Data processing method, multi-agent processing system

Abstract

The embodiment of the specification provides a data processing method and a multi-agent processing system. The method comprises the following steps: determining a target analysis task, generating a target user profile corresponding to the target analysis task by using a user agent; generating a task execution plan corresponding to the target analysis task by using a management agent according to the target analysis task, task data and the target user profile; generating a verification analysis plan and a verification subtask corresponding to the verification analysis plan according to the task execution plan by using the management agent and an analysis model; executing the verification subtask corresponding to the verification analysis plan by using the management agent, obtaining a task execution result, and generating a task analysis report corresponding to the target analysis task according to the task execution result. Through the interaction and cooperation between the multi-agents, the analysis task can be autonomously planned, the attribution hypothesis can be dynamically generated and verified when the target analysis task is processed, and a high-value task analysis report can be generated.

Description

Technical Field

[0001] The embodiments in this specification relate to the field of artificial intelligence technology, and in particular to a data processing method and a multi-agent processing system. Background Technology

[0002] In the modern business environment, companies rely heavily on key performance indicators (KPIs) such as sales conversion rate and customer satisfaction to evaluate performance and guide decision-making. However, accurately understanding the core drivers behind the fluctuations in these KPIs (i.e., attribution analysis) remains a significant challenge.

[0003] Traditional attribution analysis methods often rely on pre-set rules, manual assumptions, or single-dimensional statistics, making it difficult to capture the dynamic impact and interactions of numerous potential factors in complex project scenarios (such as multi-turn dialogue interactions). This ambiguity in attribution analysis directly leads to insufficient decision-making basis, inappropriate resource allocation, and a lack of targeted optimization strategies, ultimately hindering the effective enhancement of project value. Especially in dialogue-intensive scenarios involving massive amounts of unstructured dialogue text, traditional attribution methods struggle to systematically and automatically uncover the deep, interconnected factors affecting key project indicators, often resulting in one-sided or superficial analysis. Summary of the Invention

[0004] In view of the above, embodiments of this specification provide a data processing method. One or more embodiments of this specification also relate to a data processing apparatus, a data processing method for a dialogue analysis task, a data processing apparatus for a dialogue analysis task, a multi-agent processing system, a computing device, an electronic device, a computer-readable storage medium, and a computer program product, to address the technical deficiencies existing in the prior art.

[0005] According to a first aspect of the embodiments of this specification, a data processing method is provided, comprising:

[0006] A target analysis task is determined, and a target user profile corresponding to the target analysis task is generated using a user intelligent agent, wherein the target analysis task carries corresponding task data;

[0007] Based on the target analysis task, the task data, and the target user profile, a management agent is used to generate a task execution plan corresponding to the target analysis task.

[0008] Using the management agent and the analysis model, a verification analysis plan and a corresponding verification subtask are generated according to the task execution plan, wherein the analysis model is determined by the management agent based on the analysis results of the task execution plan;

[0009] Using the management agent, the verification sub-tasks corresponding to the verification analysis plan are executed, the task execution results are obtained, and a task analysis report corresponding to the target analysis task is generated based on the task execution results.

[0010] According to a second aspect of the embodiments of this specification, a data processing apparatus is provided, comprising:

[0011] The profile generation module is configured to determine the target analysis task and generate the target user profile corresponding to the target analysis task using the user intelligent agent, wherein the target analysis task carries the corresponding task data;

[0012] The plan generation module is configured to generate a task execution plan corresponding to the target analysis task based on the target analysis task, the task data, and the target user profile, using a management agent.

[0013] The subtask generation module is configured to use the management agent and the analysis model to generate a verification analysis plan and a verification subtask corresponding to the verification analysis plan according to the task execution plan, wherein the analysis model is determined by the management agent based on the analysis results of the task execution plan;

[0014] The report generation module is configured to use the management agent to execute the verification sub-tasks corresponding to the verification analysis plan, obtain the task execution results, and generate a task analysis report corresponding to the target analysis task based on the task execution results.

[0015] According to a third aspect of the embodiments of this specification, a data processing method for a dialogue analysis task is provided, comprising:

[0016] A target dialogue analysis task is determined, and a target user profile corresponding to the target dialogue analysis task is generated using a user intelligent agent, wherein the target dialogue analysis task carries corresponding dialogue analysis data.

[0017] Based on the target dialogue analysis task, the dialogue analysis data, and the target user profile, a management agent is used to generate a dialogue analysis task execution plan corresponding to the target dialogue analysis task.

[0018] Using the management agent and the analysis model, a verification analysis plan and a corresponding verification sub-task are generated according to the dialogue analysis task execution plan. The analysis model is determined by the management agent based on the parsing results of the dialogue analysis task execution plan.

[0019] Using the management agent, the verification sub-tasks corresponding to the verification analysis plan are executed, the task execution results are obtained, and a task analysis report corresponding to the target dialogue analysis task is generated based on the task execution results.

[0020] According to a fourth aspect of the embodiments of this specification, a data processing apparatus for a dialogue analysis task is provided, comprising:

[0021] The first profile generation module is configured to determine the target dialogue analysis task and generate a target user profile corresponding to the target dialogue analysis task using a user agent, wherein the target dialogue analysis task carries corresponding dialogue analysis data.

[0022] The first plan generation module is configured to generate a dialogue analysis task execution plan corresponding to the target dialogue analysis task based on the target dialogue analysis task, the dialogue analysis data, and the target user profile, using a management agent.

[0023] The first subtask generation module is configured to use the management agent and the analysis model to generate a verification analysis plan and a verification subtask corresponding to the verification analysis plan according to the dialogue analysis task execution plan. The analysis model is determined by the management agent based on the parsing results of the dialogue analysis task execution plan.

[0024] The first report generation module is configured to use the management agent to execute the verification sub-tasks corresponding to the verification analysis plan, obtain the task execution results, and generate a task analysis report corresponding to the target dialogue analysis task based on the task execution results.

[0025] According to a fifth aspect of the embodiments of this specification, a multi-agent processing system is provided, including a user agent and a management agent, wherein:

[0026] The user intelligent agent generates a target user profile corresponding to the target analysis task, wherein the target analysis task carries corresponding task data;

[0027] The management agent generates a task execution plan corresponding to the target analysis task based on the target analysis task, the task data, and the target user profile.

[0028] The management agent determines an analysis model and generates a verification analysis plan and corresponding verification subtasks based on the task execution plan. The analysis model is determined by the management agent based on the analysis results of the task execution plan.

[0029] The management agent executes the verification sub-tasks corresponding to the verification analysis plan, obtains the task execution results, and generates a task analysis report corresponding to the target analysis task based on the task execution results.

[0030] According to a sixth aspect of the embodiments of this specification, a computing device is provided, comprising:

[0031] Memory and processor;

[0032] The memory is used to store computer programs / instructions, and the processor is used to execute the computer programs / instructions. When the computer programs / instructions are executed by the processor, they implement the steps of the above-described data processing method or the data processing method applied to the dialogue analysis task.

[0033] According to a seventh aspect of the embodiments of this specification, an electronic device is provided, comprising:

[0034] A memory and a processor, the memory and the processor being connected via a bus;

[0035] The memory is used to store computer programs / instructions, and the processor is used to execute the computer programs / instructions. When the computer programs / instructions are executed by the processor, they implement the steps of the above-described data processing method or the data processing method applied to the dialogue analysis task.

[0036] According to an eighth aspect of the embodiments of this specification, a computer-readable storage medium is provided that stores a computer program / instructions that, when executed by a processor, implement the steps of the above-described data processing method or a data processing method applied to a dialogue analysis task.

[0037] According to a ninth aspect of the embodiments of this specification, a computer program product is provided, including a computer program / instructions that, when executed by a processor, implement the steps of the above-described data processing method or a data processing method applied to a dialogue analysis task.

[0038] One embodiment of this specification implements a data processing method, comprising: determining a target analysis task; generating a target user profile corresponding to the target analysis task using a user agent, wherein the target analysis task carries corresponding task data; generating a task execution plan corresponding to the target analysis task using a management agent based on the target analysis task, the task data, and the target user profile; generating a verification analysis plan and a verification sub-task corresponding to the verification analysis plan using the management agent and an analysis model, wherein the analysis model is determined by the management agent based on the parsing results of the task execution plan; executing the verification sub-task corresponding to the verification analysis plan using the management agent, obtaining task execution results, and generating a task analysis report corresponding to the target analysis task based on the task execution results.

[0039] Specifically, this data processing method first automatically parses the target analysis task when generating target user profiles corresponding to the target analysis task, identifying user intent, preferences, and behavioral patterns, and dynamically generating accurate target user profiles. This reduces analysis costs while improving the efficiency and accuracy of target user profile generation. Second, a management agent automatically plans and generates executable task execution plans based on the target analysis task, task data, and target user profiles. Further, combined with the analysis model, it dynamically generates verification analysis plans (i.e., attribution hypotheses) and corresponding verification sub-tasks to verify these plans, enabling in-depth data understanding. Then, the management agent executes the verification sub-tasks corresponding to the verification analysis plans, generating task execution results and task analysis reports. This allows for understanding the reasons and mechanisms behind the data, identifying causal relationships between data, and upgrading attribution analysis from descriptive statistics to executable causal decision-making, significantly improving the depth of attribution analysis and revealing the root cause chains affecting key project indicators in the target analysis task. This data processing method, through the interaction and collaboration between user agents and management agents, enables a deep attribution analysis system that can autonomously plan analysis tasks (task execution plans), dynamically generate attribution hypotheses, and verify attribution hypotheses. When processing target analysis tasks according to this deep attribution analysis system, it systematically and automatically mines deep and related factors affecting key project indicators, significantly improving the depth of attribution analysis, revealing the root cause chain, and thus providing clear, actionable, and high-value task analysis reports. Attached Figure Description

[0040] Figure 1 This is a schematic diagram illustrating an application scenario of a data processing method provided in one embodiment of this specification;

[0041] Figure 2This is a flowchart illustrating a data processing method provided in one embodiment of this specification;

[0042] Figure 3 This is a flowchart illustrating the processing procedure of a data processing method provided in one embodiment of this specification.

[0043] Figure 4 This is a schematic diagram of the structure of a memory unit in a data processing method provided in one embodiment of this specification;

[0044] Figure 5 This is a schematic diagram of the structure of a data processing apparatus provided in one embodiment of this specification;

[0045] Figure 6 This is a structural block diagram of a computing device provided in one embodiment of this specification;

[0046] Figure 7 This is a structural block diagram of an electronic device provided in one embodiment of this specification. Detailed Implementation

[0047] Many specific details are set forth in the following description to provide a full understanding of this specification. However, this specification can be implemented in many other ways than those described herein, and those skilled in the art can make similar extensions without departing from the spirit of this specification. Therefore, this specification is not limited to the specific implementations disclosed below.

[0048] The terminology used in one or more embodiments of this specification is for the purpose of describing particular embodiments only and is not intended to be limiting of the one or more embodiments of this specification. The singular forms “a,” “described,” and “the” as used in one or more embodiments of this specification and the appended claims are also intended to include the plural forms unless the context clearly indicates otherwise. It should also be understood that the term “and / or” as used in one or more embodiments of this specification refers to and includes any or all possible combinations of one or more associated listed items.

[0049] It should be understood that although the terms first, second, etc., may be used to describe various information in one or more embodiments of this specification, such information should not be limited to these terms. These terms are only used to distinguish information of the same type from one another. For example, first may also be referred to as second without departing from the scope of one or more embodiments of this specification, and similarly, second may also be referred to as first. Depending on the context, the word "if" as used herein may be interpreted as "when," "when," or "in response to a determination."

[0050] Furthermore, it should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved in one or more embodiments of this specification are all information and data authorized by the user or fully authorized by all parties. Moreover, the collection, use and processing of related data must comply with the relevant laws, regulations and standards of the relevant countries and regions, and corresponding operation entry points are provided for users to choose to authorize or refuse.

[0051] In one or more embodiments of this specification, a large model refers to a deep learning model with a large number of model parameters, typically containing hundreds of millions, tens of billions, hundreds of billions, trillions, or even tens of trillions of model parameters. A large model can also be called a foundation model. It is pre-trained using large-scale unlabeled corpora to produce a pre-trained model with hundreds of millions of parameters. Such models can adapt to a wide range of downstream tasks and have good generalization ability. Examples include Large Language Models (LLMs) and multi-modal pre-training models.

[0052] In practical applications, large models only require a small number of samples to fine-tune the pre-trained model before they can be applied to different tasks. Large models can be widely used in fields such as Natural Language Processing (NLP) and Computer Vision. Specifically, they can be applied to computer vision tasks such as Visual Question Answering (VQA), Image Captioning (IC), and Image Generation, as well as natural language processing tasks such as text-based sentiment classification, text summarization, and machine translation. The main application scenarios of large models include digital assistants, intelligent robots, search, online education, office software, e-commerce, and intelligent design.

[0053] First, the terms and concepts used in one or more embodiments of this specification will be explained.

[0054] LLM: Natural Language Processing Model based on large-scale parameters and pre-training techniques, capable of generating text, reasoning, and performing complex tasks.

[0055] Multi-Agent: A system consisting of multiple intelligent agents (agents) working together, where agents interact and make decisions to handle complex tasks; it can also be understood as a multi-agent system that coordinates the behavior and knowledge of multiple autonomous and interactive agents to solve complex or dynamic tasks that are difficult for a single system to handle effectively.

[0056] AgentMemory: A module used by intelligent agents to store short-term interaction records and long-term knowledge and experience, supporting task reasoning and decision-making; it can also be understood as an agent storage module, which is the core module inside the agent used to hierarchically store short-term interaction records and long-term knowledge and experience, and provides support for the agent's real-time task reasoning, decision planning and environmental adaptation by dynamically integrating historical data.

[0057] AgentPlanner: A module used by intelligent agents for planning and generating execution plans for complex tasks; it can also be understood as an agent planning module or an execution plan generation module, etc.

[0058] Attribution analysis: The process of determining the causes of a behavior or event, primarily used to analyze key factors that affect project metrics.

[0059] To address the aforementioned technical problems, a data processing method is provided in one or more embodiments of this specification. One or more embodiments of this specification also relate to a data processing apparatus, a data processing method for a dialogue analysis task, a data processing apparatus for a dialogue analysis task, a multi-agent processing system, a computing device, an electronic device, a computer-readable storage medium, and a computer program product, which will be described in detail in the following embodiments.

[0060] See Figure 1 , Figure 1 This is a schematic diagram illustrating an application scenario of a data processing method provided in one embodiment of this specification.

[0061] Considering the large number of model parameters in large models and the limited computing resources of mobile terminals, the data processing methods provided in one or more embodiments of this specification can be applied to, for example... Figure 1 The application scenarios shown are not limited to these. See also Figure 1 , Figure 1 This is a schematic diagram illustrating an application scenario of a data processing method provided in one embodiment of this specification. Figure 1 In the application scenario shown, the large model is deployed on server 10. Server 10 can connect to one or more client devices 20 via a local area network (LAN), wide area network (WAN), internet connection, or other types of data network. These client devices 20 may include, but are not limited to, smartphones, tablets, laptops, PDAs, personal computers, smart home devices, and in-vehicle devices. Client devices 20 can interact with users through a graphical user interface to access the large model, thereby implementing the methods provided in the embodiments of this specification.

[0062] In the embodiments described in this specification, the system consisting of a client device and a server can perform the following steps: The client device performs the following steps:

[0063] Receive the target analysis task and the corresponding task data sent by the user through the product interaction page of the client device, and send the target analysis task and the corresponding task data to the server.

[0064] The server executes a data processing method provided in one embodiment of this specification, utilizing multi-agent collaboration to perform the target analysis task and generate a task analysis report corresponding to the target analysis task. The specific implementation steps are as follows:

[0065] A target analysis task is determined, and a target user profile corresponding to the target analysis task is generated using a user intelligent agent, wherein the target analysis task carries corresponding task data;

[0066] Based on the target analysis task, the task data, and the target user profile, a management agent is used to generate a task execution plan corresponding to the target analysis task.

[0067] Using the management agent and the analysis model, a verification analysis plan and a corresponding verification subtask are generated according to the task execution plan, wherein the analysis model is determined by the management agent based on the analysis results of the task execution plan;

[0068] Using the management agent, the verification sub-tasks corresponding to the verification analysis plan are executed, the task execution results are obtained, and a task analysis report corresponding to the target analysis task is generated based on the task execution results.

[0069] Specifically, both user agents and management agents can be understood as agents that rely on large models (such as the LLM model). The user agent can be understood as the User Agent, which possesses the user's own experience and knowledge, as well as a tool for interacting with the user. The management agent can be understood as the Manager Agent, which is used to generate task execution plans, execute verification subtasks, and generate task analysis reports.

[0070] It should be noted that, provided that the client device's operating resources can meet the deployment and operation conditions of the large model, one or more embodiments of this specification can be implemented on the client device.

[0071] The data processing method provided in one or more embodiments of this specification firstly, by automatically parsing the target analysis task and identifying user intent, preferences, and behavioral patterns when generating target user profiles corresponding to the target analysis task, it dynamically generates accurate target user profiles, reducing analysis costs while improving the efficiency and accuracy of target user profile generation. Secondly, a management agent can be used to automatically plan and generate an executable task execution plan corresponding to the target analysis task based on the target analysis task, task data, and target user profiles. Further, combined with the analysis model, a verification analysis plan (i.e., attribution hypothesis) and verification sub-tasks corresponding to the verification analysis plan are dynamically generated for in-depth data understanding. Then, the management agent executes the verification sub-tasks corresponding to the verification analysis plan, generating task execution results and a task analysis report. This enables understanding the causes and mechanisms behind the data, identifying causal relationships between data, and upgrading attribution analysis from descriptive statistics to executable causal decision-making, significantly improving the depth of attribution analysis, thereby revealing the root cause chain affecting key project indicators in the target analysis task. This data processing method, through the interaction and collaboration between user agents and management agents, enables a deep attribution analysis system that can autonomously plan analysis tasks (task execution plans), dynamically generate attribution hypotheses, and verify attribution hypotheses. When processing target analysis tasks according to this deep attribution analysis system, it systematically and automatically mines deep and related factors affecting key project indicators, significantly improving the depth of attribution analysis, revealing the root cause chain, and thus providing clear, actionable, and high-value task analysis reports.

[0072] See Figure 2 , Figure 2 A flowchart of a data processing method according to an embodiment of this specification is shown, which specifically includes the following steps.

[0073] Step 202: Determine the target analysis task and use the user agent to generate the target user profile corresponding to the target analysis task.

[0074] The target analysis task carries corresponding task data.

[0075] The data processing method provided in one or more embodiments of this specification can be applied to dialogue analysis scenarios in various fields, such as customer service, sales, finance, and education. Of course, it is not limited to dialogue analysis scenarios in these fields; it can also be adapted to scenarios involving massive amounts of unstructured text analysis. For ease of understanding, the data processing method provided in one or more embodiments of this specification is described in detail using a dialogue analysis scenario in the customer service field as an example.

[0076] Specifically, the target analysis tasks vary depending on the field in which this data processing method is applied. For example, if this data processing method is applied to a dialogue analysis scenario in the customer service field, then the target analysis tasks could be customer service satisfaction, complaint rate, and other similar tasks. If this data processing method is applied to a dialogue analysis scenario in the financial field, then the target analysis tasks could be delinquency rate, fraud rate, and other similar tasks. If this data processing method is applied to a dialogue analysis scenario in the education field, then the target analysis tasks could be error rate, learning efficiency, and other similar tasks.

[0077] In practical applications, user agents can be used to generate target user profiles corresponding to target analysis tasks through multi-turn dialogues. For example, if the target analysis task is to perform attribution analysis on recent satisfaction indicators, the user agent can generate target user profiles corresponding to the target analysis task by automatically parsing the multi-turn dialogues, identifying user intent, and generating target user profiles corresponding to the target analysis task. In other words, the target user profile is a collection of information collected and organized around the user's target analysis task, used to determine the task execution parameters of the target analysis task, such as the data source corresponding to the target analysis task, the key indicators to be analyzed, and the display format of the task analysis report.

[0078] To quickly generate target user profiles corresponding to the target analysis task, avoid repeatedly confirming known information, and reduce the cost of repetitive interactions, historical analysis tasks corresponding to the target analysis task can be introduced when generating target user profiles. This allows the user agent to extract reusable experience information from the historical user profiles of historical analysis tasks based on the task requirements of the target analysis task, thus quickly and accurately generating target user profiles. The specific implementation method is as follows:

[0079] The step of determining the target analysis task and generating the target user profile corresponding to the target analysis task using a user agent includes:

[0080] Receive the target analysis task and the task data corresponding to the target analysis task;

[0081] The user intelligent agent is used to parse the target analysis task and determine the task type of the target analysis task;

[0082] Using the user agent, historical analysis tasks matching the task type of the target analysis task are selected from the memory unit according to the task type of the target analysis task;

[0083] Based on the target analysis task and the historical analysis task, a target user profile corresponding to the target analysis task is generated using a user agent.

[0084] In practical applications, before generating the target user profile corresponding to the target analysis task based on the target analysis task and historical analysis tasks using the user intelligent agent, the user intelligent agent receives the target analysis task and its corresponding task data uploaded by the user through the product interaction page provided by the user. Through parsing the target analysis task, and based on the task type determined by the parsing, it selects a historical analysis task matching the target analysis task from the memory unit. That is, before generating the target user profile corresponding to the target analysis task using the user intelligent agent based on the target analysis task and historical analysis tasks, it receives the target analysis task and its corresponding task data; it uses the user intelligent agent to parse the target analysis task to determine its task type; it uses the user intelligent agent to select a historical analysis task matching the task type of the target analysis task from the memory unit; and finally, it generates the target user profile corresponding to the target analysis task based on the historical analysis tasks.

[0085] The target analysis tasks vary depending on the application scenario of this data processing method. For ease of understanding, we will use the dialogue analysis scenario in the customer service field as an example, and introduce the target analysis tasks such as customer service satisfaction and complaint rate in detail. For example, the target analysis task is: Help me perform attribution analysis on recent satisfaction indicators. Then, the task data corresponding to the target analysis task can be understood as any format (such as .xlsx, .csv, etc.), the original data file or document of the satisfaction survey, etc., which includes, but is not limited to, user identifiers, customer identifiers, dates, or user attributes.

[0086] Therefore, if the objective analysis task is to help me perform attribution analysis on recent satisfaction indicators, the task type of the objective analysis task can be understood as: satisfaction attribution analysis, while the historical analysis task is the same as the objective analysis task: satisfaction attribution analysis.

[0087] In practice, the user agent first receives the target analysis task and the corresponding task data uploaded by the user through the product interaction page; then, it parses the target analysis task and determines the task type based on the parsing results. For example, the user agent performs semantic understanding of the target analysis task, provides keywords, and determines the task type through keyword matching; or the user agent relies on an intent recognition model to perform intent recognition on the target analysis task and determines the task type through the task type label output by the intent recognition model. Then, based on the task type of the target analysis task, it selects historical analysis tasks that match the task type from the memory unit.

[0088] The specific implementation of "selecting historical analysis tasks matching the task type of the target analysis task from the memory unit" can be achieved through automatic interaction between the user agent and the memory agent. For example, the user agent can send the task type of the target analysis task to the memory agent. After receiving the task type, the memory agent selects an initial set of historical analysis tasks matching the task type from the associated memory unit through keyword matching or semantic vector matching, and returns this initial set of historical analysis tasks to the user agent. The user agent can then select one or more historical analysis tasks from the initial set of historical analysis tasks as the historical analysis tasks corresponding to the target analysis task. For instance, the user agent can select the initial historical analysis task with the most recent execution completion time from the initial historical analysis task set as the historical analysis task corresponding to the target analysis task; alternatively, all initial historical analysis tasks in the entire initial set can be used as the historical analysis tasks corresponding to the target analysis task, and the user agent can retrieve the corresponding data from these historical analysis tasks according to actual needs.

[0089] After selecting a historical analysis task, the target user profile corresponding to the target analysis task can be generated based on the target analysis task and the historical analysis tasks. In other words, by parsing the target analysis task, UserAgent finds task-related information from UserHistory (user history, i.e., historical analysis tasks), so that some information that needs to be confirmed (such as data source, indicators to be analyzed, or display format, etc.) can be pre-filled using previous operating habits, and the target user profile corresponding to the target analysis task can be generated quickly.

[0090] For example, if the historical analysis task is "help me perform attribution analysis on recent satisfaction metrics", and the user agent has undergone multiple confirmations of requirements (such as clarification regarding data sources and display formats), then when the target analysis task is "help me perform attribution analysis on recent satisfaction metrics", the experience from the historical analysis task can be directly utilized without needing to clarify the target analysis task again.

[0091] The data processing method provided in one or more embodiments of this specification involves a user agent receiving a target analysis task and the corresponding task data. The user agent parses the target analysis task to determine its task type and automatically interacts with a memory agent based on this task type. The user agent selects a historical analysis task matching the task type from the memory unit associated with the memory agent. Subsequently, relevant information required for the target analysis task can be retrieved from the historical analysis task. This allows information requiring confirmation in the target analysis task to be pre-filled using parameters from the historical analysis task execution, eliminating the need for multiple confirmations from the user, reducing interaction costs, and improving user experience.

[0092] Therefore, after the user agent selects a historical analysis task that matches the task type of the target analysis task, a target user profile corresponding to the target analysis task can be generated based on the historical user profiles corresponding to the historical analysis tasks and the task requirements of the target analysis task. The specific implementation method is as follows:

[0093] The step of generating a target user profile corresponding to the target analysis task using a user agent based on the target analysis task and the historical analysis task includes:

[0094] The user intelligent agent is used to analyze the target analysis task and determine the task requirements of the target analysis task.

[0095] Using the user intelligent agent, determine the historical user profile corresponding to the historical analysis task;

[0096] Using the user intelligent agent, the task requirements of the target analysis task are compared with the historical user profiles, and a target user profile corresponding to the target analysis task is generated based on the comparison results.

[0097] Continuing with the previous example, if the target analysis task is to perform attribution analysis on recent satisfaction indicators, then the task requirements can be understood as: what specific data to use, what indicators to analyze, and what format of the analysis report to output; for example, the data source is a1, the indicator is b1, and the presentation format is c1. In practice, the user agent can rely on an NLP (Natural Language Processing) model to parse the requirements of the target analysis task, thereby obtaining the task requirements.

[0098] In this context, the historical user profile corresponding to the historical analysis task can be understood as a historical profile, representing the past behaviors of the user who sent the target analysis task. For example, if the user sent an analysis task yesterday asking for "attribution analysis of recent satisfaction metrics," and the user agent confirmed the task requirements (i.e., clarifying the data source, metrics, and presentation format) multiple times through dialogue, then the analysis task was executed. Therefore, if this analysis task is a historical analysis task corresponding to the task type of the target analysis task, the user agent does not need to confirm the task requirements multiple times through multiple dialogues for that target analysis task; it only needs to initiate a single confirmation with the user to directly reuse the historical profile.

[0099] In practical applications, the requirements of a target analysis task may differ from historical user profiles. For example, the target analysis task might include a new requirement: adding the b2 metric. In such cases, when generating the target user profile for the target analysis task, the user agent needs to incrementally modify the historical user profile based on this new requirement of the target analysis task.

[0100] In practice, the user agent parses the target analysis task, determines the task requirements of the target analysis task, and determines the historical user files corresponding to the historical analysis task. Then, it compares the task requirements of the target analysis task with the historical user files and generates the target user files corresponding to the target analysis task based on the comparison results.

[0101] For example, suppose the target analysis task is to analyze satisfaction, which should include indicators b1 and b2, and be displayed in the form of c1. The user agent parses the target analysis task and obtains the task requirements of the target analysis task as {"indicators": ["b1"," b2"], "display form": " c1"}.

[0102] In addition, the historical analysis task corresponding to the target analysis task is historical task A, and the historical profile of historical task A is {"Data source": "a1" ,"Indicator": "b1", "Display format": "c2"}.

[0103] Therefore, when the user agent generates the target user profile for the target analysis task based on the task requirements of the target analysis task and the historical user profiles of the historical analysis tasks, it compares the task requirements of the target analysis task with the historical user profiles of the historical analysis tasks. If the task requirements of the target analysis task do not declare a data source, it can directly inherit the data source from the historical profile of historical task A. Additionally, although the task requirements of the target analysis task define indicators, if indicator b2 is added, then indicator b2 is extended. Finally, although both the task requirements of the target analysis task and the historical user profiles of the historical analysis tasks define display formats, but these display formats differ, the task requirements of the target analysis task can be used to override the display formats in the historical user profiles of the historical analysis tasks. Therefore, the final generated target user profile for the target analysis task is {"Data Source": "a1", "Indicator": ["b1", "b2"], "Display Format": "c1"}.

[0104] In another possible embodiment, the target analysis task is merely a vague task, such as the above: help me perform attribution analysis on recent satisfaction indicators. In this case, the task requirements of the target analysis task are unclear. At this time, the historical user profiles of the historical analysis tasks can be used directly to generate the target user profile of the target analysis task, without having to make multiple clarifications to the user. It is only necessary to send the generated target user profile of the target analysis task to the user for confirmation once.

[0105] The data processing method provided in one or more embodiments of this specification parses the target analysis task through a user intelligent agent, determines the task requirements of the target analysis task, compares the task requirements with the historical user files of the historical analysis tasks corresponding to the target analysis task, extracts reusable experience information from the historical user files based on the comparison results, adds task requirements not present in the historical user files, and / or updates information with different values ​​for the same key in the historical user files, inherits user preferences through the historical user files, ensures that the depth and breadth of analysis conform to user habits, avoids repeated confirmation of known information, reduces the cost of repeated interaction, and realizes the memory-based priority of target analysis task execution.

[0106] Step 204: Based on the target analysis task, the task data, and the target user profile, generate a task execution plan corresponding to the target analysis task using a management agent.

[0107] After the user agent generates the target user profile corresponding to the target analysis task based on the task requirements of the target analysis task and the historical user profile of the historical analysis tasks, the target analysis task, task data and target user profile can be automatically sent to the management agent so that the management agent can generate the task execution plan corresponding to the target analysis task based on these parameters.

[0108] In practical applications, to reduce the uncertainty in task execution plan generation, before generating the task execution plan corresponding to the target analysis task using a management agent based on the target analysis task, task data, and target user profile, a data agent is used to parse the task data, obtain a summary of the task data, and participate in the generation of the task execution plan corresponding to the target analysis task. The specific implementation method is as follows:

[0109] The management intelligent agent includes a data intelligent agent;

[0110] Before generating the task execution plan corresponding to the target analysis task using the management agent based on the target analysis task, the task data, and the target user profile, the method further includes:

[0111] The data intelligence agent is used to parse the task data to obtain a summary information of the task data;

[0112] The step of generating a task execution plan corresponding to the target analysis task using a management agent based on the target analysis task, the task data, and the target user profile includes:

[0113] Based on the target analysis task, the data summary information, and the target user profile, the management agent generates a task execution plan corresponding to the target analysis task.

[0114] The data summary information of the task data can be understood as the summary of the data file. It is a quick overview of the key characteristics of the data in the task data, such as its structure, quality, and distribution, and is used to guide the specification of subsequent task execution plans.

[0115] Taking the above example, if the target analysis task is to help me perform attribution analysis on recent satisfaction indicators, and the corresponding task data is customer satisfaction survey data, the file content of this task data includes hundreds of questionnaires, and the fields include customer ID, satisfaction rating (1-5 points), feedback text, number of purchases, etc.; then, by using a data intelligence agent to parse this task data, the summary of the task data obtained includes, but is not limited to: basic statistics (such as the mean satisfaction rating, the distribution of each rating level), data quality checks (the proportion of missing values, outlier identification), and text field summaries (word frequency statistics of feedback text, sentiment distribution), etc.

[0116] After the data agent analyzes the task data and obtains the data summary information, the management agent can generate the task execution plan corresponding to the target analysis task based on the target analysis task, the data summary information, and the target user profile.

[0117] The data processing method provided in one or more embodiments of this specification, before generating a task execution plan using a management agent, parses the task data corresponding to the target analysis task to generate a summary of the task data using a data agent. This task data summary can reveal data quality problems through statistical descriptions (such as the proportion of missing values ​​and the distribution of outliers), and can also unify heterogeneous data into a structured summary to avoid plan execution failure due to format differences. Secondly, the feature distribution in the task data summary (such as field correlation and data volume) will also directly affect the logic of the management agent in generating the task execution plan, so that when the management agent subsequently generates the task execution plan corresponding to the target analysis task based on the target analysis task, data summary information, and target user profile, it can generate a high-quality, executable task execution plan that avoids invalid paths based on the data summary information.

[0118] Of course, in practical applications, a series of rules and logical conditions can be set in advance (such as "if the task data contains time series, then perform trend analysis") to automatically generate the task execution plan corresponding to the target analysis task based on the matching rules of task and data characteristics; or an execution template can be designed in advance for the target analysis task, and after determining the task type of the target analysis task, the corresponding template can be automatically applied according to the task type, and specific parameters can be filled in according to the task data and target user profile to generate the task execution plan corresponding to the target analysis task, etc.

[0119] Specifically, the method for generating a task execution plan corresponding to the target analysis task using a management agent based on the target analysis task, data summary information, and target user profile is as follows:

[0120] The management agent includes a demand management agent;

[0121] The step of generating a task execution plan corresponding to the target analysis task using the management agent based on the target analysis task, the data summary information, and the target user profile includes:

[0122] Based on the target analysis task, the data summary information, and the target user profile, the demand management agent generates an initial execution plan and performs information gap detection on the initial execution plan to obtain the information gap detection results.

[0123] If it is determined from the information gap detection results that there is no information gap in the initial execution plan, the initial execution plan is determined as the task execution plan corresponding to the target analysis task;

[0124] or

[0125] If, based on the information gap detection results, it is determined that there is an information gap in the initial execution plan, the user agent is used to supplement the initial execution plan to generate a task execution plan corresponding to the target analysis task.

[0126] Among them, the demand management intelligent agent can be understood as the AgentPlanner in the above terminology explanation, which is used to generate the task execution plan corresponding to the target analysis task.

[0127] In practice, the demand management agent first generates an initial execution plan based on the target analysis task, data summary information, and target user profiles. It then performs information gap detection on the initial execution plan to obtain the results. If the information gap detection results indicate that the initial execution plan has no information gaps, it is designated as the task execution plan corresponding to the target analysis task. Conversely, if the information gap detection results indicate that the initial execution plan has information gaps, the user agent supplements the initial execution plan, thereby generating the task execution plan corresponding to the target analysis task. The initial execution plan can be understood as a preliminary analysis of the target analysis task, including basic execution steps and detected information gaps. For example, it may include several basic steps (such as data extraction, overall trend analysis, and dimensional analysis) and discovered missing information (such as missing customer service groupings). The task execution plan, on the other hand, can be understood as a complete analysis plan formed by updating the initial execution plan after supplementing the information. This task execution plan includes all tasks that close the information gaps and is an executable plan. In other words, the demand management agent first generates an initial execution plan based on the target analysis task, data summary information, and target user profile, and accurately identifies missing information. The user agent then completes the closed-loop supplementation, and the demand management agent updates and generates an accurate and executable task execution plan based on this supplementation.

[0128] In practical applications, the demand management agent might first extract task requirements based on the target analysis task and target user profiles; for example, identifying the analysis objectives, time frame, and analysis metrics from the target analysis task and target user profiles. Secondly, it compares the extracted task requirements with the data summary information, such as checking data availability (whether the file contains the required metrics), time frame (whether the date range in the file covers recent data), and data quality (whether there are missing or outlier values). Then, based on the above information, it generates an initial execution plan. For example, the demand management agent might design an initial execution plan through information parsing and understanding, resource matching, and tool selection. The initial execution plan includes, but is not limited to, analysis steps, required variables, and expected outputs. Finally, the demand management agent uses natural language processing technology to detect missing information in the initial execution plan, identifying information that needs to be supplemented, such as: an unclear time frame (analyzing "recent" satisfaction, but without specifying the number of days), or missing key metrics (analyzing return rates, but the data summary indicates that the task data does not contain these metrics). At this point, the demand management agent can automatically send a request to the user agent to supplement missing information. The user agent can then query other systems or files to fill in the missing information. For missing information requiring user confirmation, a clarification question can be generated and sent to the user agent, which will then present it to the user through the product interaction page, awaiting the user's response. After obtaining the missing information supplemented by the user or automatically acquired by the user agent, the initial execution plan can be supplemented based on this information to generate a complete task execution plan. This multimodal closed-loop supplementation mechanism, including user supplementation and user agent supplementation, automatically fills in missing information, avoiding rework caused by later data issues and shortening the overall data processing cycle.

[0129] Using the previous example, the target analysis task is: help me analyze the reasons for the recent decline in satisfaction; the task data is: a satisfaction data table (the data summary shows the date and analysis indicator b1), the target user profile is: analysis indicator = b1, and the output format is c1.

[0130] The requirement management agent first extracts preliminary requirements based on the target analysis task and target user profile: Target = Attribution Analysis, Analysis Metric = b1, Time Range = Recent (to be clarified). Next, the extracted preliminary requirements are compared with the data summary information. At this point, "Recent" is not defined. The requirement management agent generates an initial execution plan and performs information gap detection on the initial execution plan. The information gap detection result is: Time Range undefined, needs to be supplemented. The requirement management agent can then send a supplement request to the user agent: Please confirm the time range is the last 30 days? The last 7 days? Or something else? Upon receiving the gap information returned by the user agent, the requirement management agent can supplement the initial execution plan based on the gap information, generating a complete task execution plan.

[0131] In practical applications, information gap detection results include both the presence and absence of information gaps. For gaps where information gaps exist, common gap types include: missing data fields (e.g., requirement dimensions not present in the file column); data quality defects (e.g., key column missing rate > preset threshold); triggering data cleaning assistance (e.g., date column has 15% null values, should we imput them by previous and next values?); and conflicting analysis logic (e.g., conflict between requirement and Profile configurations). Specific strategies include: resolving conflicts (e.g., your historical data uses a weekly dimension, but the file contains daily data, should we change it to daily trend analysis?).

[0132] The data processing method provided in one or more embodiments of this specification involves a demand management intelligent agent first generating an initial execution plan based on the target analysis task, data summary information, and target user profiles. This initial execution plan is then automatically subjected to information gap detection to obtain the results, thereby improving demand conversion efficiency, increasing gap detection accuracy, and ensuring analysis reliability. If the information gap detection results indicate that the initial execution plan has no information gaps, it can be directly designated as the task execution plan corresponding to the target analysis plan. Conversely, if the information gap detection results indicate that the initial execution plan has information gaps, only the missing information needs to be supplemented to generate an executable, unambiguous, and clearly defined task execution plan. Through the intelligent demand parsing, dynamic planning, and feedback iteration of the demand management intelligent agent, the accuracy and operability of the task execution plan are significantly improved.

[0133] Step 206: Using the management agent and analysis model, generate a verification analysis plan and the corresponding verification sub-tasks according to the task execution plan.

[0134] The analysis model is determined by the management agent based on the analysis results of the task execution plan.

[0135] In practical applications, a task execution plan can be understood as the basic framework for a target analysis task. It transforms the task requirements of the target analysis task into an executable data analysis framework, ensuring coverage of all dimensions (such as time, customer service, product, etc.) within the task requirements. However, it doesn't delve into attribution (such as the root causes affecting satisfaction). Simply executing the task execution plan to generate a report lacks exploration and verification of multiple possible causes, resulting in a report that only scratches the surface, not identifying the root causes. Therefore, in one or more embodiments of this specification, after generating the task execution plan, the management agent will also combine the analysis model to generate a verification analysis plan and corresponding verification subtasks. Only after executing the verification subtasks and confirming the validity of the verification analysis plan will the final attribution analysis report be generated based on the task execution results of the verification subtasks. The specific implementation method is as follows:

[0136] The management agent includes a task decomposition agent;

[0137] The step of generating a verification analysis plan and corresponding verification subtasks based on the task execution plan using the management agent and analysis model includes:

[0138] The task execution plan is analyzed using the task decomposition agent to determine multiple analytical dimensions of the task execution plan;

[0139] Based on the multiple analysis dimensions, multiple corresponding analysis models are determined, and the task execution plan is input into the multiple analysis models respectively to obtain multiple verification analysis plans;

[0140] The multiple verification analysis plans are decomposed to obtain multiple verification sub-tasks corresponding to the multiple verification analysis plans.

[0141] In practice, the task decomposition agent first parses the task execution plan and determines multiple analysis dimensions of the task execution plan. Then, based on the multiple analysis dimensions, it determines a suitable analysis model for each analysis dimension and inputs the task execution plan into the multiple analysis models corresponding to the multiple analysis dimensions to obtain multiple verification analysis plans output by each analysis model. The task decomposition agent then decomposes each of the multiple verification analysis plans to obtain multiple verification subtasks corresponding to each verification analysis plan.

[0142] For example, if the task execution plan is as follows: "Analysis Objective": "Satisfaction Index Attribution", "Core Steps": ["Data Extraction (Data Source a1, last 60 days)", "Calculate Weekly Average Net Promoter Score and Month-on-Month Change", "Analyze Differences by Customer Service Group (Users have supplemented rules)", "Analyze the Correlation between Return Rate and Net Promoter Score", "Identify Abnormal Time Points"], "Output Requirements": "Output format is c1".

[0143] Therefore, by analyzing the task execution plan, the task decomposition agent can determine the multiple analytical dimensions of the plan: service dimension, product dimension, and market dimension. Then, the task decomposition agent can select analytical models that fit these three dimensions, such as analytical model 1, analytical model 2, and analytical model 3. Analytical model 1 can focus on the service dimension, analytical model 2 on the product dimension, and analytical model 3 on the market dimension. In practical applications, some analytical dimensions can be pre-matched with a series of known analytical models, and then a suitable analytical model can be selected for each dimension based on this matching relationship.

[0144] After determining multiple analysis models based on multiple analysis dimensions, the task decomposition agent inputs the task execution plan into analysis model 1, analysis model 2, and analysis model 3 respectively to obtain the verification analysis plan 1 output by analysis model 1, the verification analysis plan 2 output by analysis model 2, and the verification analysis plan 3 output by analysis model 3. Each verification analysis plan can also be understood as the attribution hypothesis output by each analysis model (such as "customer service response delay is the main cause" or "product quality problem is the cause").

[0145] Then, the task decomposition AI will generate multiple verification subtasks for each attribution hypothesis. These verification subtasks will be executed, and after the execution of the verification subtasks, the validity of each attribution hypothesis can be determined based on the task execution results. The verified attribution hypotheses will then be unified into the final task analysis report.

[0146] Following the example above, the validation analysis plan 1 and its corresponding validation subtasks can be: "Attribution Hypothesis": [{"Hypothesis ID": "Attribution Hypothesis 1","Description": "Customer service response delays lead to decreased satisfaction","Source Model": "Qwen","Validation Subtask": [{"Validation Subtask ID": "T1-1", "Description": "Extract customer service response time data"},{"Validation Subtask ID": "T1-2", "Description": "Calculate the correlation coefficient between response time and Net Promoter Score"},{"Validation Subtask ID": "T1-3", "Description": "Cluster analysis of high-latency customer service groups"}]}.

[0147] The validation analysis plan 2 and its corresponding validation subtasks can be: {"Hypothesis ID": "Attribution Hypothesis 2","Description": "Product quality issues (return rate) cause dissatisfaction","Source Model": "Claude","Validation Subtask": [{"Validation Subtask ID": "T2-1", "Description": "Analyze abnormal time points of return rate"},{"Validation Subtask ID": "T2-2", "Description": "Construct a regression model of Net Promoter Score and return rate"}]}.

[0148] The validation analysis plan 3 and its corresponding validation subtasks can be: {"Hypothesis ID": "Attribution Hypothesis 3","Description": "Competitor promotions lead to increased user expectations","Source Model": "GPT-4o","Validation Subtask":[{"Validation Subtask ID": "T3-1", "Description": "Obtain competitor promotion information during the same period"},{"Validation Subtask ID": "T3-2", "Description": "Match the time series of the promotion period and the decline in Net Promoter Score"}]}.

[0149] As can be seen from the above, the verification analysis plan is a detailed execution plan generated by the task decomposition agent based on the task execution plan and combined with the attribution hypotheses of multiple analysis models corresponding to multiple analysis dimensions. It optimizes the deepening (increasing depth) and expansion (increasing breadth) of the task execution plan, and is subsequently used to verify various attribution hypotheses. Each verification analysis plan contains a series of verification sub-tasks, which are more detailed and diverse than the tasks in the task execution plan.

[0150] Furthermore, in practical applications, each analysis model can have one or more attribution hypotheses. If each analysis model has multiple attribution hypotheses, the task decomposition agent can perform operations such as deduplication, sorting, and conflict detection on these multiple attribution hypotheses to generate an integrated and more accurate list of attribution hypotheses. Detailed verification subtasks are then designed for each attribution hypothesis in the list. For ease of understanding, this specification uses the example of each analysis model outputting one attribution hypothesis for detailed explanation in one or more embodiments.

[0151] The data processing method provided in one or more embodiments of this specification involves a task decomposition intelligent agent parsing a task execution plan, determining multiple analytical dimensions of the task execution plan, obtaining verification analysis plans output by each analytical model based on multiple analytical models corresponding to the multiple analytical dimensions, and decomposing each verification analysis plan to obtain multiple verification sub-tasks corresponding to each verification analysis plan. This method enables the generation of complementary attribution hypotheses based on the task execution plan through a heterogeneous large model, and decomposes each attribution hypothesis into parallel executable atomic sub-tasks, achieving a large depth and multi-dimensional coverage of attribution analysis. It transforms ambiguous attribution requirements into verifiable, quantifiable, and executable specific analytical processes, delving into attribution to explore and verify multiple causes, ensuring that the subsequently generated task analysis report does not merely remain at the surface level. This provides users with clear, actionable, and high-value task analysis reports, such as accurately identifying key factors affecting sales success or failure, driving project optimization and growth.

[0152] Step 208: Using the management agent, execute the verification sub-tasks corresponding to the verification analysis plan, obtain the task execution results, and generate a task analysis report corresponding to the target analysis task based on the task execution results.

[0153] After the task decomposition agent combines multiple analysis models to generate multiple verification analysis plans and multiple verification subtasks corresponding to each verification analysis plan, the management agent needs to determine the task execution order and task execution resources for the multiple verification subtasks corresponding to each verification analysis plan. Based on the task execution order and task execution resources, each verification subtask is executed to obtain the overall task execution result. The specific implementation is as follows:

[0154] The management agent includes a task execution agent;

[0155] Using the management agent, the verification subtasks corresponding to the verification analysis plan are executed to obtain the task execution results, including:

[0156] Using the task execution agent, a target verification subtask is determined from multiple verification subtasks corresponding to the multiple verification analysis plans according to a preset execution order, and corresponding task execution resources are allocated to the target verification subtask.

[0157] Using the task execution agent, the target verification subtask is executed according to the task execution resources corresponding to the target verification subtask, and the subtask execution result of the target verification subtask is obtained;

[0158] Using the task execution agent, if it is determined that all verification subtasks corresponding to the multiple verification analysis plans have been completed, the task execution result is determined based on the subtask execution results of the target verification subtask.

[0159] The preset execution order can be set according to actual needs. For example, the preset execution order can allow multiple verification subtasks corresponding to each verification analysis plan in multiple verification analysis plans to be executed asynchronously and simultaneously in the order of tasks. For example, if verification analysis plan 1 includes verification subtasks T1-1, T1-2, and T1-3, verification analysis plan 2 includes verification subtasks T2-1 and T2-2, and verification analysis plan 3 includes verification subtasks T3-1 and T3-2, then the execution will first asynchronously execute (T1-1, T2-1, T3-1), then asynchronously execute (T1-2, T2-2, T3-2), and then execute T1-3.

[0160] For each target validation subtask, allocate corresponding task execution resources, including but not limited to selecting a model for executing the target validation subtask (such as any one of qwen, claude, gpt4o, or other models), constructing prompt words (constructing prompt words for the target validation subtask based on the selected model), and constructing a runtime environment (such as a computing environment and a query environment).

[0161] In practice, the task execution agent, according to a preset execution order, first determines the target verification subtask from the multiple verification subtasks corresponding to each verification analysis plan. For example, the task execution agent will first determine T1-1, T2-1, and T3-1 as target verification subtasks. Then, it allocates corresponding task execution resources to each target verification subtask; based on the task execution resources corresponding to each target verification subtask, it executes each target verification subtask and obtains the subtask execution result of each target verification subtask; it continues to execute the next round of target verification subtasks, for example, the task execution agent will then determine T1-2, T2-2, and T3-2 as target verification subtasks, and repeat the above steps until all verification subtasks corresponding to each verification analysis plan have been executed. Finally, the subtask verification results of each verification subtask corresponding to each verification analysis plan are combined to form the final task execution result.

[0162] The data processing method provided in one or more embodiments of this specification utilizes a task execution intelligence agent to execute each verification subtask corresponding to each verification analysis plan according to a preset execution order and the task execution resources allocated to each verification subtask corresponding to each verification analysis plan. This allows the task execution intelligence agent to dynamically execute each verification subtask through intelligent model matching, environment construction, and prompt word construction, achieving efficient and reliable task implementation. Each verification subtask corresponding to each verification analysis plan is executed asynchronously and simultaneously. This parallel processing of multiple verification subtasks allows more work to be completed per unit time, greatly improving the processing speed of the entire data processing method and enhancing the user experience. Furthermore, asynchronous execution allows the task execution intelligence agent to dynamically allocate resources according to the resource requirements of the verification subtasks, thereby more effectively utilizing hardware and software resources and increasing its flexibility.

[0163] To further optimize each verification analysis plan and ensure the accuracy of subsequent task analysis reports, the corresponding verification analysis plan can be updated in real time based on the execution results of each verification subtask. This iterative process enables adaptive evolution of the analysis process. The specific implementation method is as follows:

[0164] After executing the subtask execution result of the target verification subtask according to the task execution resources corresponding to the target verification subtask, the process further includes:

[0165] Based on the subtask execution results of the target verification subtask, update the verification analysis plan corresponding to the target verification subtask and other verification subtasks, wherein the other verification subtasks are the unexecuted verification subtasks corresponding to the verification analysis plan corresponding to the target verification subtask, excluding the target verification subtask.

[0166] Continuing with the previous example, assume that verification analysis plan 1 includes verification subtasks T1-1, T1-2, and T1-3, verification analysis plan 2 includes verification subtasks T2-1 and T2-2, and verification analysis plan 3 includes verification subtasks T3-1 and T3-2. In this case, the target verification subtasks corresponding to each verification analysis plan determined by the task execution agent are (T1-1, T2-1, T3-1). The other verification subtasks corresponding to each verification analysis plan are verification subtasks T1-2 and T1-3 in verification analysis plan 1, verification subtask T2-2 in verification analysis plan 2, and verification subtask T3-2 in verification analysis plan 3.

[0167] In practice, the task execution agent first executes the target verification subtask (T1-1, T2-1, T3-1), and then updates the verification subtasks T1-2 and T1-3 in verification analysis plan 1 based on the execution results of the subtasks of the target verification subtask T1-1. It then updates the verification subtask T2-2 in verification analysis plan 2 based on the execution results of the subtasks of the target verification subtask T2-1, and updates the verification subtask T3-2 in verification analysis plan 3 based on the execution results of the subtasks of the target verification subtask T3-1. And so on. When the target verification subtask is (T1-2, T2-2, T3-2), the verification subtask T1-3 in verification analysis plan 1 can be updated based on the execution results of the subtasks of the target verification subtask T1-2.

[0168] In practical applications, based on the execution results of each target verification subtask, not only can the corresponding verification analysis plan and other verification subtasks be updated, but new verification analysis plans and their corresponding verification subtasks can also be added. The execution steps for newly added verification analysis plans and their corresponding verification subtasks are described above and will not be repeated here. Alternatively, based on the execution results of each target verification subtask, unnecessary subsequent executions of verification analysis plans or specific verification subtasks can be deleted.

[0169] The data processing method provided in one or more embodiments of this specification allows the task execution agent to dynamically adjust subsequent verification subtasks by analyzing the execution results of completed verification subtasks in real time. This avoids executing invalid or low-value verification subtasks. For example, if a new clue is discovered based on the execution result of a certain verification subtask, an exploration branch and a root cause verification subtask can be added. If no new clue is found, the next original verification subtask can continue to be executed. Alternatively, if the execution result of a certain verification subtask indicates that the verification failed due to data quality issues, the aforementioned gap information supplementation process can be triggered to update the task execution plan, verification analysis plan, and the corresponding verification subtasks. Through this dynamic adjustment, new attribution hypotheses are added or the analysis direction is adjusted based on the execution results of the target verification subtask, ultimately generating a more accurate task analysis report.

[0170] In practical applications, before updating the verification analysis plan corresponding to each target verification subtask and other verification subtasks based on the subtask execution results of each target verification subtask, the subtask execution results of each target verification subtask are stored in a shareable memory unit. Then, each verification analysis plan can obtain the next new information for non-self-verifying analysis plans from this memory unit and update itself accordingly. This iterative process achieves adaptive evolution of the analysis process. The specific implementation method is as follows:

[0171] After obtaining the subtask execution result of the target verification subtask, the process further includes:

[0172] Using the task execution agent, the subtask execution result of the target verification subtask is sent to the memory agent;

[0173] The memory agent is used to store the execution results of the subtasks of the target verification subtask into the corresponding memory units.

[0174] In practical applications, a memory module can include a short-term memory unit, a medium-term memory unit, and a long-term memory unit. For example, the short-term memory unit is used to cache the execution information of target analysis tasks within a preset time (1 day), the medium-term memory unit is used to store the execution information of target analysis tasks that need to be removed from the short-term memory unit, and the long-term memory unit is used to store the execution information of target analysis tasks whose popularity exceeds a preset popularity threshold in the medium-term memory unit.

[0175] For each target verification subtask, the memory intelligence will store the subtask execution results in an appropriate location based on the characteristics of short-term memory, medium-term memory, and long-term memory units.

[0176] Specifically, after obtaining the subtask execution result of each target verification subtask, the task execution agent will send the subtask execution result of the target verification subtask to the memory agent. The memory agent will then store the subtask execution result of the target verification subtask into the corresponding memory unit according to the unit characteristics of the short-term memory unit, medium-term memory unit, and long-term memory unit.

[0177] At this point, the subtask execution results stored in the corresponding memory unit can not only optimize the corresponding verification analysis plan, but also be transformed into reusable structured knowledge to build the system's long-term cognitive capabilities. For example, the conclusion of customer service resource shortage in the subtask execution results can be stored in the user profile, and the customer service dimension analysis can be automatically added in the next similar task. That is, the memory unit is used to store the knowledge results (such as facts and rules obtained from analysis) generated during the execution of each verification subtask, which can be used to support intelligent decision-making for future analysis tasks. For example, if the historical profile of a certain target analysis task includes "indicator": ["Net Promoter Score"], and the verification subtask analyzes the Net Promoter Score and finds that "night shift delay affects the Net Promoter Score", then the execution result of the subtask is stored in the memory unit. The next execution of a similar task to the target analysis task will automatically add the night shift analysis dimension, so that the memory unit accumulates domain knowledge in the continuous execution of verification subtasks.

[0178] The data processing method provided in one or more embodiments of this specification, after obtaining the subtask execution results of each target verification subtask, stores them in the corresponding memory unit according to the interaction between the task execution agent and the memory agent. This not only enables information sharing among all verification analysis plans and achieves adaptive optimization, but also transforms them into reusable structured knowledge. It optimizes the historical profiles of analysis tasks stored in the memory unit, increases their analysis dimensions, and optimizes the generation of user profiles for similar analysis tasks, thereby improving the generation speed and accuracy of task analysis reports for similar analysis tasks.

[0179] Therefore, after all verification subtasks corresponding to all verification analysis plans have been completed, the execution results of all verification subtasks can be summarized to obtain the final task execution result, and a task analysis report meeting the user's requirements can be generated based on this result. The specific implementation method is as follows:

[0180] The management agent includes a result aggregation agent;

[0181] The step of generating a task analysis report corresponding to the target analysis task based on the task execution results includes:

[0182] The result-summarizing agent obtains the task execution results and generates a task analysis report corresponding to the target analysis task based on the task execution results and the task requirements of the target analysis task.

[0183] Specifically, after obtaining the task execution results of the target analysis task, the result aggregation agent will transform the multi-dimensional task execution results into a task analysis report that the user can understand and deliver it. For example, the result aggregation agent will transform the multi-dimensional task execution results into a task analysis report that the user can understand and send it to the report generation agent (WriterAgent). The report generation agent will then generate a personalized task analysis report that meets the user's needs based on the task requirements of the target analysis task and present it to the user.

[0184] The data processing method provided in one or more embodiments of this specification, after all verification subtasks have been completed, ultimately uses code generation to convert the analysis text and charts in the task execution results into HTML (Hypertext Markup Language) code, resulting in a user-understandable, richly illustrated, in-depth document. This method integrates analysis conclusions (text) and data evidence (icons) into a structured, interactive HTML document through programming, clearly demonstrating task completion, identifying bottlenecks and anomalies, and supporting subsequent optimization strategy formulation. Simultaneously, by presenting key indicators through data visualization, it improves overall operational efficiency and enhances user experience.

[0185] The data processing method provided in one or more embodiments of this specification can automatically generate multi-dimensional verification analysis plans based on the user's target analysis task and historical execution information of historical analysis tasks corresponding to the target analysis task, through a conversational interaction. It can also dynamically update the verification analysis plan and other unexecuted verification sub-tasks in the verification analysis plan by sharing the task execution results of each verification sub-task, or by removing unreasonable verification sub-tasks. This dynamic update mechanism optimizes the entire execution process, significantly improving analysis efficiency and intelligence. In other words, during the execution of the target analysis task, this data processing method can provide real-time feedback based on the results of verification sub-tasks and dynamically adjust subsequent tasks, avoiding invalid branches and improving the accuracy and adaptability of task execution. Simultaneously, it supports the parallel execution of multiple verification sub-tasks and the aggregation of sub-task execution results, making it suitable for complex task analysis scenarios such as causal analysis. Furthermore, by introducing a sub-intelligent agent mechanism and combining it with professional analysis tools, the professionalism and visualization effects of the data are ensured. Finally, text and charts can be integrated using HTML code generation to output a richly illustrated in-depth task analysis report. It integrates large-scale model capabilities with a multi-agent framework, possessing high intelligence, modularity, and scalability, enabling the data analysis field to evolve towards automation, dynamism, and intelligence.

[0186] The following is in conjunction with the appendix Figure 3Taking the application of the data processing method provided in this specification in a dialogue analysis scenario in the customer service field as an example, the data processing method will be further explained. Among other things, Figure 3 A flowchart illustrating the processing procedure of a data processing method according to an embodiment of this specification is shown, specifically including the following steps.

[0187] Step 302: The user agent receives the user query and the data file corresponding to the user query sent by the user through the product interaction page.

[0188] In this context, the user query can be understood as the target analysis task in the above embodiments, such as: help me perform attribution analysis on recent satisfaction indicators.

[0189] Step 304: The user agent generates a user profile based on the user query and the historical analysis task corresponding to the user query, and sends the user query, the data file for the user query, and the user profile to the manager agent.

[0190] Specifically, the user agent parses the user query, determines the corresponding task type and task requirements, and interacts with the memory agent to select the historical analysis task (i.e., ...) based on the task type of the user query. Figure 3 The user profile is generated based on the user's historical records (the historical profile of the analysis task) and the task requirements corresponding to the user's query.

[0191] Step 306: The demand management agent in the manager agent receives the user query, the data file for the user query, and the user profile, and generates a task execution plan based on the user query, the summary of the data file for the user query, and the user profile.

[0192] The summary of the data file for the user's query is generated by the data agent in the management agent by parsing the data file for the user's query.

[0193] Step 308: The demand management agent sends the generated task execution plan to the task decomposition agent in the management agent. The task decomposition agent combines the multi-dimensional analysis model corresponding to the task execution plan to generate multiple verification analysis plans and multiple verification sub-tasks corresponding to each verification analysis plan.

[0194] The task decomposition agent combines the multi-dimensional analysis model corresponding to the task execution plan to generate multiple verification analysis plans and the specific implementation of multiple verification sub-tasks corresponding to each verification analysis plan. For details, please refer to the above embodiments, which will not be repeated here.

[0195] For example, multiple verification analysis plans can be Plan A, Plan B, and Plan C in the diagram, and Plan A, Plan B, and Plan C each correspond to multiple verification sub-tasks.

[0196] Step 310: The task decomposition agent sends the generated multiple verification analysis plans and the multiple verification sub-tasks corresponding to each verification analysis plan to the task execution agent in the management agent; the task execution agent executes each verification sub-task according to the preset execution order and the task execution resources allocated to each verification sub-task.

[0197] Continuing with the previous example, if subtask A is included in the multiple verification subtasks of plan A, subtask B is included in the multiple verification subtasks of plan B, and subtask C is included in the multiple verification subtasks of plan C, and subtask A, subtask B, and subtask C are the first verification subtasks in their respective plans, then the first round of execution of the task execution agent will execute subtask A, subtask B, and subtask C.

[0198] Step 312: The task execution agent executes each verification subtask and obtains the execution results of each verification subtask.

[0199] Using the previous example, when the task execution agent executes subtasks A, B, and C, the execution results of subtasks A, B, and C are obtained.

[0200] Furthermore, the execution results of each verification subtask are used to update the verification subtasks that have not been executed in each verification analysis plan. Simultaneously, the execution results of each verification subtask are also stored in the corresponding memory unit using a memory agent.

[0201] See Figure 4 , Figure 4 A schematic diagram of the structure of a memory unit in a data processing method provided in one embodiment of this specification is shown.

[0202] Figure 4 It includes short-term memory 402, medium-term memory 404, and long-term memory 406. These three units can be understood as three different memory units, all managed by the memory agent 400.

[0203] Among them, long-term memory 406 includes structured storage, storing user profiles and agent profiles. The user profile includes static information such as user profiles (basic attributes / behavioral characteristics, historical interaction records), user knowledge base (knowledge in certain domains), and dynamic information such as user characteristics (preference / habit tags). The agent profile includes static information such as agent profiles (basic attributes, capabilities / permissions), historical experience (task cases / solutions), and dynamic information such as agent characteristics (task response patterns). The data in the long-term memory unit is relatively stable and stores key information about users and agents.

[0204] The intermediate memory 404 can be understood as a cache filter layer. It receives input from the short-term memory (inserted into the intermediate memory) and has an update mechanism based on the popularity of memory segments. It is used to store memory segments whose popularity meets the preset popularity threshold into the long-term memory (updated into the long-term memory) and to delete memory segments (deleted memory segments), such as deleting the first m memory segments.

[0205] Short-term memory 402 can be understood as a real-time workspace used to process the context of the current target analysis task. It retrieves information from user input (pushed to short-term memory), writes the context information for processing the current target analysis task into the page using a first-in-first-out mechanism, and inserts the context information in the page into medium-term memory when the page is full.

[0206] In practice, when a user issues a query task, the user's AI agent interacts with the memory AI agent. The memory AI agent retrieves data from short-term memory, medium-term memory, and long-term memory respectively, and returns the retrieved historical tasks that are the same or similar to the query task (i.e., the retrieved memories in the figure) to the user's AI agent for the generation of the user profile for the query task.

[0207] Step 314: After confirming that all verification subtasks have been executed completely, the information aggregation agent in the management agent generates a task analysis report corresponding to the user's query based on the subtask execution results of all verification subtasks, and presents the task analysis report to the user through the product interaction page.

[0208] The data processing method provided in one or more embodiments of this specification can automatically generate multi-dimensional verification analysis plans based on user queries and historical execution information of historical analysis tasks corresponding to those queries, through a conversational interaction. It can dynamically update the verification analysis plan and other unexecuted verification sub-tasks within the plan by sharing the execution results of each verification sub-task, or by removing unreasonable verification sub-tasks. This dynamic update mechanism optimizes the entire execution process, significantly improving analysis efficiency and intelligence. Specifically, during the execution of the target analysis task, this data processing method can provide real-time feedback based on the results of verification sub-tasks and dynamically adjust subsequent tasks, avoiding invalid branches and improving the accuracy and adaptability of task execution. Simultaneously, it supports the parallel execution of multiple verification sub-tasks and the aggregation of sub-task execution results, making it suitable for complex task analysis scenarios such as causal analysis. Furthermore, by introducing a sub-intelligent agent mechanism combined with professional analysis tools, the professionalism and visualization effects of the data are ensured. Finally, text and charts can be integrated using HTML code generation to output a richly illustrated in-depth task analysis report.

[0209] Corresponding to the above method embodiments, this specification also provides data processing apparatus embodiments. Figure 5 A schematic diagram of the structure of a data processing apparatus according to one embodiment of this specification is shown. Figure 5 As shown, the device includes:

[0210] The file generation module 502 is configured to determine the target analysis task and generate the target user file corresponding to the target analysis task using the user intelligent agent, wherein the target analysis task carries the corresponding task data;

[0211] The plan generation module 504 is configured to generate a task execution plan corresponding to the target analysis task using a management agent based on the target analysis task, the task data, and the target user profile.

[0212] The subtask generation module 506 is configured to use the management agent and the analysis model to generate a verification analysis plan and a verification subtask corresponding to the verification analysis plan according to the task execution plan, wherein the analysis model is determined by the management agent based on the analysis results of the task execution plan;

[0213] The report generation module 508 is configured to use the management agent to execute the verification sub-tasks corresponding to the verification analysis plan, obtain the task execution results, and generate a task analysis report corresponding to the target analysis task based on the task execution results.

[0214] Optionally, the document generation module 502 is further configured as follows:

[0215] Receive the target analysis task and the task data corresponding to the target analysis task;

[0216] The user intelligent agent is used to parse the target analysis task and determine the task type of the target analysis task;

[0217] Using the user agent, historical analysis tasks matching the task type of the target analysis task are selected from the memory unit according to the task type of the target analysis task;

[0218] Based on the target analysis task and the historical analysis task, a target user profile corresponding to the target analysis task is generated using a user agent.

[0219] Optionally, the file generation module 502 is further configured to:

[0220] The user intelligent agent is used to analyze the target analysis task and determine the task requirements of the target analysis task.

[0221] Using the user intelligent agent, determine the historical user profile corresponding to the historical analysis task;

[0222] Using the user intelligent agent, the task requirements of the target analysis task are compared with the historical user profiles, and a target user profile corresponding to the target analysis task is generated based on the comparison results.

[0223] Optionally, the management agent includes a data agent;

[0224] The device further includes:

[0225] The data summary acquisition module is configured as follows:

[0226] The data intelligence agent is used to parse the task data to obtain a summary information of the task data;

[0227] The plan generation module 504 is further configured as follows:

[0228] Based on the target analysis task, the data summary information, and the target user profile, the management agent generates a task execution plan corresponding to the target analysis task.

[0229] Optionally, the management agent includes a demand management agent;

[0230] The plan generation module 504 is further configured as follows:

[0231] Based on the target analysis task, the data summary information, and the target user profile, the demand management agent generates an initial execution plan and performs information gap detection on the initial execution plan to obtain the information gap detection results.

[0232] If it is determined from the information gap detection results that there is no information gap in the initial execution plan, the initial execution plan is determined as the task execution plan corresponding to the target analysis task;

[0233] or

[0234] If, based on the information gap detection results, it is determined that there is an information gap in the initial execution plan, the user agent is used to supplement the initial execution plan to generate a task execution plan corresponding to the target analysis task.

[0235] Optionally, the management agent includes a task decomposition agent;

[0236] The subtask generation module 506 is further configured as follows:

[0237] The task execution plan is analyzed using the task decomposition agent to determine multiple analytical dimensions of the task execution plan;

[0238] Based on the multiple analysis dimensions, multiple corresponding analysis models are determined, and the task execution plan is input into the multiple analysis models respectively to obtain multiple verification analysis plans;

[0239] The multiple verification analysis plans are decomposed to obtain multiple verification sub-tasks corresponding to the multiple verification analysis plans.

[0240] Optionally, the management agent includes a task execution agent;

[0241] The report generation module 508 is further configured to:

[0242] Using the task execution agent, a target verification subtask is determined from multiple verification subtasks corresponding to the multiple verification analysis plans according to a preset execution order, and corresponding task execution resources are allocated to the target verification subtask.

[0243] Using the task execution agent, the target verification subtask is executed according to the task execution resources corresponding to the target verification subtask, and the subtask execution result of the target verification subtask is obtained;

[0244] Using the task execution agent, if it is determined that all verification subtasks corresponding to the multiple verification analysis plans have been completed, the task execution result is determined based on the subtask execution results of the target verification subtask.

[0245] Optionally, the device further includes:

[0246] The task update module is configured as follows:

[0247] Based on the subtask execution results of each target verification subtask, update the verification analysis plan corresponding to the target verification subtask and other verification subtasks, wherein the other verification subtasks are the unexecuted verification subtasks corresponding to the verification analysis plan corresponding to the target verification subtask, excluding the target verification subtask.

[0248] Optionally, the device further includes:

[0249] The storage module is configured as follows:

[0250] Using the task execution agent, the subtask execution result of the target verification subtask is sent to the memory agent;

[0251] The memory agent is used to store the execution results of the subtasks of the target verification subtask into the corresponding memory units.

[0252] Optionally, the management agent includes a result-summarizing agent;

[0253] The report generation module 508 is further configured to:

[0254] The result-summarizing agent obtains the task execution results and generates a task analysis report corresponding to the target analysis task based on the task execution results and the task requirements of the target analysis task.

[0255] The data processing apparatus provided in one or more embodiments of this specification firstly introduces historical analysis tasks corresponding to the target analysis task when generating the target user profile corresponding to the target analysis task. This allows the user agent to extract reusable experience information from the historical user profiles of the historical analysis tasks according to the task requirements of the target analysis task, quickly generating the target user profile corresponding to the target analysis task, avoiding repeated confirmation of known information and reducing the cost of repeated interactions. Secondly, the management agent can automatically plan and generate an executable task execution plan corresponding to the target analysis task based on the target analysis task, task data, and target user profile. Further, combined with the analysis model, it dynamically generates a verification analysis plan (i.e., attribution hypothesis) and verification sub-tasks corresponding to the verification analysis plan, enabling in-depth data understanding. Then, the management agent executes the verification sub-tasks corresponding to the verification analysis plan, generating task execution results and a task analysis report. This achieves an understanding of the causes and mechanisms behind the data, identifies causal relationships between data, and upgrades attribution analysis from descriptive statistics to executable causal decision-making, significantly improving the depth of attribution analysis and revealing the root cause chain affecting key project indicators in the target analysis task. This data processing method, through the interaction and collaboration between user agents and management agents, enables a deep attribution analysis system that can autonomously plan analysis tasks (task execution plans), dynamically generate attribution hypotheses, and verify attribution hypotheses. When processing target analysis tasks according to this deep attribution analysis system, it systematically and automatically mines deep and related factors affecting key project indicators, significantly improving the depth of attribution analysis, revealing the root cause chain, and thus providing clear, actionable, and high-value task analysis reports.

[0256] The above is an illustrative scheme of a data processing apparatus according to this embodiment. It should be noted that the technical solution of this data processing apparatus and the technical solution of the data processing method described above belong to the same concept. For details not described in detail in the technical solution of the data processing apparatus, please refer to the description of the technical solution of the data processing method described above.

[0257] One or more embodiments of this specification also provide a data processing method for dialogue analysis tasks, including:

[0258] A target dialogue analysis task is determined, and a target user profile corresponding to the target dialogue analysis task is generated using a user intelligent agent, wherein the target dialogue analysis task carries corresponding dialogue analysis data.

[0259] Based on the target dialogue analysis task, the dialogue analysis data, and the target user profile, a management agent is used to generate a dialogue analysis task execution plan corresponding to the target dialogue analysis task.

[0260] Using the management agent and the analysis model, a verification analysis plan and a corresponding verification sub-task are generated according to the dialogue analysis task execution plan. The analysis model is determined by the management agent based on the parsing results of the dialogue analysis task execution plan.

[0261] Using the management agent, the verification sub-tasks corresponding to the verification analysis plan are executed, the task execution results are obtained, and a task analysis report corresponding to the target dialogue analysis task is generated based on the task execution results.

[0262] In practice, the specific implementation of the data processing method applied to the dialogue analysis task is the same as the specific implementation of the data processing method described above, and will not be repeated here.

[0263] Corresponding to the above-described data processing method embodiments for dialogue analysis tasks, this specification also provides embodiments of a data processing apparatus for dialogue analysis tasks, the apparatus comprising:

[0264] The first profile generation module is configured to determine the target dialogue analysis task and generate a target user profile corresponding to the target dialogue analysis task using a user agent, wherein the target dialogue analysis task carries corresponding dialogue analysis data.

[0265] The first plan generation module is configured to generate a dialogue analysis task execution plan corresponding to the target dialogue analysis task based on the target dialogue analysis task, the dialogue analysis data, and the target user profile, using a management agent.

[0266] The first subtask generation module is configured to use the management agent and the analysis model to generate a verification analysis plan and a verification subtask corresponding to the verification analysis plan according to the dialogue analysis task execution plan. The analysis model is determined by the management agent based on the parsing results of the dialogue analysis task execution plan.

[0267] The first report generation module is configured to use the management agent to execute the verification sub-tasks corresponding to the verification analysis plan, obtain the task execution results, and generate a task analysis report corresponding to the target dialogue analysis task based on the task execution results.

[0268] The above is an illustrative scheme of a data processing apparatus applied to a dialogue analysis task according to this embodiment. It should be noted that the technical solution of this data processing apparatus applied to a dialogue analysis task belongs to the same concept as the technical solution of the data processing method applied to a dialogue analysis task described above. Details not described in detail in the technical solution of the data processing apparatus applied to a dialogue analysis task can be found in the description of the technical solution of the data processing method applied to a dialogue analysis task described above.

[0269] In another possible implementation, one or more embodiments of this specification also provide a multi-agent processing system, including user agents and management agents, wherein:

[0270] The user intelligent agent generates a target user profile corresponding to the target analysis task based on the target analysis task and the historical analysis task, wherein the target analysis task carries corresponding task data.

[0271] The management agent generates a task execution plan corresponding to the target analysis task based on the target analysis task, the task data, and the target user profile.

[0272] The management agent determines an analysis model and generates a verification analysis plan and corresponding verification subtasks based on the task execution plan. The analysis model is determined by the management agent based on the analysis results of the task execution plan.

[0273] The management agent executes the verification sub-tasks corresponding to the verification analysis plan, obtains the task execution results, and generates a task analysis report corresponding to the target analysis task based on the task execution results.

[0274] Specifically, the implementation steps of the user agent and the management agent in this multi-agent processing system are the same as those of the user agent and the management agent in the above data processing method, and will not be repeated here.

[0275] The above is an illustrative scheme of a multi-agent processing system according to this embodiment. It should be noted that the technical solution of this multi-agent processing system and the technical solution of the data processing method described above belong to the same concept. For details not described in detail in the technical solution of the multi-agent processing system, please refer to the description of the technical solution of the data processing method described above.

[0276] Figure 6 A structural block diagram of a computing device 600 provided according to one embodiment of this specification is shown.

[0277] The computing device 600 includes:

[0278] Memory 610 and processor 620;

[0279] The memory 610 is used to store computer programs / instructions, and the processor 620 is used to execute the computer programs / instructions, which, when executed by the processor 620, implement the steps of the data processing method.

[0280] In one or more embodiments of this specification, the computing device can be understood as an integrated smart terminal, including but not limited to a server, desktop computer, PC (Personal Computer), all-in-one model machine, mobile phone, tablet computer or other portable smart terminal, etc., and the computing device may have the model described in the above embodiments of this application pre-installed.

[0281] Specifically, this computing device can pre-install various types of models, including but not limited to models in natural language processing, visual processing, speech processing, code processing, and multimodal task processing, thus providing diverse model selection. In different product forms, this computing device can support one or more model usage methods, including but not limited to model training, model invocation, model fine-tuning, model deployment, model inference, and application. In some product forms, this computing device also supports model management, including but not limited to multi-type model management (supporting the management of discriminative, generative, and other model types), model version control (supporting the control of different model versions), and model evaluation (evaluating model performance and effectiveness based on model evaluation tools). In other product forms, this computing device can also create applications based on models, providing API (Application Programming Interface) calling capabilities. Users can call models into created applications through the API interface, and application management tools are also provided to manage and monitor the applications.

[0282] Furthermore, the computing device can also include data management (supporting the creation and management of model tuning datasets), a training center (providing abundant training resources to help users learn and master AI (Artificial Intelligence) technology), and basic control capabilities (providing enterprise-level basic control capabilities to ensure the security and efficient operation of the system). Through the above functions, it provides a comprehensive and integrated device for AI development, training, deployment, and application.

[0283] Figure 7 A structural block diagram of an electronic device 700 provided according to one embodiment of this specification is shown.

[0284] A memory 710 and a processor 720 are connected via a bus 730;

[0285] The memory 710 is used to store computer programs / instructions, and the processor 720 is used to execute the computer programs / instructions, which, when executed by the processor 720, implement the steps of the method.

[0286] Specifically, the components of the electronic device 700 include, but are not limited to, a memory 710 and a processor 720. The processor 720 is connected to the memory 710 via a bus 730, and the database 750 is used to store data.

[0287] Electronic device 700 also includes access device 740, which enables electronic device 700 to communicate via one or more networks 760. Examples of these networks include Public Switched Telephone Network (PSTN), Local Area Network (LAN), Wide Area Network (WAN), Personal Area Network (PAN), or combinations of communication networks such as the Internet. Access device 740 may include one or more of any type of wired or wireless network interface (e.g., network interface card (NIC)), such as an IEEE 802.11 Wireless Local Area Network (WLAN) wireless interface, Wi-MAX (Worldwide Interoperability for Microwave Access) interface, Ethernet interface, Universal Serial Bus (USB) interface, cellular network interface, Bluetooth interface, Near Field Communication (NFC) interface, and so on.

[0288] In one embodiment of this specification, the above-described components of the electronic device 700 and Figure 7 Other components, not shown, can also be connected to each other, for example, via a bus. It should be understood that... Figure 7 The block diagram of the electronic device shown is for illustrative purposes only and is not intended to limit the scope of this specification. Those skilled in the art can add or replace other components as needed.

[0289] Electronic device 700 can be any type of stationary or mobile electronic device, including mobile computers or mobile electronic devices (e.g., tablet computers, personal digital assistants, laptop computers, notebook computers, netbooks, etc.), mobile phones (e.g., smartphones), wearable electronic devices (e.g., smartwatches, smart glasses, etc.) or other types of mobile devices, or stationary electronic devices such as desktop computers or personal computers (PCs). Electronic device 700 can also be a mobile or stationary server.

[0290] The above is an illustrative scheme of an electronic device according to this embodiment. It should be noted that the technical solution of this electronic device and the technical solution of the data processing method described above belong to the same concept. For details not described in detail in the technical solution of the electronic device, please refer to the description of the technical solution of the data processing method described above.

[0291] An embodiment of this specification also provides a computer-readable storage medium storing a computer program / instructions that, when executed by a processor, implement the steps of the above-described data processing method.

[0292] The above is an illustrative scheme of a computer-readable storage medium according to this embodiment. It should be noted that the technical solution of this storage medium and the technical solution of the data processing method described above belong to the same concept. For details not described in detail in the technical solution of the storage medium, please refer to the description of the technical solution of the data processing method described above.

[0293] An embodiment of this specification also provides a computer program product, including a computer program / instructions that, when executed by a processor, implement the steps of the above-described data processing method.

[0294] The above is an illustrative scheme of a computer program product according to this embodiment. It should be noted that the technical solution of this computer program product and the technical solution of the data processing method described above belong to the same concept. For details not described in detail in the technical solution of the computer program product, please refer to the description of the technical solution of the data processing method described above.

[0295] The foregoing has described specific embodiments of this specification. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps recited in the claims may be performed in a different order than that shown in the embodiments and may still achieve the desired result. Furthermore, the processes depicted in the drawings do not necessarily require the specific or sequential order shown to achieve the desired result. In some embodiments, multitasking and parallel processing are possible or may be advantageous.

[0296] The computer program / instructions include computer program code, which may be in the form of source code, object code, executable file, or certain intermediate forms. The computer-readable medium may include: any entity or device capable of carrying the computer program code, recording media, USB flash drive, portable hard drive, magnetic disk, optical disk, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signals, telecommunication signals, and software distribution media, etc. It should be noted that the content included in the computer-readable medium may be appropriately added or removed according to the requirements of patent practice. For example, in some regions, according to patent practice, computer-readable media may not include electrical carrier signals and telecommunication signals.

[0297] It should be noted that, for the sake of simplicity, the foregoing method embodiments are all described as a series of actions. However, those skilled in the art should understand that the embodiments in this specification are not limited to the described order of actions, because according to the embodiments in this specification, some steps can be performed in other orders or simultaneously. Furthermore, those skilled in the art should also understand that the embodiments described in this specification are all preferred embodiments, and the actions and modules involved are not necessarily essential to the embodiments in this specification.

[0298] In the above embodiments, the descriptions of each embodiment have different focuses. For parts not described in detail in a certain embodiment, please refer to the relevant descriptions of other embodiments.

[0299] The preferred embodiments disclosed above are merely illustrative of this specification. The optional embodiments do not exhaustively describe all details, nor do they limit the invention to the specific implementations described. Clearly, many modifications and variations can be made based on the embodiments described herein. These embodiments are selected and specifically described in this specification to better explain the principles and practical applications of the embodiments, thereby enabling those skilled in the art to better understand and utilize this specification. This specification is limited only by the claims and their full scope and equivalents.

Claims

1. A data processing method, comprising: A target analysis task is determined, and a target user profile is generated through multi-turn dialogue using a user intelligent agent. The target analysis task carries corresponding task data, and the target user profile is generated by automatically parsing the multi-turn dialogue and recognizing user intent. Based on the target analysis task, the task data, and the target user profile, a management agent is used to generate a task execution plan corresponding to the target analysis task. The task execution plan is parsed using a task decomposition agent to determine multiple analysis dimensions of the task execution plan; multiple analysis models are determined based on the multiple analysis dimensions, and the task execution plan is input into the multiple analysis models respectively to obtain multiple verification analysis plans, wherein the verification analysis plans are the attribution hypotheses output by the analysis models, and the multiple analysis models are heterogeneous analysis models; the multiple verification analysis plans are decomposed to obtain multiple verification subtasks corresponding to the multiple verification analysis plans, wherein the analysis models are determined by the management agent based on the parsing results of the task execution plan, and the management agent includes the task decomposition agent; Using the management agent, multiple verification sub-tasks corresponding to the verification analysis plan are executed, the task execution results are obtained, and a task analysis report corresponding to the target analysis task is generated based on the task execution results. The execution results of each verification subtask are shared, and the verification analysis plan corresponding to each verification subtask and other unexecuted verification subtasks in the verification analysis plan are dynamically updated.

2. The data processing method according to claim 1, wherein determining the target analysis task and generating a target user profile corresponding to the target analysis task using a user agent includes: Receive the target analysis task and the task data corresponding to the target analysis task; The user intelligent agent is used to parse the target analysis task to determine the task type of the target analysis task; Using the user agent, historical analysis tasks matching the task type of the target analysis task are selected from the memory unit according to the task type of the target analysis task; Based on the target analysis task and the historical analysis task, a target user profile corresponding to the target analysis task is generated using a user agent.

3. The data processing method according to claim 2, wherein generating a target user profile corresponding to the target analysis task using a user agent based on the target analysis task and the historical analysis task includes: The user intelligent agent is used to analyze the target analysis task and determine the task requirements of the target analysis task. Using the user intelligent agent, determine the historical user profile corresponding to the historical analysis task; Using the user intelligent agent, the task requirements of the target analysis task are compared with the historical user profiles, and a target user profile corresponding to the target analysis task is generated based on the comparison results.

4. The data processing method according to claim 1, wherein the management agent includes a data agent; Before generating the task execution plan corresponding to the target analysis task using the management agent based on the target analysis task, the task data, and the target user profile, the method further includes: The data intelligence agent is used to parse the task data to obtain a summary information of the task data; The step of generating a task execution plan corresponding to the target analysis task using a management agent based on the target analysis task, the task data, and the target user profile includes: Based on the target analysis task, the data summary information, and the target user profile, the management agent generates a task execution plan corresponding to the target analysis task.

5. The data processing method according to claim 4, wherein the management agent includes a demand management agent; The step of generating a task execution plan corresponding to the target analysis task using the management agent based on the target analysis task, the data summary information, and the target user profile includes: Based on the target analysis task, the data summary information, and the target user profile, the demand management agent generates an initial execution plan and performs information gap detection on the initial execution plan to obtain the information gap detection results. If it is determined from the information gap detection results that there is no information gap in the initial execution plan, the initial execution plan is determined as the task execution plan corresponding to the target analysis task; or If, based on the information gap detection results, it is determined that there is an information gap in the initial execution plan, the user agent is used to supplement the initial execution plan to generate a task execution plan corresponding to the target analysis task.

6. The data processing method according to claim 1, wherein the management agent includes a task execution agent; The process of using the management agent to execute multiple verification sub-tasks corresponding to the verification analysis plan and obtain task execution results includes: Using the task execution agent, a target verification subtask is determined from multiple verification subtasks corresponding to the multiple verification analysis plans according to a preset execution order, and corresponding task execution resources are allocated to the target verification subtask. Using the task execution agent, the target verification subtask is executed according to the task execution resources corresponding to the target verification subtask, and the subtask execution result of the target verification subtask is obtained; Using the task execution agent, if it is determined that all verification subtasks corresponding to the multiple verification analysis plans have been completed, the task execution result is determined based on the subtask execution results of the target verification subtask.

7. The data processing method according to claim 6, wherein sharing the task execution results of each verification subtask and dynamically updating the verification analysis plan corresponding to each verification subtask and other unexecuted verification subtasks of the verification analysis plan includes: Based on the subtask execution results of the target verification subtask, update the verification analysis plan corresponding to the target verification subtask and other verification subtasks, wherein the other verification subtasks are the unexecuted verification subtasks corresponding to the verification analysis plan corresponding to the target verification subtask, excluding the target verification subtask.

8. A data processing method for dialogue analysis tasks, comprising: A target dialogue analysis task is determined, and a target user profile is generated through multi-turn dialogue using a user intelligent agent. The target dialogue analysis task carries corresponding dialogue analysis data, and the target user profile is generated by automatically parsing the multi-turn dialogue and recognizing user intent. Based on the target dialogue analysis task, the dialogue analysis data, and the target user profile, a management agent is used to generate a dialogue analysis task execution plan corresponding to the target dialogue analysis task. The task execution plan is parsed using a task decomposition agent to determine multiple analysis dimensions of the task execution plan; multiple analysis models are determined based on the multiple analysis dimensions, and the task execution plan is input into the multiple analysis models respectively to obtain multiple verification analysis plans, wherein the verification analysis plans are the attribution hypotheses output by the analysis models, and the multiple analysis models are heterogeneous analysis models; the multiple verification analysis plans are decomposed to obtain multiple verification subtasks corresponding to the multiple verification analysis plans, wherein the analysis models are determined by the management agent based on the parsing results of the dialogue analysis task execution plan, and the management agent includes the task decomposition agent; Using the management agent, the verification sub-tasks corresponding to the verification analysis plan are executed, the task execution results are obtained, and a task analysis report corresponding to the target dialogue analysis task is generated based on the task execution results. The execution results of each verification subtask are shared, and the verification analysis plan corresponding to each verification subtask and other unexecuted verification subtasks in the verification analysis plan are dynamically updated.

9. A multi-agent processing system, comprising a user agent and a management agent, wherein: The user intelligent agent generates a target user profile corresponding to the target analysis task through multi-turn dialogue. The target analysis task carries corresponding task data, and the target user profile is generated by automatically parsing the multi-turn dialogue and recognizing user intent. The management agent generates a task execution plan corresponding to the target analysis task based on the target analysis task, the task data, and the target user profile. The management agent includes a task decomposition agent, which analyzes the task execution plan to determine multiple analysis dimensions of the task execution plan; determines multiple corresponding analysis models based on the multiple analysis dimensions; inputs the task execution plan into the multiple analysis models to obtain multiple verification analysis plans, wherein the verification analysis plans are attribution hypotheses output by the analysis models, and the multiple analysis models are heterogeneous analysis models; and decomposes the multiple verification analysis plans to obtain multiple verification subtasks corresponding to the multiple verification analysis plans, wherein the analysis models are determined by the management agent based on the analysis results of the task execution plan. The management agent executes the verification sub-tasks corresponding to the verification analysis plan, obtains the task execution results, and generates a task analysis report corresponding to the target analysis task based on the task execution results. The management agent shares the task execution results of each verification subtask and dynamically updates the verification analysis plan corresponding to each verification subtask as well as other unexecuted verification subtasks in the verification analysis plan.

10. A computing device, comprising: Memory and processor; The memory is used to store computer programs / instructions, and the processor is used to execute the computer programs / instructions, which, when executed by the processor, implement the steps of the method according to any one of claims 1 to 8.

11. An electronic device, comprising: A memory and a processor, the memory and the processor being connected via a bus; The memory is used to store computer programs / instructions, and the processor is used to execute the computer programs / instructions, which, when executed by the processor, implement the steps of the method according to any one of claims 1 to 8.

12. A computer-readable storage medium storing a computer program / instructions that, when executed by a processor, implement the steps of the method according to any one of claims 1 to 8.

13. A computer program product comprising a computer program / instructions that, when executed by a processor, implement the steps of the method according to any one of claims 1 to 8.

Images