned a verification rule with a relatively complex verification strategy, and the second sub-agent with a lower decision weight needs to be assigned a verification rule with a relatively simple verification strategy to achieve dynamic balance between resources and effects.

[0062] Specifically, in the present example, the parent agent assigns a first verification rule to the first sub-agent, the first verification rule is used to instruct the first sub-agent to verify its data processing result from multiple verification dimensions, and assigns a second verification rule to the second sub-agent, the second verification rule is used to instruct the second sub-agent to verify its data processing result from a single verification dimension, the number of verification dimensions is used to represent the coverage of the sub-agent to the risk prevention and control of its data processing result, that is, the more the number of verification dimensions is, the more comprehensive the risk prevention and control of the data processing result is. Among them, the multiple verification dimensions contained in the first verification rule include but are not limited to: data source authority verification, data logic consistency verification and timeliness verification, etc., that is, to avoid non-authoritative data pollution, to avoid contradictions in data itself, and to avoid misleading of expired data, etc., to comprehensively cover the risks of data processing result in multiple dimensions, while the single verification dimension contained in the second verification rule is a preset basic consistency verification, for example, only needs to verify the consistency of its content source, without verifying whether the content source is authoritative, etc., to simplify the verification strategy of non-core tasks. Further, in addition to the number of verification dimensions, the first verification rule can also contain multi-level verification depth, that is, multi-level progressive verification is adopted, after completing the first-level verification, if the result meets the preset condition, for example, the numerical fluctuation exceeds the reasonable range, then enter the second-level verification, for example, cross-data source cross-verification, until the verification is passed or the abnormality is confirmed, to further improve the reliability of the output content of the first sub-agent.

[0063] As can be seen, in the present example, the parent agent assigns differentiated verification rules to the sub-agents based on the decision weight to adapt to the importance level of the tasks undertaken by different sub-agents, to constrain the sub-agents with high decision weight with more complex verification strategies, and to constrain the sub-agents with low decision weight with simpler verification strategies, to further improve the overall processing efficiency and result reliability of multi-agent collaboration.

[0064] In one possible example, the abnormal scenario includes a data abnormal scenario, the data abnormal scenario refers to an abnormal scenario in which a child agent cannot obtain data from a target data source, and the parent agent allocates the abnormal coping rule to all child agents, including: the parent agent allocates a data abnormal coping rule to all child agents, the data abnormal coping rule is used to instruct the child agent to extract calculated data from an associated data source as a data processing result, and the associated data source refers to a data source that has data association or logical association with the target data source and is in a non-abnormal state.

[0065] When in an abnormal scenario, the child agent cannot normally perform a data processing task, resulting in task interruption or reduced output result reliability. In a data abnormal scenario, the child agent cannot obtain data from a target data source, for example, a target data table abnormality causes a data retrieval agent to be unable to obtain a target field from the data table, or a target knowledge base abnormality causes a knowledge question and answer agent to be unable to obtain a knowledge node from the knowledge base.

[0066] In this example, the abnormal coping rule allocated by the parent agent to all child agents includes a data abnormal coping rule to instruct the child agent to extract calculated data from an associated data source as a data processing result, and the associated data source refers to a data source that has data association or logical association with the target data source and is in a non-abnormal state.

[0067] For example, the target data required by the user is "gross profit rate of Company A in the third quarter of 2024", and the target data source, for example, "financial statements of Company A in the third quarter of 2024", is abnormal and cannot be accessed. According to the data exception handling rule, the data retrieval agent extracts the calculated data from the associated data source as the data processing result. For example, it is determined that the associated data source includes: the financial statements of Company A in the first and second quarters of 2024, which belong to the financial data system of Company A in the same year, have direct data association, and the data state is normal; and also includes: the financial statements of Company A in the third quarter of 2023, which are historical data of the same quarter as the target data source, have logical association, can reflect the fluctuation rule of quarterly gross profit rate, and the data state is normal; and also includes: the financial statements of Company C in the third quarter of 2024, which are similar to the business model and scale of Company A, have industry logical association with the target data source, and the data state is normal. In some embodiments, the data retrieval agent performs the following operations according to the data exception handling rule: first, extracts the gross profit rate data from the financial statements of Company A in the first and second quarters of 2024, calculates the average gross profit rate of the first two quarters of the year; second, extracts the gross profit rate data from the financial statements of Company A in the third quarter of 2023, and combines the historical data to obtain the fluctuation range of the third quarter of Company A compared to the average of the first two quarters; and finally, cross-verify with the gross profit rate of Company C in the third quarter of 2024 to obtain the calculated data of the gross profit rate of Company A in the third quarter of 2024, and output the calculated data as the data processing result, and mark the data as calculated data and associated data source.

[0068] As can be seen, in this example, the parent agent assigns data exception handling rules to all child agents, so that the child agents can still extract calculated data from the associated data source as the data processing result in the data exception scenario, which not only guarantees the continuity of the data processing task, but also ensures the traceability of the result by clearly marking the source and logic, avoiding the problem of task interruption or ambiguous result credibility caused by data loss.

[0069] In one possible example, the parent agent fuses the data processing results of each child agent to obtain the target reply content, including: in response to the absence of the abnormal scenario, calculating the output content proportion of each child agent according to the decision weight; and fusing the data processing results of each child agent according to the output content proportion to obtain the target reply content.

[0070] When there is no abnormal scenario, the sub-agent can normally acquire data, perform data processing tasks according to normal rules, and return data processing results, and finally the parent agent fuses the data processing results to obtain the target reply content. Specifically, the parent agent calculates the output content proportion of the sub-agent according to the decision weight of the sub-agent, and then determines the proportion of each data processing result in the target reply content according to the output content proportion. After extracting the core information and removing the duplicates of the data processing results of each sub-agent, the data processing results are fused according to the allocated proportion, and the target reply content conforming to the natural language expression habit is output.

[0071] It can be seen that in this example, when there is no abnormal scenario, the parent agent calculates the output content proportion of each sub-agent according to the decision weight, and fuses the data processing results of each sub-agent according to the output content proportion to obtain the target reply content. In this way, the core data and key analysis most related to the user's demand can occupy the main proportion in the reply, reducing irrelevant information interference and improving the overall user question and answer experience.

[0072] In one possible example, the parent agent fuses the data processing results of each sub-agent to obtain the target reply content, including: in response to the existence of the abnormal scenario, determining an abnormal sub-agent based on the data processing results, the abnormal sub-agent being a sub-agent affected by the abnormal scenario; and reducing the decision weight of the abnormal sub-agent based on the abnormal influence degree, and increasing the decision weight of other sub-agents by the same proportion; and calculating the output content proportion of each sub-agent according to the adjusted decision weight; and fusing the data processing results of each sub-agent according to the output content proportion to obtain the target reply content.

[0073] When there is an abnormal scenario, the abnormal sub-agent affected by the abnormal scenario returns a data processing result according to an abnormal coping rule, and the parent agent determines the abnormal sub-agent based on an abnormal mark therein. At the same time, due to the decrease in the reliability of the output result of the abnormal sub-agent, the parent agent reduces the decision weight of the abnormal sub-agent based on the abnormal influence degree, and increases the decision weight of other sub-agents by the same proportion, and then calculates the output content proportion according to the adjusted decision weight and fuses it to obtain the target reply content, so as to avoid the proportion of unreliable calculated data in the reply being too high, and reduce the interference degree of the abnormal scenario on the accuracy of the output result.

[0074] The abnormal influence degree is used to represent the interference degree of the abnormal scene on the abnormal sub-agent. Taking a data abnormal scene as an example, based on a preset abnormal judgment rule, the abnormal influence degree can be determined according to the proportion of abnormal data. For example, when the data is completely missing, the abnormal influence degree is serious, and the corresponding weight adjustment range is the largest. When the data is partially missing, the abnormal influence degree is general, and the corresponding weight adjustment range is relatively small.

[0075] It can be seen that in the present example, when there is an abnormal scene, the parent agent reduces the decision weight of the abnormal sub-agent, and increases the decision weight of other sub-agents by the same proportion, thereby avoiding the unreliable calculation data from having too high a proportion in the reply, and making full use of the effective information of the normal sub-agents, thereby improving the fault tolerance of the multi-agent cooperation.

[0076] For the above-mentioned embodiments, please refer to Figure 5 , Figure 5 is a structural block diagram of a data processing device based on multi-agent cooperation provided by the embodiment of the present application. The data processing device based on multi-agent cooperation 50 comprises: a calculation unit 501 configured to make the parent agent calculate the decision weight of each sub-agent according to user question data, wherein the decision weight is used to represent the matching degree of the sub-agent and the user demand; an allocation unit 502 configured to make the parent agent allocate a data processing task and a task execution rule to each sub-agent according to the decision weight, wherein the task execution rule is used to indicate the constraint condition of the sub-agent when executing the data processing task; an execution unit 503 configured to make the plurality of sub-agents execute the data processing task according to the task execution rule, and return the data processing result to the parent agent; and a fusion unit 504 configured to make the parent agent fuse the data processing result of each sub-agent to obtain the target reply content.

[0077] In one possible example, in terms of the calculation unit 501 being configured to make the parent agent calculate the decision weight of each sub-agent according to user question data, the calculation unit 501 is specifically configured to: make the parent agent perform word segmentation processing on the user question data to obtain a plurality of keywords; perform feature extraction on the user question data to obtain a plurality of demand features; determine the sub-agent corresponding to each demand feature and the score value based on a preset feature matching rule, wherein the feature matching rule comprises the corresponding relationship among the demand feature, the associated sub-agent and the score value; for each sub-agent, match the plurality of keywords with the keyword library corresponding to the sub-agent to calculate a first matching score of the sub-agent; calculate a second matching score of the sub-agent according to the base score value and the total score value of the sub-agent; and calculate the decision weight of the sub-agent according to the first matching score and the second matching score.

[0078] In a possible example, the task execution rule includes an exception handling rule and a normal execution rule, the exception handling rule is used to indicate a constraint condition of the sub-agent when returning a data processing result in an exception scenario, and the normal execution rule is used to indicate a constraint condition of the sub-agent when not executing a data processing task in the exception scenario, the exception scenario refers to a scenario that affects the normal execution of the data processing task, and in terms of the allocation of the data processing task and the task execution rule to each sub-agent by the parent agent according to the decision weight, the allocation unit 502 is specifically configured to: cause the parent agent to allocate a first data processing task and a second data processing task to a first sub-agent, and allocate the first data processing task to a second sub-agent, the first sub-agent refers to a sub-agent whose decision weight is not less than a first threshold, the second sub-agent refers to a sub-agent whose decision weight is less than the first threshold, the first data processing task refers to a necessary processing step that meets a user demand, and the second data processing task refers to an associated processing step that assists in understanding the reply content; allocate a first normal execution rule to the first sub-agent, and allocate a second normal execution rule to the second sub-agent, the constraint strength of the first normal execution rule is higher than that of the second normal execution rule; and allocate the exception handling rule to all sub-agents.

[0079] In a possible example, the normal execution rule includes a verification rule, the verification rule refers to a verification strategy of the sub-agent for the data processing result, and in terms of the allocation of the first normal execution rule to the first sub-agent and the allocation of the second normal execution rule to the second sub-agent by the parent agent, the allocation unit 502 is specifically configured to: cause the parent agent to allocate a first verification rule to the first sub-agent, the first verification rule is used to indicate that the first sub-agent verifies the data processing result from multiple verification dimensions; and allocate a second verification rule to the second sub-agent, the second verification rule is used to indicate that the second sub-agent verifies the data processing result from a single verification dimension; and the number of verification dimensions is used to represent the coverage range of the risk prevention and control of the sub-agent for the data processing result.

[0080] In a possible example, the exception scenario includes a data exception scenario, the data exception scenario refers to an exception scenario in which the sub-agent cannot obtain data from a target data source, and in terms of the allocation of the exception handling rule to all sub-agents by the parent agent, the allocation unit 502 is specifically configured to: cause the parent agent to allocate a data exception handling rule to all sub-agents, the data exception handling rule is used to indicate that the sub-agent extracts calculated data from an associated data source as a data processing result, and the associated data source refers to a data source that has data association or logical association with the target data source and is in a non-exception state.

[0081] In one possible example, in terms of fusing the data processing results of each child agent by the parent agent to obtain the target reply content, the fusion unit 504 is specifically configured to: in response to the absence of the abnormal scenario, calculate the output content proportion of each child agent according to the decision weight; and fuse the data processing results of each child agent according to the output content proportion to obtain the target reply content.

[0082] In one possible example, in terms of fusing the data processing results of each child agent by the parent agent to obtain the target reply content, the fusion unit 504 is specifically configured to: in response to the presence of the abnormal scenario, determine an abnormal child agent based on the data processing results, the abnormal child agent being a child agent affected by the abnormal scenario; and reduce the decision weight of the abnormal child agent based on the abnormal influence degree, and increase the decision weight of other child agents by the same proportion; calculate the output content proportion of each child agent according to the adjusted decision weight; and fuse the data processing results of each child agent according to the output content proportion to obtain the target reply content.

[0083] It can be understood that, since the method embodiment and the device embodiment are different presentation forms of the same technical concept, the content of the method embodiment part in the present application should be synchronously adapted to the device embodiment part, which will not be repeated here.

[0084] In the case of using integrated units, as Figure 6 indicated, Figure 6 is another structure block diagram of a data processing apparatus based on multi-agent collaboration provided by the embodiment of the present application, in Figure 6 which the data processing apparatus based on multi-agent collaboration 50 comprises a processing module 52 and a communication module 51. The processing module 52 is configured to control and manage the actions of the data processing apparatus based on multi-agent collaboration, for example, to execute the steps of the computing unit 501, the distribution unit 502, the execution unit 503 and the fusion unit 504, and / or to execute other processes of the technology described herein. The communication module 51 is configured to support the interaction between the data processing apparatus based on multi-agent collaboration and other devices. As Figure 6 indicated, the data processing apparatus based on multi-agent collaboration can further comprise a storage module 53, which is configured to store the program code and data of the data processing apparatus based on multi-agent collaboration.

[0085] Wherein all the related content of each scenario involved in the above method embodiment can be cited to the function description of the corresponding function module, which will not be repeated here. The above data processing apparatus based on multi-agent collaboration 50 can execute the data processing method based on multi-agent collaboration as Figure 4 indicated.

[0086] Based on the description of the method embodiments and the device embodiments above, please refer to Figure 7 , Figure 7 is a structural schematic diagram of an electronic device provided in an embodiment of the present application. Figure 7 The electronic device shown in the figure includes a memory 701, a processor 702, a communication interface 703, and a bus 704. Among them, the memory 701, the processor 702, and the communication interface 703 are communicatively connected to each other through the bus 704. Among them, the electronic device can specifically refer to the first electronic device 11 or the second electronic device 12 in the above embodiments.

[0087] The memory 701 can be a read-only memory (ROM), a static storage device, a dynamic storage device, or a random access memory (RAM).

[0088] The memory 701 can store a program, and when the program stored in the memory 701 is executed by the processor 702, the processor 702 and the communication interface 703 are used to execute each step of the data processing method based on multi-agent collaboration of the embodiments of the present application.

[0089] The processor 702 can adopt a general central processing unit (CPU), a microprocessor, an application specific integrated circuit (ASIC), a graphics processing unit (GPU), or one or more integrated circuits, to execute related programs to implement the functions required by the units in the electronic device of the embodiments of the present application, or to execute the data processing method based on multi-agent collaboration of the method embodiments of the present application.

[0090] The processor 702 can also be an integrated circuit chip having a processing capability for signals. In implementation, each step of the data processing method based on multi-agent collaboration of the present application can be completed by integrated logic circuits of hardware or instructions in the form of software in the processor 702. The processor 702 described above can also be a general processor, a digital signal processor (DSP), an application specific integrated circuit (ASIC), a ready programmable gate array (FPGA) or other programmable logic device, a discrete gate or transistor logic device, a discrete hardware component. The disclosed methods, steps and logic block diagrams in the embodiments of the present application can be implemented or executed. The general processor can be a microprocessor or the processor can also be any conventional processor. The steps of the method disclosed in combination with the embodiments of the present application can be directly embodied as a hardware decoding processor for execution, or a combination of hardware and software modules in the decoding processor for execution. The software module can be located in a random access memory, a flash memory, a read only memory, a programmable read only memory or an electrically erasable programmable memory, a register or other mature storage medium in the art. The storage medium is located in the memory 701, and the processor 702 reads the information in the memory 701, and combines the hardware to complete the functions required by the units included in the electronic device in the embodiments of the present application, or executes the data processing method based on multi-agent collaboration of the method embodiments of the present application.

[0091] The communication interface 703 uses a transceiving device such as but not limited to a transceiver to realize the communication between the electronic device and other devices or communication networks. For example, data can be obtained through the communication interface 703.

[0092] The bus 704 can include a path for transmitting information between various components (for example, the memory 701, the processor 702, the communication interface 703) of the electronic device.

[0093] It should be noted that although Figure 7 The electronic device shown only shows the memory 701, the processor 702, the communication interface 703, but in the specific implementation process, those skilled in the art should understand that the electronic device also includes other devices necessary for normal operation. At the same time, according to the specific needs, those skilled in the art should understand that the electronic device can also include hardware devices for realizing other additional functions. In addition, those skilled in the art should understand that the electronic device can also only include devices necessary for the embodiments of the present application, and does not have to include all the devices shown in the Figure 7

[0094] ​The embodiments of the present application further provide a computer storage medium, wherein a computer program / instruction is stored on the computer storage medium, and the computer program / instruction is executed by a processor to implement some or all steps of any method described in the above method embodiments.

[0095] In several embodiments provided in the present application, it should be understood that the disclosed methods and devices can be implemented in other manners. For example, the above-described device embodiments are merely illustrative; for example, the division of the units is only a logical function division, and there can be another division manner in actual implementation; for example, a plurality of units or components can be combined or integrated into another system, or some features can be ignored or not executed. In addition, the displayed or discussed mutual couplings or direct couplings or communication connections between different units, can be indirect couplings or communication connections through some interfaces, devices or units, and can be electrical, mechanical or in other forms.

[0096] The units described as separated components can or can not be physically separated, and the components displayed as units can or can not be physical units, i.e., can be located in one place, or can be distributed on a plurality of network units. Some or all of the units can be selected according to actual needs to achieve the purposes of the embodiments of the present application.

[0097] In the above embodiments, all or part of the embodiments can be implemented by software, hardware, firmware or any combination thereof. When implemented by software, all or part of the embodiments can be implemented in the form of a computer program product. The computer program product includes one or more computer instructions. When loaded and executed by a computer, all or part of the computer program instructions generate processes or functions according to the embodiments of the present application. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable devices. The computer instructions can be stored in a computer-readable storage medium or transmitted by the computer-readable storage medium. The computer instructions can be transmitted from one website, computer, electronic device or data center to another website, computer, electronic device or data center by wire (such as coaxial cable, optical fiber, digital subscriber line) or wireless (such as infrared, wireless, microwave, etc.). The computer-readable storage medium can be any available medium that can be accessed by a computer or data storage device such as an electronic device, data center, etc. integrated with one or more available media. The available medium can be a read-only memory, a random access memory, a magnetic medium, such as a floppy disk, a hard disk, a magnetic tape, a magnetic disk, or an optical medium, such as a digital versatile disc, or a semiconductor medium, such as a solid-state disk, etc.

[0098] The above merely describes the specific implementation of the embodiments of the present application, but the protection scope of the embodiments of the present application is not limited thereto, any change or replacement within the technical scope disclosed by the embodiments of the present application should be covered in the protection scope of the embodiments of the present application. Therefore, the protection scope of the embodiments of the present application should be subject to the protection scope of the claims.

[0099] The apparatus embodiments described above are merely illustrative, wherein the units and modules described as separate components can or can not be physically separated. In addition, part or all of the units and modules can be selected to achieve the purpose of the embodiments of the present application according to actual needs. Those skilled in the art can understand and implement without creative labor.

[0100] Although the present application is disclosed as above, the present application is not limited thereto. Any person skilled in the art can easily think of changes or replacements without departing from the spirit and scope of the present application, and can make various changes and modifications, including the combination of different functions and implementation steps, including the software and hardware implementation, which are all within the protection scope of the present application.

Claims

1. A data processing method based on multi-agent cooperation, characterized in that, The method is applied to an electronic device including a parent agent and a plurality of child agents, and the method includes: The parent agent calculates a decision weight of each child agent according to user question data, and the decision weight is used to represent a matching degree of the child agent and a user demand; The parent agent allocates a data processing task and a task execution rule to each child agent according to the decision weight, the task execution rule includes an abnormality coping rule and a normal execution rule, the abnormality coping rule is used to indicate a constraint condition of the child agent when returning a data processing result in an abnormal scenario, and the abnormal scenario refers to a scenario affecting normal execution of the data processing task; the normal execution rule includes a verification rule, and the verification rule refers to a verification strategy of the child agent for the data processing result; the abnormality coping rule includes a data abnormality coping rule, and the data abnormality coping rule is used to indicate that the child agent extracts calculated data from an associated data source as the data processing result, and the associated data source refers to a data source having data association or logical association with a target data source and being in a non-abnormal state; the parent agent allocates a first data processing task and a second data processing task to a first child agent and allocates the first data processing task to a second child agent, the first child agent refers to a child agent with a decision weight not less than a first threshold, the second child agent refers to a child agent with a decision weight less than the first threshold, the first data processing task refers to a necessary processing step meeting a user demand, and the second data processing task refers to an associated processing step assisting in understanding reply content; the parent agent allocates a first verification rule to the first child agent, and the first verification rule is used to indicate that the first child agent verifies the data processing result from a plurality of verification dimensions; the parent agent allocates a second verification rule to the second child agent, and the second verification rule is used to indicate that the second child agent verifies the data processing result from a single verification dimension; wherein, the number of verification dimensions is used to represent a coverage range of the child agent for risk prevention and control of the data processing result; and the parent agent allocates the data abnormality coping rule to all child agents; The plurality of child agents execute the data processing task according to the task execution rule and return a data processing result to the parent agent; The parent agent determines an abnormal child agent based on the data processing result in response to the existence of the abnormal scenario, the abnormal child agent refers to a child agent affected by the abnormal scenario; and reduces the decision weight of the abnormal child agent based on an abnormal influence degree and proportionally increases the decision weight of other child agents; and calculates an output content proportion of each child agent according to the adjusted decision weight; and fuses the data processing result of each child agent according to the output content proportion to obtain target reply content.

2. The method of claim 1, wherein, The parent agent calculates a decision weight of each child agent according to user question data, including: The parent agent performs word segmentation processing on the user question data to obtain a plurality of keywords; The parent intelligent agent extracts features from the user question data to obtain a plurality of requirement features; The parent intelligent agent determines a sub-intelligent agent corresponding to each requirement feature and a score value based on a preset feature matching rule, and the feature matching rule includes a corresponding relationship among requirement features, associated sub-intelligent agents, and score values; For each sub-intelligent agent, the plurality of keywords and the keyword library corresponding to the sub-intelligent agent are matched to calculate a first matching score of the sub-intelligent agent; and a second matching score of the sub-intelligent agent is calculated according to a base score value and a total score value of the sub-intelligent agent; and a decision weight of the sub-intelligent agent is calculated according to the first matching score and the second matching score.

3. The method of claim 1, wherein, The method further includes: The parent intelligent agent calculates an output content proportion of each sub-intelligent agent according to the decision weight in response to the absence of the abnormal scene; and data processing results of each sub-intelligent agent are fused to obtain target reply content according to the output content proportion.

4. A data processing apparatus based on multi-agent collaboration, characterized by, The device is applied to an electronic device, and the electronic device includes a parent intelligent agent and a plurality of sub-intelligent agents, and the device includes: A calculation unit configured to cause the parent intelligent agent to calculate a decision weight of each sub-intelligent agent according to user question data, and the decision weight is used to represent a matching degree of the sub-intelligent agent and user requirements. an allocation unit configured to cause the parent agent to allocate data processing tasks and task execution rules to each child agent according to the decision weight, the task execution rules comprising an abnormality coping rule and a normal execution rule, the abnormality coping rule being used to indicate a constraint condition of the child agent returning a data processing result in an abnormal scenario, the abnormal scenario being a scenario affecting normal execution of the data processing task, the normal execution rule comprising a verification rule, the verification rule being a verification strategy of the child agent on the data processing result thereof, the abnormality coping rule comprising a data abnormality coping rule, the data abnormality coping rule being used to instruct the child agent to extract calculated data from an associated data source as the data processing result, the associated data source being a data source having data association or logical association with a target data source and being in a non-abnormal state, the parent agent allocating a first data processing task and a second data processing task to a first child agent and allocating the first data processing task to a second child agent, the first child agent being a child agent with a decision weight not less than a first threshold, the second child agent being a child agent with a decision weight less than the first threshold, the first data processing task being a necessary processing step meeting a user demand, the second data processing task being an associated processing step assisting in understanding reply content, the parent agent allocating a first verification rule to the first child agent, the first verification rule being used to instruct the first child agent to verify the data processing result thereof from multiple verification dimensions, the parent agent allocating a second verification rule to the second child agent, the second verification rule being used to instruct the second child agent to verify the data processing result thereof from a single verification dimension, wherein the number of verification dimensions represents a coverage range of risk prevention and control of the child agent on the data processing result thereof, and the parent agent allocating the data abnormality coping rule to all child agents; an execution unit configured to cause the multiple child agents to execute the data processing tasks according to the task execution rules and return data processing results to the parent agent; a fusion unit configured to cause the parent agent to, in response to the existence of the abnormal scenario, determine an abnormal child agent based on the data processing results, the abnormal child agent being a child agent affected by the abnormal scenario, and reduce the decision weight of the abnormal child agent and proportionally increase the decision weight of other child agents based on an abnormal influence degree, and calculate an output content proportion of each child agent according to the adjusted decision weight, and fuse the data processing results of each child agent according to the output content proportion to obtain target reply content.

5. An electronic device, comprising: A computer program comprising a processor, a memory and one or more programs stored in the memory and configured to be executed by the processor, the programs comprising instructions for performing the steps of the method of any one of claims 1-3.

6. A computer-readable storage medium having stored thereon a computer program, characterized in that, The computer program, when executed by the processor, implements the steps of the method of any one of claims 1-3.

Images