Data analysis method, device and equipment, readable storage medium and program product

Abstract

The invention provides a data analysis method and device, equipment, a readable storage medium and a program product. The scheme comprises the steps of obtaining data analysis target information in a natural language form of a user; then, the data analysis target information is input into a first large language model, first task sequence information obtained by splitting the data analysis target information through the first large language model is obtained, and the first task sequence information comprises multiple pieces of task information and task execution logic information of the multiple pieces of task information; then calling a plurality of agents to execute the plurality of pieces of task information according to the task execution logic information to obtain a plurality of pieces of task result information generated by the plurality of agents; and finally, based on the multiple pieces of task result information, determining a data analysis result corresponding to the data analysis target information.

Description

Technical Field

[0001] The present invention relates to the field of data analysis technology, and in particular to a data analysis method. The present invention also relates to a data analysis device, a computing device, a computer-readable storage medium, and a computer program product. Background Art

[0002] Currently, intelligent data analysis products typically require multiple rounds of human-machine collaboration to complete analytical tasks. For example, for indicator attribution, the typical process involves configuring data sources, creating datasets, creating metrics, creating reports, conducting statistical analysis, trend analysis, fluctuation analysis, anomaly detection, distribution analysis, and finally manually summarizing these steps into an analytical report. Because users must independently plan multiple rounds of interaction, the data analysis barrier is high. Furthermore, because users must perform multiple rounds of interaction and manually summarize the results of each round, data analysis efficiency is low.

[0003] In view of this, it is necessary to provide a simple, easy-to-use and efficient data analysis method. Summary of the Invention

[0004] In view of this, embodiments of the present application provide data analysis methods, devices, equipment, and computer-readable media to solve the problems of high threshold and low efficiency of existing data analysis methods.

[0005] According to a first aspect of an embodiment of the present application, a data analysis method is provided, which is applied to a server and includes:

[0006] Obtaining data analysis target information in the form of natural language from the user;

[0007] Inputting the data analysis target information into a first large language model to obtain first task sequence information obtained by decomposing the data analysis target information by the first large language model; the first task sequence information includes a plurality of task information and task execution logic information of the plurality of task information;

[0008] Calling multiple agents to execute the multiple task information according to the task execution logic information, and obtaining multiple task result information generated by the multiple agents;

[0009] Based on the plurality of task result information, a data analysis result corresponding to the data analysis target information is determined.

[0010] According to a second aspect of an embodiment of the present application, a data analysis method is provided, which is applied to a user terminal and includes:

[0011] receiving data analysis target information in the form of natural language input by a user;

[0012] The data analysis target information is sent to a server deployed with a first large language model; the first large language model is used to decompose the data analysis target information to obtain first task sequence information; the first task sequence information includes multiple task information and task execution logic information of the multiple task information; the multiple task information is used to be executed by multiple agents according to the task execution logic information to generate data analysis results corresponding to the data analysis target information;

[0013] Obtaining the data analysis result fed back by the server;

[0014] Display the data analysis results on the page of the user terminal

[0015] According to a third aspect of an embodiment of the present application, a data analysis device is provided, which is applied to a server and includes:

[0016] A data analysis target information acquisition module is used to obtain the user's data analysis target information in the form of natural language;

[0017] a target decomposition module, configured to input the data analysis target information into a first large language model, and obtain first task sequence information obtained by decomposing the data analysis target information by the first large language model; the first task sequence information includes a plurality of task information and task execution logic information of the plurality of task information;

[0018] A task result information determination module is used to call multiple agents to execute the multiple task information according to the task execution logic information, and obtain multiple task result information generated by the multiple agents;

[0019] The data analysis result generating module is used to determine the data analysis result corresponding to the data analysis target information based on the multiple task result information.

[0020] According to a fourth aspect of an embodiment of the present application, a data analysis device is provided, which is applied to a user terminal and includes:

[0021] A data analysis target information receiving module is used to receive data analysis target information in natural language form input by a user;

[0022] a data analysis target information sending module, configured to send the data analysis target information to a server deployed with a first large language model; the first large language model is configured to decompose the data analysis target information to obtain first task sequence information; the first task sequence information includes a plurality of task information and task execution logic information for the plurality of task information; the plurality of task information is configured to be executed by a plurality of intelligent agents according to the task execution logic information to generate data analysis results corresponding to the data analysis target information;

[0023] A data analysis result acquisition module, configured to acquire the data analysis result fed back by the server;

[0024] The data analysis result display module is used to display the data analysis result on the page of the user terminal.

[0025] According to a fifth aspect of an embodiment of the present application, a computing device is provided, comprising a memory, a processor, and computer instructions stored in the memory and executable on the processor, wherein the processor implements the steps of the data analysis method when executing the computer instructions.

[0026] According to a sixth aspect of an embodiment of the present application, a computer-readable storage medium is provided, which stores computer instructions, and when the computer instructions are executed by a processor, the steps of the data analysis method are implemented.

[0027] According to a seventh aspect of the embodiments of this specification, a computer program product is provided, comprising a computer program / instruction, which implements the steps of the above-mentioned data analysis method when executed by a processor.

[0028] One embodiment of the present specification can at least achieve the following beneficial effects: by decomposing the user's data analysis target information in the form of natural language into tasks, task sequence information including multiple task information and task execution logic information for the multiple task information is obtained, and then multiple intelligent agents are called to execute the multiple task information according to the task execution logic information to obtain multiple task result information, and then the data analysis results corresponding to the data analysis target information are determined based on the multiple task result information to feed back to the user. In view of this, in response to a natural language question asked by the user, the task can be automatically decomposed and multiple intelligent agents can be called to obtain the execution results of each task, and then the results can be automatically summarized and fed back to the user. The user does not need to understand how to decompose the task, does not need to issue data analysis instructions multiple times, and does not need to integrate the data analysis results of multiple intelligent agents. The operation threshold for users to obtain data analysis results is low and the overall execution efficiency is high. BRIEF DESCRIPTION OF THE DRAWINGS

[0029] In order to more clearly illustrate the embodiments of this specification or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are only some embodiments recorded in this application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative labor.

[0030] Figure 1A schematic diagram of an application scenario of an Agentic AI system for data analysis provided in an embodiment of this specification;

[0031] Figure 2 A flowchart of a data analysis method provided in an embodiment of this specification;

[0032] Figure 3 A schematic diagram of a plan execution process of an intelligent agent provided in an embodiment of this specification;

[0033] Figure 4 A schematic diagram of a data processing flow of an intelligent agent in an actual application scenario provided in an embodiment of this specification;

[0034] Figure 5 A flowchart of a data analysis method provided in an embodiment of this specification;

[0035] Figure 6 A schematic diagram of a user terminal page during a data analysis process provided in an embodiment of this specification;

[0036] Figure 7 The embodiments of this specification provide corresponding Figure 2 A structural diagram of a data analysis device;

[0037] Figure 8 The embodiments of this specification provide corresponding Figure 5 A structural diagram of a data analysis device;

[0038] Figure 9 A structural block diagram of a computing device provided in an embodiment of this specification. DETAILED DESCRIPTION

[0039] The following description sets forth many specific details to facilitate a thorough understanding of the present application. However, the present application can be implemented in many other ways than those described herein, and those skilled in the art can make similar generalizations without violating the scope of the present application. Therefore, the present application is not limited to the specific implementations disclosed below.

[0040] The terms used in one or more embodiments of the present application are for the purpose of describing specific embodiments only and are not intended to limit one or more embodiments of the present application. The singular forms "a", "the" and "the" used in one or more embodiments of the present application and the appended claims are also intended to include plural forms, unless the context clearly indicates otherwise. It should also be understood that the term "and / or" used in one or more embodiments of the present application refers to and includes any or all possible combinations of one or more associated listed items.

[0041] It should be understood that although the terms first, second, etc. may be used to describe various information in one or more embodiments of the present application, such information should not be limited to these terms. These terms are only used to distinguish information of the same type from each other. For example, without departing from the scope of one or more embodiments of the present application, the first may also be referred to as the second, and similarly, the second may also be referred to as the first. Depending on the context, the word "if" as used herein may be interpreted as "at the time of" or "when" or "in response to determining".

[0042] 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 this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of relevant data must comply with the relevant laws, regulations and standards of the relevant regions, and provide corresponding operation entrances for users to choose to authorize or refuse.

[0043] Currently, intelligent data analysis products typically require multiple rounds of human-machine collaboration to complete analytical tasks. For example, for indicator attribution, the typical process involves configuring data sources, creating datasets, creating metrics, creating reports, conducting statistical analysis, trend analysis, fluctuation analysis, anomaly detection, distribution analysis, and finally manually summarizing these steps into an analytical report.

[0044] During multiple rounds of human-machine collaboration to complete analysis tasks, on the one hand, the entry threshold for data analysis is high. For example, data analysis users need to understand professional data analysis knowledge such as data sources, data sets, reports, statistical analysis, trend analysis, fluctuation analysis, and anomaly detection. On the other hand, the data analysis tasks that can be achieved in each interaction are single and of low quality. For example, data can only be retrieved through simple text2SQL, and then the results can be visualized and summarized with simple conclusions. In addition, the data analysis process of multiple interactions is fragmented, and the input and output products of each link cannot be effectively connected in series. Data analysts need to manually summarize the output products of multiple links to obtain the final analysis report.

[0045] In view of this, in the embodiments of this specification, an agent-based artificial intelligence (Agentic AI) system that combines artificial intelligence and data analysis technology is provided, which can independently complete the whole process of operations such as understanding the user's data analysis needs, breaking down the user's data analysis needs into tasks, dynamically planning and executing tasks, generating and delivering data analysis results, etc.

[0046] Unlike the related art that requires users to input a single execution instruction multiple times and obtain a single execution result, and then manually summarize the execution results, in actual application, the embodiment of this specification only requires users to provide data analysis target information in natural language form once. The agentic AI system provided by the embodiment of this application can autonomously plan and complete tasks that originally required multiple rounds of human-computer collaboration, thereby delivering data analysis needs with high quality and efficiency.

[0047] Taking indicator attribution as an example, if the analysis task is completed through multiple rounds of human-computer collaboration, it usually takes a data analysis user a day to complete. Not only is the data analysis threshold high, but the data analysis efficiency is very low. The solution based on the embodiment of the present application may only require the staff to input the data analysis target information in natural language form once, and the data analysis results can be delivered within 5 minutes.

[0048] First, the terms involved in one or more embodiments of the present application are explained.

[0049] Agentic AI (Agentic AI) is capable of independently completing complex, multi-step tasks through autonomous perception, reasoning, planning, and execution. Compared to traditional AI tools, it possesses proactiveness, autonomy, and environmental adaptability, enabling decision-making and interaction without direct human control. Agentic AI is a multi-agent architecture that leverages the collaboration of multiple specialized, modular agents to achieve complex, high-level goals. Agents are coordinated through a centralized coordinator or decentralized protocols to form an agent network. Agentic AI represents a form of system-level intelligence characterized by dynamic inter-agent collaboration. Agentic AI is capable of goal decomposition, automatically parsing user-specified target tasks and dividing them into smaller, more manageable subtasks. These subtasks are then assigned to the agent network. Agentic AI's reflective reasoning and memory capabilities enable it to store context across multiple interactions with the agent network, evaluate past decisions, and iteratively optimize its strategies. Agentic AI facilitates the dynamic sequencing of these subtasks through multi-step reasoning and planning mechanisms, enabling the system to adapt in real time to environmental changes or partial task failures, ensuring robust task execution even under uncertainty.

[0050] The A2A protocol, short for Agent-to-Agent protocol, enables AI agents to securely exchange information and coordinate actions across different enterprise platforms or applications. The A2A protocol's goal is to standardize communication between different agents. In other words, the A2A protocol shields differences in underlying technologies, enabling agents to seamlessly collaborate regardless of the underlying technology, automating complex enterprise workflows. The A2A protocol uses HTTP / HTTPS as the transport layer and JSON-RPC 2.0 for message encapsulation.

[0051] The Model Context Protocol (MCP) is an open standard launched by Anthropic at the end of November 2024. It aims to unify the communication protocols between large models and external data sources and tools. The main purpose of MCP is to solve the problem that current AI models cannot fully realize their potential due to the limitations of data silos. MCP enables AI applications to securely access and operate local and remote data, providing AI applications with an interface that connects everything. MCP defines a common format that allows AI models to dispatch search engines, databases, calculators, code executors, and even other models or API services like calling functions. Through the MCP protocol, AI applications can easily introduce external services, functions, or retrieve more data, thereby having richer capabilities.

[0052] The technical solutions provided by the embodiments of this specification are described in detail below with reference to the accompanying drawings.

[0053] Figure 1 A schematic diagram of an application scenario of an Agentic AI system for data analysis provided in an embodiment of this specification.

[0054] like Figure 1 As shown, the Agentic AI system provided in the embodiment of this specification includes a planning engine server 10, a data analysis agent server 20, and a data analysis tool server 30. The user terminal 40 can access the Agentic AI system by interacting with the planning engine server 10 through a network.

[0055] Among them, a task planner can be deployed on the planning engine server 10. The task planner is built based on a large language model, for example, it can be built based on the DeepSeek series large model, the GPT series large model, the Tongyi Qianwen series large model, etc. The task planner can use the large language model to decompose the user's data analysis target information into task sequences, dynamic planning task sequences, etc. Figure 1For example, the task planner can decompose the data analysis target information into a task sequence including a first task Task1, a second task Task2, a third task Task3, a fourth task Task4, and a fifth task Task5 connected according to a certain execution logic.

[0056] Multiple agents can be deployed on the data analysis agent server 20. In actual applications, multiple agents can be deployed in one or more servers. Figure 1 An example of an intelligent agent is shown in FIG, which may include a step planner. The step planner is constructed based on a large language model, for example, a DeepSeek series large model, a GPT series large model, a Tongyi Qianwen series large model, etc. The step planner decomposes the received task information into a step sequence, a dynamic planning step sequence, etc. by using a large language model. Figure 1 In the example, the step planner in the agent can decompose the task information into a step sequence including the first step Step1, the second step Step2, and the third step Step3 connected according to a certain execution logic.

[0057] Multiple data analysis tools may be deployed on the data analysis tool server 30. In actual application, multiple data analysis tools may be deployed in one or more servers.

[0058] In such Figure 1 In the application scenario shown, the server can be connected to one or more user terminals via a local area network connection, a wide area network connection, an Internet connection, or other types of data networks. Servers can communicate with each other via a local area network connection, a wide area network connection, an Internet connection, or other types of data networks. Figure 1 The server may include but is not limited to any device, equipment, platform, equipment cluster, etc. with computing and processing capabilities. Figure 1 The user terminals in the present invention may include but are not limited to smart phones, tablet computers, laptops, PDAs, personal computers, smart home devices, car-mounted devices, etc.

[0059] In actual application, the user can interact with the planning engine server 10 through the user terminal 40 to call the Agentic AI system, thereby achieving the purpose of data analysis.

[0060] The planning engine server 10 can receive the user's data analysis target information in the natural language form sent by the user terminal; then the data analysis target information can be disassembled using a large language model to obtain task sequence information containing a series of task information for realizing the data analysis target information; thereafter, multiple agents in the data analysis agent server 20 can be called to execute each task information in the task sequence information.

[0061] For example, the user terminal sends the user's data analysis target information in the form of natural language as "Help me analyze this data and generate a report and analysis report". The planning engine server 10 may decompose the data analysis target information into task sequence information including multiple task information such as "data preparation", "create report", "generate data analysis ideas", "generate analysis ideas", "generate analysis report", etc., and call the corresponding agent in the data analysis agent server 20 to execute the task sequence information.

[0062] The data analysis agent server 20 can respond to the call of the planning engine server 10 and use the agents corresponding to the respective task information to implement the respective task information. For example, the "data preparation agent" can be called to execute the task information of "data preparation", the "report making agent" can be called to execute the task information of "creating report", the "number-reading thinking agent" can be called to execute the task information of "generating number-reading thinking", the "deep analysis agent" can be called to execute the task information of "generating analysis thinking", the "report generation agent" can be called to execute the task information of "generating analysis report", and so on. Thus, by executing multiple task information in sequence according to the task execution logic information in the task sequence information, multiple task result information generated by multiple agents can be obtained.

[0063] In actual application, the planning engine server 10 can call the data analysis agent server 20 multiple times, and each call instructs the data analysis agent server 20 to execute a task information, and obtains the task result information generated by executing the task information fed back by the data analysis agent server 20; then, the planning engine server 10 can dynamically adjust the task sequence information according to the execution result of the previous task information, so as to determine the next task information to be executed.

[0064] In the embodiments of the present specification, the planning engine server 10 and the data analysis agent server 20 can communicate based on the A2A protocol, thereby making it possible for the input and output information of different agents to follow a unified format, facilitating the introduction of the output of the previous agent into the processing logic for the next agent, and realizing the synergistic effect of using multiple agents to solve the data analysis target information.

[0065] Furthermore, any agent in the data analysis agent server 20 can further decompose the task information in the process of executing the corresponding task information to obtain step sequence information containing a series of data processing step information for implementing the task information; thereafter, multiple data analysis tools in the data analysis tool server 30 can be called to execute each data processing step information in the step sequence information.

[0066] For example, when the "Kanshu Sigui Intelligent Agent" executes the task information of "generating Kanshu Sigui", the "Kanshu Sigui Intelligent Agent" may decompose the task information into step sequence information including multiple data processing step information such as "opening the report in the browser", "generating Kanshu Sigui report content", and "publishing the report in the browser", and call the corresponding tool in the data analysis tool server 30 to execute the step sequence information.

[0067] The data analysis tool server 30 can respond to the call of the data analysis agent server 20 and use the data analysis tools corresponding to each step sequence information to execute the data processing step information. For example, the "browser-open report" tool can be called to execute the data processing step information of "browser open report", the "generate view number idea report content" tool can be called to execute the data processing step information of "generate view number idea report content", the "browser-publish report" tool can be called to execute the data processing step information of "browser-publish report", and the like. Thus, by sequentially executing multiple data processing step information according to the step execution logic information in the step sequence information, multiple step execution results generated by multiple data analysis tools can be obtained.

[0068] In actual application, the data analysis intelligent agent server 20 can call the data analysis tool server 30 multiple times, and each call instructs the data analysis tool server 30 to execute a data processing step information, and obtains the step execution result generated by executing the data processing step information fed back by the data analysis tool server 30; then, the data analysis intelligent agent server 20 can dynamically adjust and plan the step sequence information according to the step execution result corresponding to the previous data processing step information, so as to determine the next data processing step information to be executed.

[0069] In an embodiment of the present specification, the data analysis agent server 20 and the data analysis tool server 30 can communicate based on MCP, thereby ensuring that the input and output information of different data analysis tools follow a unified format, making it easier to introduce the output of the previous data analysis tool into the processing logic for the next data analysis tool, thereby achieving a synergistic effect of using multiple data analysis tools to solve task information.

[0070] The Agentic AI system for data analysis based on the embodiment of this specification can integrate or call a rich set of intelligent agents with various task implementation functions and a rich set of data analysis tools with various data processing functions through a unified protocol. In practical applications, it can be applied to a variety of application scenarios such as data report generation, data depth analysis, abnormal indicator attribution, networked data analysis, etc., but is not limited to this. For example, in the data report generation scenario, based on the system's ability to support the input of multiple types of data sources and the ability to support the generation of multiple types of data charts, it can efficiently generate high-quality data reports based on the data analysis target information in the form of natural language input by the user at one time. For example, in the data depth analysis scenario, based on the system's support for multiple types of intelligent analysis tools, such as statistical analysis, descriptive analysis, predictive analysis, diagnostic analysis and other intelligent analysis tools, all intermediate links can be combined to automatically generate complete and in-depth reports. For example, in the abnormal indicator attribution scenario, based on the system's support for the root cause location and analysis of abnormal points based on indicators, a deep attribution analysis report can be produced. For example, in the networked data analysis scenario, based on the system's support for multi-channel data collection and light processing capabilities such as networked data, professional data, and platform data, it can effectively complete multi-channel data joint analysis and obtain comprehensive data analysis results.

[0071] The Agentic AI system for data analysis based on the embodiments of this specification can provide data analysis results in the form of online products or offline products. Online products can include but are not limited to various forms such as data sets, workbooks, reports, and analysis reports, and offline products can include but are not limited to various forms such as PDF files, Excel files, and Markdown files.

[0072] In the present application, a data analysis method is provided. The present application also relates to a data analysis apparatus, a computing device, a computer-readable storage medium and a computer program product, which are described in detail one by one in the following embodiments.

[0073] Figure 2 A flowchart of a data analysis method provided in an embodiment of this specification.

[0074] From a program perspective, the execution body of the process can be a program installed on a server. It is understood that the method can be executed by any device, equipment, platform, or device cluster with computing and processing capabilities.

[0075] like Figure 2 As shown, the process may include the following steps:

[0076] Step 202: Obtain data analysis target information in the user's natural language format.

[0077] The data analysis target information may be information input by the user in the user interaction interface of the user terminal for instructing data analysis. The user terminal may send the data analysis target information to the server to trigger the embodiment of this specification. Figure 2 The method flow shown.

[0078] In the embodiments of this specification, the data analysis goal expressed by the data analysis goal information may be a long-term data analysis goal. A long-term data analysis goal may refer to a data analysis goal for which the data analysis results cannot be directly obtained through data retrieval and aggregation. The process of achieving a long-term data analysis goal typically needs to be broken down into the implementation processes of multiple tasks, and the execution results of multiple tasks need to be integrated to obtain the result.

[0079] For example, "Help me analyze this data and generate a report and analysis report" describes a long-term data analysis goal. The realization of this goal requires obtaining the execution results of multiple tasks such as "data preparation", "report creation", "generating data analysis ideas", "generating analysis ideas", and "generating analysis reports" in sequence.

[0080] Step 204: Input the data analysis target information into the first large language model to obtain first task sequence information obtained by decomposing the data analysis target information by the first large language model; the first task sequence information includes multiple task information and task execution logic information of the multiple task information.

[0081] In actual application, prompt words can be constructed based on the data analysis target information, and then the constructed prompt words are input into the first large language model to obtain first task sequence information output by the first large language model based on the indication of the prompt words.

[0082] Furthermore, the content of the first language model inputted via the prompt word may also include dataset characteristic information corresponding to the data analysis target information. This dataset characteristic information is used to indicate characteristics of the dataset to be analyzed, targeted by the data analysis target information. In practical applications, this dataset characteristic information can be used to reflect field information contained in the dataset to be analyzed, association information between fields, and the like.

[0083] Furthermore, the content of the first language model input through the prompt word may also include data analysis knowledge. Optionally, the data analysis knowledge may include task planning logic knowledge, task planning example knowledge, and other knowledge used to assist the first language model in task planning. Optionally, the data analysis knowledge may include domain knowledge used to assist the first language model in planning different task sequence information for data analysis target information in different fields. The type of data analysis knowledge is not limited to this and can be provided according to actual business needs.

[0084] In an optional embodiment, the first large language model may refer to a preset data analysis template when decomposing the data analysis target information.

[0085] Specifically, the data analysis method may further include: selecting a data analysis template that matches the data analysis target information from a preset template set; the data analysis template including a template task sequence and agent call instruction information corresponding to each template task in the template task sequence. Accordingly, inputting the data analysis target information into a first large language model to obtain first task sequence information obtained by decomposing the data analysis target information using the first large language model may specifically include: inputting the data analysis target information and the data analysis template into the first large language model to obtain first task sequence information obtained by decomposing the data analysis target information using the data analysis template.

[0086] The data analysis templates in the preset template set may be pre-configured and published by a template configuration user. Specifically, the template configuration user may enter the data analysis template on the template configuration terminal and publish the entered data analysis template. The published data analysis template may be used by the template configuration user or other users. The published data analysis template may be stored in a template database, for example, on a server.

[0087] Furthermore, the template setting user enters the data analysis template in the template setting terminal, which may specifically include entering a data analysis framework. The data analysis framework may include a template task sequence and agent call instruction information corresponding to each template task in the template task sequence. When the data analysis template is used, the multiple template task sequences contained therein can be instantiated into specific task sequences.

[0088] Among them, each data analysis task can be in natural language form, semi-structured language form or structured language form. In the embodiments of this specification, a data analysis template including a data analysis task in natural language form can be called a natural language template; a data analysis template including a data analysis task in semi-structured form can be called a semi-structured template; and a data analysis template including a data analysis task in structured form can be called a structured template. In actual application, a data analysis template can include one or more data analysis tasks in natural language form, semi-structured form and structured form.

[0089] Furthermore, the data analysis template may be a structured template or a semi-structured template; the structured template or the semi-structured template includes a preset filling position for filling in task feature information.

[0090] Specifically, if the data analysis template includes a preset filling position for filling in task feature information; then, the first large language model decomposes the data analysis target information according to the data analysis template to obtain the first task sequence information, which may specifically include: extracting task feature information from the data analysis target information; filling the task feature information into the preset filling position to obtain the first task sequence information for realizing the data analysis target information.

[0091] In an optional embodiment, when setting up a data analysis template, the template user may further set routing rules corresponding to the data analysis template. The routing rules may be used as a basis for selecting the data analysis template. Alternatively, the data analysis template and its corresponding routing rules may be stored in correspondence. Thus, when a data analysis template is needed in a generation environment, the data analysis template that conforms to the routing rules may be called based on the routing rules of the data analysis template.

[0092] Specifically, the selecting of a data analysis template that matches the data analysis target information from the preset template set may specifically include: selecting a data analysis template that matches the data analysis target information from the preset template set according to the preset routing rules of each preset template in the preset template set; the preset routing rules are used to indicate the conditions that should be met for matching the corresponding preset template.

[0093] In an optional embodiment, when setting up a data analysis template, the template user can further specify the business domain of the data analysis template. This business domain can be used as a basis for selecting the data analysis template. Alternatively, the business domain can be used as a tag for the data analysis template. Thus, when a data analysis template is needed in a production environment, the data analysis template with the tag that meets the conditions can be called based on the data analysis template's tag.

[0094] In the embodiment of the present specification, the user who provides the data analysis target information and the user who sets the template may be the same user, or may be different users.

[0095] In an embodiment of the present specification, the data analysis templates in the preset template set may be data analysis templates in a template database. Furthermore, the data analysis templates in the preset template set may be at least some of the templates selected by a user from the template database. Thus, when the execution subject of an embodiment of the present specification selects a data analysis template, it may select from the preset template set. Since the preset template set reflects the user's preference information, the data analysis framework reflected in the resulting task sequence information is more consistent with the user's preferences.

[0096] Step 206: Call multiple agents to execute the multiple task information according to the task execution logic information, and obtain multiple task result information generated by the multiple agents.

[0097] The task execution logic information may be used to indicate the execution conditions of each task or the flow logic between different tasks.

[0098] like Figure 1 In the example, the task sequence information planned by the planning engine server 10 includes multiple task information connected according to a certain flow logic. The "data preparation" task is executed first; when the "data preparation" task is successfully executed, the "report creation" task is executed; when the "report creation" task is successfully executed, the "generate data analysis ideas" task and the "generate analysis ideas" task are executed; when the "generate data analysis ideas" task and the "generate analysis ideas" task are successfully executed, the "generate analysis report" task is executed.

[0099] Furthermore, the task execution logic information may also include dynamic planning logic information. For example, if an exception occurs during the execution of a task, such as the "data preparation" task failing to obtain the required data, how should the subsequent process be carried out? For example, a new data acquisition task can be planned and executed. For another example, if the "generate data analysis ideas" task and the "generate analysis ideas" task are only partially completed or the execution results obtained do not meet expectations, how should the subsequent process be carried out?

[0100] In actual application, each task information can be executed in sequence according to the order of the task information in the first task sequence information. One task information can call one intelligent agent to execute.

[0101] Specifically, calling multiple agents to execute the multiple task information according to the task execution logic information to obtain multiple task result information generated by the multiple agents can specifically include: calling a first agent to execute the first task information among the multiple task information to obtain the first task result information; determining the second task information based on the first task result information and the task execution logic information; calling a second agent to execute the second task information to obtain the second task result information.

[0102] Generally, different tasks can be performed by different agents. The second agent and the first agent can be different agents.

[0103] The second task information may be task information in the first task sequence information or task information in the second task sequence information; the second task sequence information may be obtained by updating the first task sequence information based on the first task result information.

[0104] In an optional embodiment, the first task result information may be first-type result information; the first-type result information may be used to represent task execution result information obtained after the first task information is successfully executed. Accordingly, determining the second task information based on the first task result information and the task execution logic information may specifically include: if the task execution result information satisfies the target execution condition corresponding to the target execution logic in the task execution logic information, then determining the first target task information corresponding to the target execution logic in the first task sequence information as the second task information.

[0105] Specifically, the first target task information downstream of the first task information in the first task sequence information can be determined as the second task information based on the task execution result information and the task execution logic information. When the first task information includes multiple downstream task information, it can be determined whether to execute the corresponding downstream task information based on the execution logic corresponding to each downstream task information. Optionally, only one downstream task information may be executed, multiple downstream task information may be executed at the same time, or there is no need to execute any downstream task information, thus ending the complete execution process of the task sequence information.

[0106] In an optional embodiment, the first task result information may be second-type result information; the second-type result information may be used to indicate task execution failure. Accordingly, determining the second task information based on the first task result information and the task execution logic information may specifically include: updating the first task sequence information based on the data analysis target information in response to the second-type result information to obtain second task sequence information; and determining the second target task information in the second task sequence information as the second task information.

[0107] To facilitate understanding, the following examples are given.

[0108] Assume that in step 202, the user's data analysis target information obtained is: help me query the transaction amount in Beijing in the past 30 days, then analyze the trend of the transaction amount, and then analyze the abnormal changes. If there are abnormal changes, generate a report and generate an analysis report.

[0109] Then, in step 204, the data analysis target information can be decomposed to obtain an execution plan expressed by the first task sequence information, assuming that the execution plan is as follows: Task 1: The data acquisition agent performs data preparation; Task 2: The analysis agent performs data analysis; Task 3: The reporting agent generates an analysis report.

[0110] Then, in step 206, the data acquisition agent can be called first to prepare the data; assuming that the data acquisition agent returns a successful data acquisition result, the analysis agent will continue to be called to perform trend and change analysis; assuming that the analysis agent returns an analysis result indicating that the indicator has changed, the subsequent tasks will continue to be executed according to the first task sequence information, and the reporting agent will be called to generate an analysis report.

[0111] Assuming that the analysis agent returns an analysis result indicating that the indicator has not changed, there is no need to call the reporting agent to generate an analysis report according to the first task sequence information. In this case, the task sequence information can be re-planned and re-executed.

[0112] If the data acquisition agent returns a data acquisition result indicating that the data acquisition failed, the task of analyzing the data and generating an analysis report in the first task sequence information cannot be continued. In this case, the task sequence information can be re-planned and re-executed.

[0113] Step 208: Based on the plurality of task result information, determine the data analysis result corresponding to the data analysis target information.

[0114] In actual application, optionally, a preset data integration program can be called to integrate the task result information fed back by each intelligent agent to obtain the final data analysis result.

[0115] Optionally, the task result information fed back by each agent can be provided to the large language model, so that the large language model can integrate the task result information to obtain the final data analysis result.

[0116] In practical applications, such as Figure 2 Steps 202 to 206 can be applied to Figure 1 Agentic AI systems for data analysis, more specifically, can be applied to Figure 1 The planning engine server 10 can obtain the data analysis target information in the natural language form of the user sent by the user terminal, decompose the data analysis target information to obtain first task sequence information including multiple task information and task execution logic information of the multiple task information, and then call multiple intelligent agents to execute the multiple task information according to the task execution logic information to obtain multiple task result information fed back by the multiple intelligent agents. Finally, the data analysis result corresponding to the data analysis target information can be determined based on the multiple task result information, and the data analysis result can be fed back to the user terminal.

[0117] Among them, the planning engine server 10 can continuously obtain the execution results of each agent in the process of calling the data analysis agent server 20 to execute task information. In actual application, the planning engine server 10 can stream the obtained execution results of each agent to the user terminal so that the user terminal can display them to the user through the user interaction interface.

[0118] It should be understood that in the methods described in one or more embodiments of this specification, the order of some steps can be adjusted according to actual needs, or some steps can be omitted.

[0119] Figure 2 The method decomposes the user's data analysis target information in the natural language form into tasks to obtain task sequence information including multiple task information and task execution logic information for the multiple task information, and then calls multiple agents to execute the multiple task information according to the task execution logic information to obtain multiple task result information, and then determines the data analysis results corresponding to the data analysis target information based on the multiple task result information to feed back to the user. In view of this, in response to a user's question in the natural language form, the task can be automatically decomposed and multiple agents can be called to obtain the execution results of each task, and then the results are automatically summarized and fed back to the user. The user does not need to understand how to decompose the task, does not need to issue data analysis instructions multiple times, and does not need to integrate the data analysis results of multiple agents. The operation threshold for the user to obtain data analysis results is low and the overall execution efficiency is high.

[0120] based on Figure 2The present specification also provides some improved implementation methods of the method, which are described below.

[0121] In one or more embodiments of the present specification, a specific solution is further provided for generating task result information based on task information after the intelligent agent is called.

[0122] Specifically, the calling of multiple intelligent agents to execute the multiple task information according to the task execution logic information to obtain multiple task result information generated by the multiple intelligent agents can specifically include: inputting the task information and the data set feature information corresponding to the task information into a second large language model to obtain step sequence information for implementing the task information determined by the second large language model, wherein the step sequence information includes multiple data processing step information and step execution logic information of the multiple data processing step information; calling multiple data analysis tools to execute the multiple data processing step information according to the step execution logic information to obtain multiple step execution results generated by the multiple data analysis tools; inputting the multiple step execution results into a third large language model to obtain task result information corresponding to the task information generated by the third large language model.

[0123] The dataset characteristic information is used to represent the characteristics of the dataset to be analyzed targeted by the data analysis target information. In practical applications, the dataset characteristic information can be used to reflect the field information contained in the dataset to be analyzed, the association information between fields, etc.

[0124] The second largest language model may be a model for step planning. In practical applications, it may optionally be a model obtained by fine-tuning the base large model through supervised training. The samples used for supervised fine-tuning may include step planning examples. A step planning example may include a task example and a step sequence example for implementing the task example. Alternatively, through a prompt word project, the prompt words input to the second largest language model may include thought chain information for describing the preset step planning logic or step sequence examples generated according to the preset step planning logic, so as to instruct the second largest language model to generate step sequence information according to the preset step planning logic. In this way, the step planning accuracy of the second largest model can be improved, which is conducive to obtaining higher quality task results.

[0125] The third language module can be a model for generating task result information, that is, a model for generating a task report for task information. In practical applications, it can optionally be a model obtained by fine-tuning the base model through supervised training, and the samples used for supervised fine-tuning can include report examples. Alternatively, through prompt word engineering, the prompt words input to the third language model can include thought chain information for describing the preset report generation logic or report examples generated according to the preset report generation logic, so as to instruct the third language model to generate a task report for the task information according to the preset report generation logic. In this way, the standardization of the task report generated by the third language model can be improved.

[0126] In one or more embodiments of the present specification, the realization of the task information depends on the analysis results of at least one data indicator; the data analysis method may further include: obtaining indicator knowledge; the indicator knowledge includes at least one of a first association knowledge between data indicators, a second association knowledge between data indicators and data analysis dimensions, and a third association knowledge between data indicators and data filtering conditions. Accordingly, the step of inputting the task information and the dataset feature information corresponding to the task information into a second language model to obtain the step sequence information for realizing the task information determined by the second language model may specifically include: inputting the task information, the dataset feature information corresponding to the task information, and the indicator knowledge into a second language model to obtain the step sequence information for realizing the task information determined by the second language model with reference to the indicator knowledge.

[0127] Indicator knowledge can be constructed around a dataset by professionals with domain knowledge (such as data analysts). Typically, indicator knowledge takes the form of a knowledge graph, providing a structured and semantic description of the analysis object. Specifically, indicator knowledge includes at least one of the following: first-order association knowledge between data indicators, second-order association knowledge between data indicators and data analysis dimensions, and third-order association knowledge between data indicators and data filtering conditions.

[0128] The first association knowledge between data indicators can reflect the association between two or more data indicators. Furthermore, it can be used to reflect the numerical correlation between different data indicators. For example, there is a formula relationship between "GMV" and "order volume" and "average order value", and there may be a positive correlation between "GMV" and "advertising investment".

[0129] The second association knowledge between data indicators and data analysis dimensions can reflect the relationship between the data indicator and one or more data analysis dimensions. Furthermore, it can be used to reflect the commonly used data analysis dimensions of the data indicator. For example, the commonly used analysis dimensions for the data indicator "GMV" include "time", "region", and "product category".

[0130] The third association knowledge between the data indicator and the data filtering condition can reflect the association relationship between the data indicator and one or more data filtering conditions, and further, can be used to reflect the data filtering conditions commonly used when analyzing the data indicator.

[0131] In the embodiments of this specification, indicator knowledge can provide key domain knowledge context for subsequent intelligent planning and analysis.

[0132] In one or more embodiments of the present specification, a preset analysis step template may be used during the planning phase of step sequence information.

[0133] Specifically, the data analysis method may further include: determining task analysis type information corresponding to the task information; obtaining an analysis step template corresponding to the task analysis type information; wherein the analysis step template includes a template step sequence and tool call instruction information corresponding to each template step in the template step sequence. Accordingly, inputting the task information and the dataset feature information corresponding to the task information into a second language model to obtain step sequence information determined by the second language model for implementing the task information may specifically include: inputting the task information, the dataset feature information corresponding to the task information, and the analysis step template into the second language model to obtain step sequence information generated by the second language model with reference to the analysis step template.

[0134] In actual application, prompt word information can be generated based on the task information, the dataset feature information corresponding to the task information, the analysis step template, and the indicator information. This prompt word can be used to indicate a data processing step that conforms to the analysis step template and is determined by referring to the indicator knowledge to analyze the dataset feature information in response to the task information. Then, based on this prompt word, a data processing step that conforms to the analysis step template and is determined by referring to the indicator knowledge to analyze the dataset feature information in response to the task information is called.

[0135] The task analysis type information may include data trend prediction analysis, indicator anomaly analysis, indicator anomaly attribution analysis, and other types. In practical applications, task analysis type options may be pre-set, and when determining the task analysis type based on the task information, the task analysis type may be selected from the task analysis type options. Alternatively, task analysis type options may not be set, and the large language model may automatically determine the task analysis type based on the task information.

[0136] In the embodiments of this specification, a corresponding analysis step template can be pre-set for each task analysis type option. Therefore, in actual applications, if the analysis task to be performed involves a certain task analysis type option, the pre-set analysis step template can be used to plan the step sequence information for implementing the analysis task. In this way, the accuracy and professionalism of the planned step sequence information can be improved by leveraging domain template knowledge.

[0137] As an example, for the task analysis type option of "gross profit margin anomaly", an analysis step template of "profit anomaly diagnosis" can be preset. This analysis step template contains a fixed analysis framework: anomaly confirmation and measurement; core indicator decomposition and attribution; key dimension drill-down attribution and related factor exploration.

[0138] Specifically, first, anomaly confirmation and measurement, that is, use the anomaly detection operator to confirm whether the indicator really constitutes an anomaly in a statistical sense, and quantify its degree of deviation; then, core indicator decomposition and attribution, that is, call the indicator decomposition contribution attribution operator, decompose "gross profit margin" into its constituent elements (such as "sales revenue" and "sales cost"), and determine which core element's change caused the gross profit margin anomaly; then, key dimension drill-down attribution, that is, based on the common dimensions recommended by the knowledge center (such as "product category" and "sales channel"), call the dimension decomposition contribution attribution operator to find out which specific category or channel contributed to the main anomaly; finally, related factors exploration, that is, based on other indicators related to "gross profit margin" in the knowledge center (such as "discount rate" and "return rate"), call the correlation analysis operator to explore whether there are other potential influencing factors.

[0139] Based on the embodiment of the use of the analysis step template in this specification, after identifying whether the task information meets the task analysis type, the corresponding analysis step template can be retrieved and loaded first. The second largest language model does not need to plan the step sequence information from scratch, but can obtain instantiated step sequence information by intelligently filling in the specific information in the current analysis task (such as time, specific indicator name, etc.) into the analysis step template. Since the analysis step template can be formulated based on expert experience, it usually has a relatively more complete and rigorous logic. Therefore, step sequence information with expert thinking logic can be obtained, so that the quality of the generated step sequence information is high, which is conducive to improving the quality of the final data analysis results and can meet the user's professional analysis needs.

[0140] In one or more embodiments of the present specification, during the planning stage of the step sequence information, a comparative analysis step can be proactively introduced, thereby improving the professionalism of the generated task results and further improving the professionalism of the ultimately generated data analysis results.

[0141] Specifically, the realization of the task information depends on the analysis results of the target data indicator; accordingly, the task information and the data set feature information corresponding to the task information are input into the second largest language model to obtain the step sequence information for realizing the task information determined by the second largest language model, which may specifically include: generating a first prompt word based on the task information, the data set feature information corresponding to the task information and the target data indicator, the first prompt word being used to indicate the execution of a comparative analysis step for the target data indicator, the comparative analysis step being used to compare and analyze a first indicator value of the target data indicator at a first time with a second indicator value of the target data indicator at a second time; calling the second largest language model based on the first prompt word to obtain the step sequence information determined by the second largest language model and including the comparative analysis step.

[0142] In practice, you can define default baselines for comparison in the Knowledge Center. For example, you can configure default comparison baselines for key metrics (such as GMV and active users) in the Knowledge Center. For example, the default comparison baselines for the GMV metric are "Daily Year-over-Year" and "Weekly Year-over-Year."

[0143] Based on the solution of the embodiments of this specification, when planning step sequence information, if the task information involves querying or analyzing a core indicator, the second language model can automatically add one or more steps to compare the core indicator with the baseline. In view of this, through comparative analysis, it can proactively provide users with further in-depth analysis results.

[0144] In one or more embodiments of the present specification, an exploratory step may be introduced during the planning phase of step sequence information.

[0145] Specifically, the data analysis method further includes: determining whether the task information and the characteristic information of the dataset corresponding to the task information meet a preset verification trigger condition; if so, obtaining verification task information corresponding to the verification trigger condition. Accordingly, inputting the task information and the characteristic information of the dataset corresponding to the task information into a second language model to obtain step sequence information for implementing the task information, as determined by the second language model, may specifically include: inputting the task information, the characteristic information of the dataset corresponding to the task information, and the verification task information into the second language model to obtain step sequence information including verification steps for implementing the verification task, as determined by the second language model.

[0146] In practical applications, the knowledge graph can store common business hypotheses, such as the positive correlation between "big sales promotions" and "GMV," and the positive correlation between "holidays" and "new user registrations." When task information or the characteristics of the data targeted by the task information touch upon a pre-defined business hypothesis, the second language model can generate analytical steps to verify or falsify the business hypothesis. These analytical steps can then be executed to verify or falsify the business hypothesis, and the verification or falsification results can be fed back to the user.

[0147] Based on the embodiment of introducing the exploratory verification step in this specification, the intelligent agent has preliminary reasoning ability, can actively explore the potential causal relationship behind the data, and realize enhanced analysis of the data.

[0148] In one or more embodiments of the present specification, the agent's planning of step sequence information may be dynamic.

[0149] Specifically, the calling of multiple data analysis tools to execute the multiple data processing step information according to the step execution logic information to obtain multiple step execution results generated by the multiple data analysis tools may specifically include: calling a first data analysis tool to execute the first step information in the multiple data processing step information to obtain the first step execution result; determining the second step information based on the first step execution result and the step execution logic information; calling a second data analysis tool to execute the second step information to obtain the second step execution result.

[0150] Optionally, determining the second step information based on the execution result of the first step and the step execution logic information may specifically include: determining the predetermined second step information after the first step information based on the step execution logic information; generating a third prompt word based on the execution result of the first tool and the predetermined second step information, the third prompt word being used to indicate that the reference information in the predetermined second step information is replaced with entity information based on the tool execution result; and inputting the third prompt word into the fourth language model to obtain the second step information obtained after the fourth language model replaces the reference information in the predetermined second step information with entity information.

[0151] Optionally, determining the second step information based on the first step execution result and the step execution logic information may specifically include: determining the predetermined second step information after the first step information based on the step execution logic information; generating a step inference prompt word based on the first tool execution result and the predetermined second step information; the step inference prompt word is used to indicate whether the predetermined second step information needs to be executed based on the tool execution result and determine the predetermined third step information to be executed when the predetermined second step information does not need to be executed; inputting the step inference prompt word into the fourth language model to obtain the second step information determined by the fourth language model; the second step information is the predetermined second step information or the predetermined third step information.

[0152] like Figure 3 , which shows a schematic diagram of the plan execution process of an intelligent agent provided in an embodiment of this specification.

[0153] In actual application, after determining the step sequence information, each data processing step can be executed in sequence according to the order of each data processing step information in the step sequence information.

[0154] When executing step N (S302), determine whether step N is the first step (S304). If step N is the first step, call the data analysis tool corresponding to the step (S312) to obtain the step execution result of step N.

[0155] If step N is not the first step, the execution result of the preceding step can be extracted (S306). Specifically, the execution results of the preceding step 1 to step N-1 are extracted as context and provided together with the subsequent step N to the fourth language model so that the fourth language model can use the context to perform reference resolution and conditional judgment (S308).

[0156] Coreference resolution can mean replacing the pronouns in step N (e.g., "the channel," "its performance") with the specific entities identified in steps 1 to N-1. For example, explicitly replacing "the channel" in "predict the performance of the channel over the next three days" with "XX Mall" found in the previous step.

[0157] Conditional execution can refer to determining whether the current step N is still necessary based on the results of the previous steps 1 to N-1 (S310). For example, if the previous step finds that "gross profit margin is normal," the subsequent step of "analyzing the cause of the abnormality" can be automatically skipped.

[0158] If the conditional judgment result indicates that step N does not need to be skipped, the updated step N after reference resolution is used to call the data analysis tool (S312) to obtain the step execution result of step N.

[0159] If the conditional judgment result indicates that step N needs to be skipped, the subsequent steps to be executed after skipping step N are determined according to the step execution logic information, for example, step N+1, and then the reference resolution and conditional judgment steps for step N+1 are continued as for step N.

[0160] After calling the data analysis tool in step N to obtain the step execution result, it can be determined whether the steps in the current step sequence information have been executed (S314). If so, the execution process of the step sequence information ends; if not, the next step is executed until the steps in the step sequence information are executed.

[0161] In one or more embodiments of this specification, in accordance with Figure 3 After the step sequence information is executed in the process shown, the obtained multiple step execution results can be further analyzed and improved to improve the quality of the task result information finally generated.

[0162] Specifically, after calling multiple data analysis tools to execute the multiple data processing step information according to the step execution logic information and obtaining the multiple step execution results generated by the multiple data analysis tools, the method may further include: generating a fourth prompt word based on the task information and the multiple step execution results; the fourth prompt word is used to indicate the generation of modification suggestion information for the multiple step execution results; inputting the fourth prompt word into a fifth language model to obtain newly added step information output by the fifth language model; and calling a data analysis tool to execute the newly added step information to obtain newly added execution result information. Correspondingly, inputting the multiple step execution results into a third language model to obtain task result information generated by the third language model and corresponding to the task information may specifically include: inputting the multiple step execution results and the newly added execution result information into the third language model to obtain task result information generated by the third language model and corresponding to the task information.

[0163] The third language model can be used to expand the steps. In practical applications, prompt word engineering can be used to include indicative information for inferring additional steps in the prompt words input to the fifth language model. For example, such instructions might include, "Based on the above findings, are there any missing analysis angles?" or "Is there any related information worth further exploration?" The fifth language model can then output additional step information to provide further, more in-depth analysis suggestions.

[0164] Based on the solution of the embodiment of this specification, by allowing the fifth language model to infer the newly added step information, the initial step sequence information can be checked for omissions and supplemented, and a more complete step sequence can be obtained. Based on the newly added steps, a more in-depth analysis can be carried out, and more comprehensive and professional analysis results can be obtained, ultimately improving the data analysis depth and professionalism of the data analysis results for the user's data analysis target information feedback.

[0165] In one or more embodiments of the present specification, the data analysis tools called by the intelligent agent may include external tools and internal tools.

[0166] Among them, calling external tools can be through the MCP protocol. Generally, external tools can include data analysis field tools, general tools, business intelligence field tools, etc., but are not limited to these. Among them, data analysis field tools can include data set connection tools, data set creation tools, report creation tools, data analysis tools, report interpretation tools, chart making tools, etc., but are not limited to these. General tools can include browser use tools, Python execution tools, file reading tools, web page search tools, etc., but are not limited to these. Business intelligence field tools can include search industry consulting tools, research industry analysis tools, custom data processing tools, etc., but are not limited to these.

[0167] Internal tools can be pre-built, standardized, independently executable computing units that receive clear input information, execute fixed analysis algorithms within a limited data range, and produce structured results. Operators in the operator library can be abstracted from common analysis scenarios. One operator corresponds to one analysis scenario. As examples, analysis scenarios may include statistical description, time series forecasting, trend analysis, correlation analysis, proportion analysis, distribution analysis, anomaly detection, dimensional attribution, indicator attribution, etc. In actual applications, the input information provided to the operator can be dataset field information, such as indicators, dimensions, and filtering conditions.

[0168] Based on the embodiments of this specification, for the parts that require internal calculations of the intelligent agent, they are implemented by calling internal operators. Since these operators are deterministic computing units rather than relying on AI-generated code, the execution results are highly accurate and can minimize the impact of model hallucinations on the data processing results of the intelligent agent.

[0169] Furthermore, in the embodiments of this specification, to ensure the richness and readability of the analysis conclusions, a unique rich text display syntax can be preset for each operator. In actual applications, after the operator is executed, the operator's execution results can be rendered on the user terminal according to the preset rich text display syntax, including key conclusions, data tables, and visual charts (such as line charts and bar charts). The illustrated data card with both text and images ensures that the process results are clear and easy to understand.

[0170] Figure 4 A schematic diagram of the data processing flow of an intelligent agent in an actual application scenario provided in an embodiment of this specification.

[0171] like Figure 4 As shown in , the agent includes a first step planner, which can be constructed based on the second language model. The first step planner can disassemble the task information provided by the agent caller (for example, the user terminal or the planning engine server) to obtain step sequence information. Figure 4 As shown in , step sequence information including Step 1, Step 2, Step 3, etc. can be obtained.

[0172] In actual application, when the first step planning module determines the step sequence information, it can refer to the knowledge in the knowledge center. In the embodiment of this specification, the knowledge in the knowledge center may include but is not limited to indicator knowledge, analysis step templates, verification task information, etc.

[0173] Then, the step execution logic in the step sequence information can be used to determine the steps to be executed. The step execution module can then call data analysis tools to obtain the step execution results. The obtained step execution results can be fed back to the agent caller and ultimately displayed to the user on a page on the user terminal. In practical applications, the continuously updated step execution results can be streamed to the user terminal page.

[0174] Then, the step sequence verification module can use the step execution result of the previous step as context to verify whether the next step should be executed in the order of steps determined in the step sequence information. If so, the step execution module continues to execute the next step by calling the data analysis tool; if not, a new next step is determined and executed.

[0175] When the entire sequence of steps used to implement the task information is completed, the execution results of all the steps can be reflected and analyzed to determine whether further new steps need to be executed. Specifically, the second step planning module based on the fifth language model in the intelligent agent can be used to generate new steps; then, the data analysis tool can be called through the step execution module to obtain the new execution result information and append it to the step execution results.

[0176] Afterward, all previously obtained step execution result information (including newly added execution result information) can be input into a data integration module based on the third language model for data integration. The data integration module then outputs task result information corresponding to the task information provided by the agent caller. Furthermore, the task result information can be fed back to the agent caller and ultimately displayed to the user on a page on the user terminal.

[0177] In the embodiments of this specification, a first language model, a second language model, a third language model, a fourth language model, and a fifth language model are proposed, which can be used to represent different models. However, optionally, some of these models can use the same model in actual applications, by providing different prompt word information to achieve different functions in different application scenarios.

[0178] Figure 5A flowchart of a data analysis method provided in an embodiment of this specification.

[0179] From a program perspective, the execution body of the process can be a program installed on the user terminal. It is understood that the method can be executed by any device or equipment with computing and processing capabilities. In actual application, the user terminal can have a user interface.

[0180] like Figure 5 As shown, the process may include the following steps:

[0181] Step 502: Receive data analysis target information in natural language form input by the user.

[0182] Step 504: Send the data analysis target information to a server deployed with a first large language model; the first large language model is used to decompose the data analysis target information to obtain first task sequence information; the first task sequence information includes multiple task information and task execution logic information of the multiple task information; the multiple task information is used to be executed by multiple intelligent agents according to the task execution logic information to generate data analysis results corresponding to the data analysis target information.

[0183] Step 506: Obtain the data analysis result fed back by the server.

[0184] Step 508: Display the data analysis result on the page of the user terminal.

[0185] In actual applications, users can enter data analysis target information in natural language through a dialogue in the user interface of the user terminal. The data analysis target information can be information used to instruct data analysis. In actual applications, users can also enter or specify the data set to be analyzed for the data analysis target information; when specifying the data set to be analyzed, users can select or enter data set feature information.

[0186] The user terminal can send the data analysis target information to the Agentic AI system for data analysis. Specifically, it can be sent to the Agentic AI system such as Figure 1The planning engine server 10 shown in the figure triggers the entire Agentic AI system to perform the data analysis task in response to the data analysis target information. In actual application, the first large language model is deployed on the planning engine server 10 to implement the task planning function of the planning engine server 10. Thus, the data analysis target information is input into the first large language model to obtain the first large language model to decompose the data analysis target information to obtain the first task sequence information, and then obtain the data analysis result obtained by calling multiple agents on the data analysis agent server 20 in the Agentic AI system to execute the first task sequence information based on the first task sequence information.

[0187] In an embodiment of the present specification, after obtaining the data analysis result fed back by the server, the user terminal may display the terminal analysis result on a page of the user terminal.

[0188] In at least some embodiments of the present specification, the data analysis results can be obtained based on multiple task result information generated by the multiple intelligent agents; thus, the data analysis method may also include: the user terminal obtains the multiple task result information fed back by the server; the first progress of the user terminal obtaining the multiple task result information is consistent with the second progress of the server generating the multiple task result information; then, the multiple task result information is displayed on the page of the user terminal according to the first progress.

[0189] In at least some embodiments of the present specification, the task result information can be obtained based on multiple step execution results generated by multiple data analysis tools; thus, the data analysis method may further include: obtaining the multiple step execution results fed back by the server; the third progress of the user terminal obtaining the multiple step execution results is consistent with the fourth progress of the server generating the multiple step execution results; and then, displaying the multiple step execution results on the page of the user terminal according to the third progress.

[0190] Based on the solution of the embodiments of this specification, in the process of the Agentic AI system decomposing the user's data analysis target information into multiple tasks and calling multiple intelligent agents to obtain task result information fed back by each intelligent agent, the task result information can be sent to the user terminal in a timely manner after each task result information fed back by the intelligent agent is obtained. Therefore, on the one hand, the user can understand the progress of data analysis through the task result information fed back by each intelligent agent displayed in a streaming manner on the user terminal. On the other hand, the user can trace the intermediate process information of the final generated data analysis results (for example, data analysis reports, statements, etc.) by viewing the task results generated by a single intelligent agent for a single decomposed task, so as to facilitate understanding of the final data analysis results.

[0191] Furthermore, when each agent executes task information, it decomposes the task information into multiple data processing steps and calls multiple data analysis tools to obtain the step execution results fed back by each data analysis tool. Each time a step execution result fed back by a data analysis tool is obtained, the step execution result can be sent to the user terminal. Thus, on the one hand, the user can understand the data analysis progress through the step execution results fed back by each processing tool displayed in a streaming manner on the user terminal. On the other hand, the user can trace the intermediate process information of the agent generating the task result by viewing the step execution results generated by a single data analysis tool for a single decomposed data processing step, thereby facilitating understanding of the task result information output by the agent and the final data analysis results. Thus, the user can trace the data analysis results.

[0192] In at least some embodiments of this specification, after the data analysis results are displayed on the page of the user terminal, the following may also be included: obtaining the user's editing operation on the data analysis results; and updating the data analysis results in response to the editing operation. In actual application, some editing operations, such as operations to modify the data analysis framework or outline, can also be fed back to the Agentic AI system so that the Agentic AI system can perform additional analysis based on the modified data analysis framework or outline. In this way, the user can edit the data analysis results.

[0193] In at least some embodiments of this specification, after obtaining the data analysis results, the data analysis target information can be associated and stored with the data analysis result information. In actual application, a trigger control for the data analysis target information can be displayed on a page of a user terminal. This trigger control can be used to display the data analysis result information corresponding to the data analysis target information on the page of the user terminal in response to a user trigger operation.

[0194] Furthermore, the streaming display of the multiple step execution results, the streaming display of the multiple task result information, and the display of the data analysis results can be recorded to generate sequence display information. The sequence display information can be stored in association with the data analysis target information. In actual application, a trigger control for the data analysis target information can be displayed on the page of the user terminal. The trigger control can be used to display the sequence display information on the page of the user terminal in response to the user's trigger operation. Specifically, the multiple step execution results, the multiple task result information, and the data analysis results are displayed again in the order and form in which they were previously displayed. In this way, the user can review the data analysis process.

[0195] In order to facilitate understanding of the embodiments of this specification, the display process of each information on the user terminal is as follows: Figure 6 This is a schematic diagram of a user terminal page in a data analysis process provided by an embodiment of this specification.

[0196] like Figure 6 As shown, when the Agentic AI system of the embodiment of this specification analyzes the data analysis target information of "help me deeply analyze whether the sum is abnormal and the potential causes of the abnormality", the intermediate analysis results can be streamed to the user terminal and displayed on the page of the user terminal.

[0197] like Figure 6 As shown, the user terminal page can include a dialogue area and an analysis result display area, wherein the dialogue area further includes an analysis progress display area and a user question input area. In actual application, after the user enters data analysis target information, the analysis progress of the data analysis target information can be displayed in a streaming manner in the analysis progress display area, and the data analysis results or intermediate data analysis results consistent with the analysis progress can be displayed in the analysis result display area.

[0198] Specifically, if the process of analyzing the target data is divided into a first task and a second task, the first agent is called to perform the first task; and the first agent further divides the first task into multiple steps, such as the first step, the second step, and the third step, and then calls the data analysis tool to perform the first step. Figure 6 1.1 corresponds to the first step executed by the first agent calling the data analysis tool, and 1.2 corresponds to the second step executed by the second agent calling the data analysis tool, such as Figure 6 As shown in , the analysis progress shows that the first step 1.1 has been completed and the second step 1.2 is being executed.

[0199] Furthermore, in an embodiment of the present specification, an analysis result backtracking control may be included in the user terminal page. Specifically, the analysis result backtracking control may be displayed corresponding to each step execution result generated by the data analysis tool and each task result information generated by the agent, so that by triggering the control, the result corresponding to the control is displayed in the analysis result display area.

[0200] The above are only examples. In actual application, the execution results of some or all data analysis tools can be displayed in the user terminal, and the execution results of some or all intelligent agents can be displayed. In practice, they can be adjusted according to business conditions, user needs, etc.

[0201] The various technical features in the above embodiments can be arbitrarily combined as long as there is no conflict or contradiction between the combinations of features. However, due to space limitations, they are not described one by one. Therefore, the arbitrary combination of the various technical features in the above embodiments also falls within the scope of disclosure of this specification.

[0202] Based on the same idea, the embodiments of this specification also provide a device corresponding to the above method.

[0203] Figure 7 The embodiments of this specification provide corresponding Figure 2 A structural diagram of a data analysis device.

[0204] like Figure 7 As shown, the device is applied to a server, and the device may include:

[0205] The data analysis target information acquisition module 702 is used to acquire the user's data analysis target information in the form of natural language;

[0206] The target decomposition module 704 is configured to input the data analysis target information into a first large language model to obtain first task sequence information obtained by decomposing the data analysis target information by the first large language model; the first task sequence information includes a plurality of task information and task execution logic information of the plurality of task information;

[0207] The task result information determination module 706 is used to call multiple agents to execute the multiple task information according to the task execution logic information, and obtain multiple task result information generated by the multiple agents;

[0208] The data analysis result generating module 708 is configured to determine a data analysis result corresponding to the data analysis target information based on the plurality of task result information.

[0209] based on Figure 7 The present specification also provides some specific implementation plans of the method, which are described below.

[0210] Optionally, the task result information determination module 706 is specifically used to: call the first intelligent agent to execute the first task information among the multiple task information to obtain the first task result information; determine the second task information based on the first task result information and the task execution logic information; call the second intelligent agent to execute the second task information to obtain the second task result information.

[0211] Optionally, the first task result information is first type result information; the first type result information is used to represent the task execution result information obtained after the first task information is successfully executed; determining the second task information based on the first task result information and the task execution logic information specifically includes: if the task execution result information meets the target execution condition corresponding to the target execution logic in the task execution logic information, then determining the first target task information corresponding to the target execution logic in the first task sequence information as the second task information.

[0212] Optionally, the first task result information is second type result information; the second type result information is used to indicate that the task execution failed; determining the second task information based on the first task result information and the task execution logic information specifically includes: in response to the second type result information, updating the first task sequence information based on the data analysis target information to obtain second task sequence information; determining the second target task information in the second task sequence information as the second task information.

[0213] Optionally, the data analysis device method is also used to: select a data analysis template that matches the data analysis target information from a preset template set; the data analysis template includes a template task sequence and intelligent agent call instruction information corresponding to each template task in the template task sequence; the target disassembly module 704 is specifically used to: input the data analysis target information and the data analysis template into the first large language model, and obtain the first task sequence information obtained by the first large language model by disassembling the data analysis target information according to the data analysis template.

[0214] Optionally, the selecting a data analysis template that matches the data analysis target information from the preset template set specifically includes: selecting a data analysis template that matches the data analysis target information from the preset template set according to the preset routing rules of each preset template in the preset template set; the preset routing rules are used to indicate the conditions that should be met for matching the corresponding preset template.

[0215] Optionally, the data analysis template includes a preset filling position for filling in task feature information; the first large language model decomposes the data analysis target information according to the data analysis template to obtain first task sequence information, specifically including: extracting task feature information from the data analysis target information; filling the task feature information into the preset filling position to obtain the first task sequence information for realizing the data analysis target information.

[0216] Optionally, the task result information determination module 706 is specifically used to: input the task information and the data set feature information corresponding to the task information into the second largest language model to obtain step sequence information for implementing the task information determined by the second largest language model; the step sequence information includes multiple data processing step information and step execution logic information of the multiple data processing step information; call multiple data analysis tools to execute the multiple data processing step information according to the step execution logic information to obtain multiple step execution results generated by the multiple data analysis tools; input the multiple step execution results into the third largest language model to obtain task result information corresponding to the task information generated by the third largest language model.

[0217] Optionally, the realization of the task information depends on the analysis results of at least one data indicator; the data analysis device is also used to: obtain indicator knowledge; the indicator knowledge includes at least one of the first association knowledge between data indicators, the second association knowledge between data indicators and data analysis dimensions, and the third association knowledge between data indicators and data filtering conditions; the inputting of the task information and the data set feature information corresponding to the task information into the second largest language model to obtain the step sequence information for realizing the task information determined by the second largest language model, specifically includes: inputting the task information, the data set feature information corresponding to the task information and the indicator knowledge into the second largest language model to obtain the step sequence information for realizing the task information determined by the second largest language model with reference to the indicator knowledge.

[0218] Optionally, the data analysis device is also used to: determine the task analysis type information corresponding to the task information; obtain an analysis step template corresponding to the task analysis type information; the analysis step template includes a template step sequence and tool call instruction information corresponding to each template step in the template step sequence; the task information and the data set feature information corresponding to the task information are input into the second largest language model to obtain the step sequence information determined by the second largest language model for implementing the task information, specifically including: inputting the task information, the data set feature information corresponding to the task information and the analysis step template into the second largest language model to obtain the step sequence information generated by the second largest language model with reference to the analysis step template.

[0219] Optionally, the implementation of the task information depends on the analysis results of the target data indicator; the inputting of the task information and the data set feature information corresponding to the task information into the second largest language model to obtain the step sequence information for implementing the task information determined by the second largest language model specifically includes: generating a first prompt word based on the task information, the data set feature information corresponding to the task information and the target data indicator; the first prompt word is used to indicate the execution of a comparative analysis step for the target data indicator; the comparative analysis step is used to compare and analyze a first indicator value of the target data indicator at a first time with a second indicator value of the target data indicator at a second time; and calling the second largest language model based on the first prompt word to obtain the step sequence information including the comparative analysis step determined by the second largest language model.

[0220] Optionally, the data analysis device is also used to: determine whether the task information and the data set characteristic information corresponding to the task information meet a preset verification trigger condition; if so, obtain the verification task information corresponding to the verification trigger condition; the task information and the data set characteristic information corresponding to the task information are input into a second large language model to obtain step sequence information for implementing the task information determined by the second large language model, specifically including: inputting the task information, the data set characteristic information corresponding to the task information and the verification task information into a second large language model to obtain step sequence information including verification steps for implementing the verification task determined by the second large language model.

[0221] Optionally, calling multiple data analysis tools to execute the multiple data processing step information according to the step execution logic information to obtain multiple step execution results generated by the multiple data analysis tools specifically includes: calling a first data analysis tool to execute the first step information in the multiple data processing step information to obtain the first step execution result; determining the second step information based on the first step execution result and the step execution logic information; calling a second data analysis tool to execute the second step information to obtain the second step execution result.

[0222] Optionally, determining the second step information based on the execution result of the first step and the step execution logic information specifically includes: determining the predetermined second step information after the first step information based on the step execution logic information; generating a third prompt word based on the execution result of the first tool and the predetermined second step information; the third prompt word is used to indicate that the reference information in the predetermined second step information is replaced with entity information based on the tool execution result; and inputting the third prompt word into the fourth language model to obtain the second step information obtained after the fourth language model replaces the reference information in the predetermined second step information with entity information.

[0223] Optionally, determining the second step information based on the first step execution result and the step execution logic information specifically includes: determining the predetermined second step information after the first step information based on the step execution logic information; generating a step inference prompt word based on the first tool execution result and the predetermined second step information; the step inference prompt word is used to indicate whether the predetermined second step information needs to be executed based on the tool execution result and determine the predetermined third step information to be executed when the predetermined second step information does not need to be executed; inputting the step inference prompt word into the fourth language model to obtain the second step information determined by the fourth language model; the second step information is the predetermined second step information or the predetermined third step information.

[0224] Optionally, the data analysis device is also used to: generate a fourth prompt word based on the task information and the execution results of the multiple steps; the fourth prompt word is used to indicate the generation of modification suggestion information for the execution results of the multiple steps; input the fourth prompt word into the fifth language model to obtain the newly added step information output by the fifth language model; call the data analysis tool to execute the newly added step information to obtain the newly added execution result information; the inputting the multiple step execution results into the third language model to obtain the task result information generated by the third language model corresponding to the task information specifically includes: inputting the multiple step execution results and the newly added execution result information into the third language model to obtain the task result information generated by the third language model corresponding to the task information.

[0225] It is understood that the above modules refer to computer programs or program segments for performing one or more specific functions. In addition, the distinction between the above modules does not mean that the actual program codes must also be separated.

[0226] The above is a schematic diagram of a data analysis device according to this embodiment. It should be noted that the technical solution of the data analysis device and the technical solution of the above-mentioned data analysis method are based on the same concept. For details not described in detail in the technical solution of the data analysis device, please refer to the description of the technical solution of the above-mentioned data analysis method.

[0227] Based on the same idea, the embodiments of this specification also provide a device corresponding to the above method.

[0228] Figure 8 The embodiments of this specification provide corresponding Figure 5 A structural diagram of a data analysis device.

[0229] like Figure 8 As shown, the device is applied to a user terminal, and the device may include:

[0230] The data analysis target information receiving module 802 is used to receive data analysis target information in natural language input by a user;

[0231] The data analysis target information sending module 804 is configured to send the data analysis target information to a server deployed with a first large language model; the first large language model is configured to decompose the data analysis target information to obtain first task sequence information; the first task sequence information includes a plurality of task information and task execution logic information of the plurality of task information; the plurality of task information is configured to be executed by a plurality of intelligent agents according to the task execution logic information to generate data analysis results corresponding to the data analysis target information;

[0232] A data analysis result acquisition module 806 is configured to acquire the data analysis result fed back by the server;

[0233] The data analysis result display module 808 is used to display the data analysis result on the page of the user terminal.

[0234] based on Figure 8 The present specification also provides some specific implementation plans of the method, which are described below.

[0235] Optionally, the data analysis results are obtained based on multiple task result information generated by the multiple intelligent agents; the data analysis device is also used to: obtain the multiple task result information fed back by the server; the first progress of the user terminal in obtaining the multiple task result information is consistent with the second progress of the server generating the multiple task result information; and display the multiple task result information on the page of the user terminal according to the first progress.

[0236] Optionally, the task result information is obtained based on multiple step execution results generated by multiple data analysis tools; the data analysis device is also used to: obtain the multiple step execution results fed back by the server; the third progress of the user terminal obtaining the multiple step execution results is consistent with the fourth progress of the server generating the multiple step execution results; and display the multiple step execution results on the page of the user terminal according to the third progress.

[0237] It is understood that the above modules refer to computer programs or program segments for performing one or more specific functions. In addition, the distinction between the above modules does not mean that the actual program codes must also be separated.

[0238] The above is a schematic diagram of a data analysis device according to this embodiment. It should be noted that the technical solution of the data analysis device and the technical solution of the above-mentioned data analysis method are based on the same concept. For details not described in detail in the technical solution of the data analysis device, please refer to the description of the technical solution of the above-mentioned data analysis method.

[0239] Based on the same idea, the embodiments of this specification also provide devices corresponding to the above methods.

[0240] Figure 9 This is a block diagram of a computing device according to an embodiment of the present disclosure. Components of the computing device 900 include, but are not limited to, a memory 910 and a processor 920. The processor 920 is connected to the memory 910 via a bus 930, and a database 950 is used to store data.

[0241] The computing device 900 also includes an access device 940 that enables the computing device 900 to communicate via one or more networks 960. Examples of such networks include a public switched telephone network (PSTN), a local area network (LAN), a wide area network (WAN), a personal area network (PAN), or a combination of communication networks such as the Internet. The access device 940 may include one or more of any type of network interface (e.g., a network interface card (NIC)) whether wired or wireless, such as an IEEE 802.11 wireless local area network (WLAN) wireless interface, a Worldwide Interoperability for Microwave Access (Wi-MAX) interface, an Ethernet interface, a universal serial bus (USB) interface, a cellular network interface, a Bluetooth interface, a near field communication (NFC) interface, and the like.

[0242] In one embodiment of the present application, the above components of the computing device 900 and Figure 9 Other components not shown in the figure may also be connected to each other, for example, via a bus. Figure 9 The computing device structure block diagram shown is for illustrative purposes only and is not intended to limit the scope of the present application. Those skilled in the art may add or replace other components as needed.

[0243] The computing device 900 may be any type of stationary or mobile computing device, including a mobile computer or mobile computing device (e.g., a tablet computer, a personal digital assistant, a laptop computer, a notebook computer, a netbook computer, etc.), a mobile phone (e.g., a smartphone), a wearable computing device (e.g., a smartwatch, smart glasses, etc.), or other types of mobile devices, or a stationary computing device such as a desktop computer or a personal computer (PC). The computing device 900 may also be a mobile or stationary server.

[0244] When the processor 920 executes the computer instructions, the steps of the data analysis method are implemented.

[0245] The above is a schematic diagram of a computing device according to this embodiment. It should be noted that the technical solution of the computing device and the technical solution of the above-mentioned data analysis method are of the same concept. For details not described in detail in the technical solution of the computing device, please refer to the description of the technical solution of the above-mentioned data analysis method.

[0246] An embodiment of the present application further provides a computer-readable storage medium storing computer instructions, which, when executed by a processor, implement the steps of the data analysis method described above.

[0247] The above is a schematic diagram 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 aforementioned data analysis method are based on 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 aforementioned data analysis method.

[0248] An embodiment of the present specification further provides a computer program product, comprising a computer program / instruction, which implements the steps of the above-mentioned data analysis method when executed by a processor.

[0249] The above is a schematic diagram 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 aforementioned data analysis method are based on 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 aforementioned data analysis method.

[0250] Each embodiment in this specification is described in a progressive manner, and the same or similar parts between the embodiments can be referred to each other. Each embodiment focuses on the differences from other embodiments. In particular, for the embodiments of the device, equipment, computer-readable storage medium, and computer program product, since they are basically similar to the method embodiments, the description is relatively simple, and the relevant parts can be referred to the partial description of the method embodiments. The device, device, computer-readable storage medium, and computer program product provided in the embodiments of this specification correspond to the method, so the device, device, computer-readable storage medium, and computer program product also have beneficial technical effects similar to the corresponding method. Since the beneficial technical effects of the method have been described in detail above, the beneficial technical effects of the corresponding device, device, computer-readable storage medium, and computer program product will not be repeated here.

[0251] The foregoing description describes specific embodiments of the present application. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps recited in the claims can be performed in an order different from that described in the embodiments and still achieve the desired results. Furthermore, the processes depicted in the accompanying drawings do not necessarily require the specific order shown or the sequential order to achieve the desired results. In certain embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0252] In the 1990s, technological improvements could be clearly distinguished as either hardware improvements (for example, improvements to circuit structures such as diodes, transistors, and switches) or software improvements (improvements to process flows). However, with the advancement of technology, many process flow improvements today can now be considered direct improvements to hardware circuit structures. Designers almost always create the corresponding hardware circuit structure by programming the improved process flow into the hardware circuit. Therefore, it cannot be said that a process flow improvement cannot be implemented using hardware modules. For example, a programmable logic device (PLD) (such as a field programmable gate array (FPGA)) is an integrated circuit whose logical function is determined by user programming. Designers can "integrate" a digital system onto a PLD through their own programming, without having to hire a chip manufacturer to design and manufacture a dedicated integrated circuit chip. Moreover, nowadays, instead of manually fabricating integrated circuit chips, this programming is mostly done using "logic compiler" software. This is similar to the software compiler used when developing programs. Before compilation, the original code must also be written in a specific programming language, called a hardware description language (HDL). There is not just one HDL, but many, such as ABEL (Advanced Boolean Expression Language), AHDL (Altera Hardware Description Language), Confluence, CUPL (Cornell University Programming Language), HDCal, JHDL (Java Hardware Description Language), Lava, Lola, MyHDL, PALASM, RHDL (Ruby Hardware Description Language), etc. The most commonly used ones are VHDL (Very-High-Speed ​​Integrated Circuit Hardware Description Language) and Verilog. Those skilled in the art will also understand that by simply programming the method flow in one of these hardware description languages ​​and then programming it into an integrated circuit, a hardware circuit that implements the logic method flow can be easily obtained.

[0253] The controller can be implemented in any suitable manner. For example, the controller can take the form of a microprocessor or processor and a computer-readable medium storing computer-readable program code (e.g., software or firmware) executable by the (micro)processor, logic gates, switches, application-specific integrated circuits (ASICs), programmable logic controllers, and embedded microcontrollers. Examples of controllers include, but are not limited to, the following microcontrollers: ARC 625D, Atmel AT91SAM, Microchip PIC18F26K20, and Silicone Labs C8051F320. The memory controller can also be implemented as part of the control logic of the memory. Those skilled in the art will also know that in addition to implementing the controller in a purely computer-readable program code format, the controller can be implemented in the form of logic gates, switches, application-specific integrated circuits, programmable logic controllers, and embedded microcontrollers by logically programming the method steps. Therefore, such a controller can be considered a hardware component, and the devices included therein for implementing various functions can also be considered as structures within the hardware component. Or even, the devices for implementing various functions can be considered as both software modules that implement the method and structures within the hardware component.

[0254] The systems, devices, modules, or units described in the above embodiments may be implemented by computer chips or entities, or by products having certain functions. A typical implementation device is a computer. Specifically, the computer may be, for example, a personal computer, a laptop computer, a cellular phone, a camera phone, a smartphone, a personal digital assistant, a media player, a navigation device, an email device, a game console, a tablet computer, a wearable device, or a combination of any of these devices.

[0255] For the convenience of description, the above devices are described as being divided into various units according to their functions. Of course, when implementing this application, the functions of each unit can be implemented in the same or multiple software and / or hardware.

[0256] It will be understood by those skilled in the art that embodiments of the present invention may be provided as methods, systems, or computer program products. Thus, the present invention may take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware. Furthermore, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to magnetic disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0257] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowcharts and / or block diagrams, as well as combinations of processes and / or blocks in the flowcharts and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowcharts and / or block diagrams. Figure 1 a process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0258] These computer program instructions may also be stored in a computer readable memory that can direct a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 a process or multiple processes and / or boxes Figure 1 The function specified in one or more boxes.

[0259] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operational steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing the instructions executed on the computer or other programmable device for implementing the process. Figure 1 a process or multiple processes and / or boxes Figure 1 A step that specifies a function in one or more boxes.

[0260] In a typical configuration, a computing device includes one or more processors (CPUs), input / output interfaces, network interfaces, and memory.

[0261] Memory may include non-permanent storage in a computer-readable medium, random access memory (RAM) and / or non-volatile memory in the form of read-only memory (ROM) or flash RAM. Memory is an example of a computer-readable medium.

[0262] Computer-readable media include permanent and non-permanent, removable and non-removable media that can be implemented by any method or technology to store information. The information can be computer-readable instructions, data structures, program modules or other data. Examples of computer storage media include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technology, compact disc read-only memory (CD-ROM), digital versatile disc (DVD) or other optical storage, magnetic cassettes, magnetic disk storage or other magnetic storage devices or any other non-transmission media that can be used to store information that can be accessed by a computing device. As defined herein, computer-readable media does not include transitory media such as modulated data signals and carrier waves.

[0263] It should also be noted that the terms "comprises," "includes," or any other variations thereof are intended to encompass non-exclusive inclusion, such that a process, method, commodity, or apparatus that includes a series of elements includes not only those elements but also other elements not explicitly listed, or includes elements inherent to such process, method, commodity, or apparatus. In the absence of further limitations, an element defined by the phrase "comprises a ..." does not exclude the presence of other identical elements in the process, method, commodity, or apparatus that includes the element.

[0264] The present application may be described in the general context of computer-executable instructions executed by a computer, such as program modules. Generally, program modules include routines, programs, objects, components, data structures, etc. that perform specific tasks or implement specific abstract data types. The present application may also be practiced in distributed computing environments where tasks are performed by remote processing devices connected through a communications network. In a distributed computing environment, program modules may be located in local and remote computer storage media, including storage devices.

[0265] The foregoing is merely an embodiment of the present application and is not intended to limit the present application. For those skilled in the art, the present application may have various changes and variations. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principles of the present application should all be included within the scope of the claims of the present application.

Claims

1. A data analysis method, applied to a server, comprising: Obtaining data analysis target information in the form of natural language from the user; Inputting the data analysis target information into a first large language model to obtain first task sequence information obtained by decomposing the data analysis target information by the first large language model; the first task sequence information includes a plurality of task information and task execution logic information of the plurality of task information; Calling multiple agents to execute the multiple task information according to the task execution logic information, and obtaining multiple task result information generated by the multiple agents; Based on the plurality of task result information, a data analysis result corresponding to the data analysis target information is determined.

2. The method according to claim 1, wherein the calling of multiple agents to execute the multiple task information according to the task execution logic information to obtain multiple task result information generated by the multiple agents specifically comprises: Invoking a first agent to execute first task information among the plurality of task information to obtain first task result information; Determining second task information according to the first task result information and the task execution logic information; Call the second agent to execute the second task information and obtain the second task result information.

3. The method of claim 2, wherein the first task result information is first-type result information; the first-type result information is used to represent task execution result information obtained after the first task information is successfully executed; and determining the second task information based on the first task result information and the task execution logic information specifically includes: If the task execution result information satisfies the target execution condition corresponding to the target execution logic in the task execution logic information, the first target task information corresponding to the target execution logic in the first task sequence information is determined as the second task information.

4. The method according to claim 2, wherein the first task result information is second-type result information; the second-type result information is used to indicate that the task execution failed; The determining of the second task information according to the first task result information and the task execution logic information specifically includes: In response to the second type of result information, updating the first task sequence information based on the data analysis target information to obtain second task sequence information; The second target task information in the second task sequence information is determined as the second task information.

5. The method of claim 1 , further comprising: Selecting a data analysis template that matches the data analysis target information from a preset template set; The data analysis template includes a template task sequence and agent calling instruction information corresponding to each template task in the template task sequence; Inputting the data analysis target information into the first large language model to obtain first task sequence information obtained by decomposing the data analysis target information by the first large language model specifically includes: The data analysis target information and the data analysis template are input into the first large language model to obtain first task sequence information obtained by the first large language model decomposing the data analysis target information according to the data analysis template.

6. The method according to claim 5, wherein selecting a data analysis template that matches the data analysis target information from a preset template set specifically comprises: selecting, from the preset template set, a data analysis template that matches the data analysis target information according to a preset routing rule of each preset template in the preset template set; The preset routing rules are used to indicate conditions that should be met for matching the corresponding preset template.

7. The method according to claim 5, wherein the data analysis template includes a preset filling position for filling in task feature information; The first language model decomposes the data analysis target information according to the data analysis template to obtain first task sequence information, specifically including: extracting task feature information from the data analysis target information; The task characteristic information is filled into the preset filling position to obtain first task sequence information for realizing the data analysis target information.

8. The method according to claim 1, wherein the calling of multiple agents to execute the multiple task information according to the task execution logic information to obtain multiple task result information generated by the multiple agents specifically comprises: Inputting the task information and feature information of a data set corresponding to the task information into a second language model to obtain step sequence information for implementing the task information determined by the second language model; the step sequence information includes information on a plurality of data processing steps and step execution logic information for the plurality of data processing step information; Calling multiple data analysis tools to execute the multiple data processing step information according to the step execution logic information, and obtaining multiple step execution results generated by the multiple data analysis tools; The execution results of the multiple steps are input into a third language model to obtain task result information corresponding to the task information generated by the third language model.

9. The method of claim 8, wherein the realization of the task information depends on the analysis result of at least one data indicator; the method further comprising: Acquire indicator knowledge; the indicator knowledge includes at least one of first association knowledge between data indicators, second association knowledge between data indicators and data analysis dimensions, and third association knowledge between data indicators and data filtering conditions; Inputting the task information and the feature information of the data set corresponding to the task information into a second language model to obtain the step sequence information for implementing the task information determined by the second language model specifically includes: The task information, the dataset feature information corresponding to the task information, and the indicator knowledge are input into a second language model to obtain step sequence information for implementing the task information determined by the second language model with reference to the indicator knowledge.

10. The method of claim 8, further comprising: Determining task analysis type information corresponding to the task information; Obtaining an analysis step template corresponding to the task analysis type information; The analysis step template includes a template step sequence and tool call instruction information corresponding to each template step in the template step sequence; Inputting the task information and the feature information of the data set corresponding to the task information into a second language model to obtain the step sequence information for implementing the task information determined by the second language model specifically includes: The task information, the dataset feature information corresponding to the task information, and the analysis step template are input into a second language model to obtain step sequence information generated by the second language model with reference to the analysis step template.

11. The method of claim 8, wherein implementation of the task information depends on analysis results of target data indicators; wherein inputting the task information and feature information of a dataset corresponding to the task information into a second language model to obtain step sequence information for implementing the task information determined by the second language model specifically comprises: generating a first prompt word based on the task information, feature information of a data set corresponding to the task information, and the target data indicator; The first prompt word is used to instruct to perform a comparative analysis step on the target data indicator; the comparative analysis step is used to compare and analyze a first indicator value of the target data indicator at a first time with a second indicator value of the target data indicator at a second time; The second language model is called based on the first prompt word to obtain step sequence information including the comparative analysis step determined by the second language model.

12. The method of claim 8, further comprising: Determining whether the task information and the characteristic information of the data set corresponding to the task information meet a preset verification trigger condition; If the conditions are met, the verification task information corresponding to the verification trigger condition is obtained; Inputting the task information and the feature information of the data set corresponding to the task information into a second language model to obtain the step sequence information for implementing the task information determined by the second language model specifically includes: The task information, the dataset feature information corresponding to the task information, and the verification task information are input into a second language model to obtain step sequence information including verification steps for implementing the verification task determined by the second language model.

13. The method of claim 8, wherein the calling of multiple data analysis tools to execute the multiple data processing step information according to the step execution logic information to obtain multiple step execution results generated by the multiple data analysis tools specifically comprises: Invoking a first data analysis tool to execute first step information among the plurality of data processing step information to obtain a first step execution result; Determining second step information according to the execution result of the first step and the step execution logic information; Call a second data analysis tool to execute the second step information and obtain a second step execution result.

14. The method according to claim 13, wherein determining the second step information according to the first step execution result and the step execution logic information specifically comprises: determining predetermined second step information following the first step information according to the step execution logic information; generating a third prompt word based on the execution result of the first tool and the predetermined second step information; The third prompt word is used to instruct to replace the reference information in the predetermined second step information with entity information based on the execution result of the tool; The third prompt word is input into the fourth language model to obtain second step information obtained after the fourth language model replaces the reference information in the predetermined second step information with entity information.

15. The method of claim 8, wherein after calling multiple data analysis tools to execute the multiple data processing step information according to the step execution logic information and obtaining multiple step execution results generated by the multiple data analysis tools, the method further comprises: generating a fourth prompt word based on the task information and the execution results of the multiple steps; The fourth prompt word is used to instruct to generate modification suggestion information for the execution results of the multiple steps; Inputting the fourth prompt word into the fifth language model to obtain the newly added step information output by the fifth language model; Calling a data analysis tool to execute the newly added step information and obtain newly added execution result information; Inputting the execution results of the multiple steps into a third language model to obtain task result information corresponding to the task information generated by the third language model specifically includes: The execution results of the multiple steps and the newly added execution result information are input into a third language model to obtain task result information corresponding to the task information generated by the third language model.

16. A data analysis method, applied to a user terminal, comprising: receiving data analysis target information in the form of natural language input by a user; Sending the data analysis target information to a server where a first language model is deployed; The first language model is used to decompose the data analysis target information to obtain first task sequence information; the first task sequence information includes a plurality of task information and task execution logic information of the plurality of task information; the plurality of task information is used to be executed by a plurality of intelligent agents according to the task execution logic information to generate data analysis results corresponding to the data analysis target information; Obtaining the data analysis result fed back by the server; The data analysis result is displayed on a page of the user terminal.

17. The method of claim 16, wherein: The data analysis result is obtained based on the multiple task result information generated by the multiple agents; the method further includes: Acquire the plurality of task result information fed back by the server; wherein a first progress of the user terminal acquiring the plurality of task result information is consistent with a second progress of the server generating the plurality of task result information; The plurality of task result information is displayed on a page of the user terminal according to the first progress.

18. The method of claim 17, wherein: The task result information is obtained based on multiple step execution results generated by multiple data analysis tools; the method further includes: Obtaining the execution results of the multiple steps fed back by the server; a third progress of the user terminal obtaining the execution results of the multiple steps is consistent with a fourth progress of the server generating the execution results of the multiple steps; The execution results of the multiple steps are displayed on a page of the user terminal according to the third progress.

19. A data analysis device, applied to a server, comprising: A data analysis target information acquisition module is used to obtain the user's data analysis target information in the form of natural language; a target decomposition module, configured to input the data analysis target information into a first large language model, and obtain first task sequence information obtained by decomposing the data analysis target information by the first large language model; the first task sequence information includes a plurality of task information and task execution logic information of the plurality of task information; A task result information determination module is used to call multiple agents to execute the multiple task information according to the task execution logic information, and obtain multiple task result information generated by the multiple agents; The data analysis result generating module is used to determine the data analysis result corresponding to the data analysis target information based on the multiple task result information.

20. A data analysis device, applied to a user terminal, comprising: A data analysis target information receiving module is used to receive data analysis target information in natural language form input by a user; a data analysis target information sending module, configured to send the data analysis target information to a server deployed with a first large language model; the first large language model is configured to decompose the data analysis target information to obtain first task sequence information; the first task sequence information includes a plurality of task information and task execution logic information for the plurality of task information; the plurality of task information is configured to be executed by a plurality of intelligent agents according to the task execution logic information to generate data analysis results corresponding to the data analysis target information; A data analysis result acquisition module, configured to acquire the data analysis result fed back by the server; The data analysis result display module is used to display the data analysis result on the page of the user terminal.

21. A computing device comprising a memory, a processor, and computer instructions stored in the memory and executable on the processor, wherein the processor implements the steps of the data analysis method according to any one of claims 1 to 18 when executing the computer instructions.

22. A computer-readable storage medium storing computer instructions, wherein when the computer instructions are executed by a processor, the steps of the data analysis method according to any one of claims 1 to 18 are implemented.

23. A computer program product comprising a computer program / instruction, which, when executed by a processor, implements the steps of the data analysis method according to any one of claims 1 to 18.

Images