Achieving method for intelligent agent data analysis based on large language model

Abstract

The invention discloses an implementation method for intelligent agent data analysis based on a large language model, and the method specifically comprises the following steps: S1, constructing an intelligent agent model which comprises a business index system module, a task planning module and an execution module; s2, a user inputs a real-time query demand to the constructed agent model, and the agent model performs semantic analysis on the input query demand based on a large language model to obtain a query target; s3, inputting the obtained query target into a task planning module, and performing task planning by the task planning module according to the query target and generating a task query result; s4, the execution module visualizes the generated task query result and displays the visualized query result on a display screen, and data analysis and display are completed. According to the method, semantic analysis is carried out on the natural language text by means of the large model agent, the real-time query requirement of the user is quickly and accurately understood, and the interaction experience is improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of data processing, and specifically relates to a method for implementing data analysis based on a large language model intelligent agent. Background Art

[0002] The data analysis screen generally displays business data trends intuitively in the form of charts, which is used to monitor the business in real time and provide support for business decision-making and team communication.

[0003] Existing data analysis screens usually display visual data analysis results in a fixed code logic or BI logic manner. They cannot interact with users in real time, cannot respond to users' real-time or changing needs, and cannot generate visual results outside of the original data analysis rules.

[0004] Large model agent is an automated technology based on the intelligent semantic analysis, reasoning and generation capabilities of large language models, which can process and generate results based on user needs.

[0005] At present, there are few studies on the application of large language agents. Aiming at the problem that traditional data analysis large screens cannot analyze and display real-time changing needs, this paper proposes a data analysis large screen implementation method based on the automation capabilities of large model agents. Summary of the invention

[0006] The purpose of the present invention is to overcome the deficiencies in the above-mentioned prior art and to provide an implementation method for data analysis based on a large language model intelligent agent. The method of the present invention uses a large model intelligent agent to perform semantic analysis on natural language text, quickly and accurately understand the user's real-time query needs, and improve the interactive experience.

[0007] In order to achieve the above invention purpose, the technical solution provided by the present invention is as follows:

[0008] A method for implementing data analysis based on a large language model agent, the method specifically comprises the following steps:

[0009] S1, constructing an agent model, which includes a business indicator system module, a task planning module and an execution module. The business system module classifies different types of indicators according to different business needs and business scenarios. The task planning module is used to plan and execute received tasks, and the execution module is used to execute the planned tasks and output the execution results.

[0010] S2, the user inputs real-time query requirements to the constructed intelligent agent model, and the intelligent agent model performs semantic analysis on the input query requirements based on the large language model to obtain the query target;

[0011] S3, inputting the obtained query target into the task planning module, and the task planning module performs task planning according to the query target and generates a task query result;

[0012] S4, the execution module visualizes the generated task query results and displays the visualized query results on the display screen to complete data analysis and presentation.

[0013] Furthermore, the business indicator system module includes indicator classification, indicator definition and indicator correlation. Indicator classification is to classify indicators into different categories according to different business needs and application scenarios. Indicator definition is to define each indicator and determine the formula and definition of the meaning and calculation method of each indicator; indicator correlation is the relationship and influence between different indicators.

[0014] Furthermore, the business indicator system module is specifically as follows: after the intelligent agent model receives the user's query requirements, the business indicator system determines the original data that needs to be obtained based on the query results obtained and the indicator definition, and obtains data related to the business indicators from the data source; uses the calculation formula defined in the indicator definition to process the data according to the calculation formula, processes the collected data, and calculates the specific values ​​of each business indicator; based on the indicator correlation relationship, analyzes related indicators to generate analysis results and insights that conform to business logic.

[0015] Furthermore, the data sources include databases, ERP systems and CRM systems.

[0016] Furthermore, the task planning module plans the query requirements specifically by: clarifying the task objectives, decomposing the task into multiple subtasks according to the task objectives, defining the relationship between each subtask, including but not limited to forward and backward dependencies, parallel execution, and processing each subtask separately to complete the task planning.

[0017] Furthermore, the intelligent agent model performs task planning through the task planning module based on the business logic of the business system module. At the same time, the task planning module processes and outputs the results after task planning. The task planning module evaluates the output results and sends them to the execution module for visual display.

[0018] Based on the above technical solution, the implementation method of the invention based on large language model agent data analysis has achieved the following technical advantages through practical application:

[0019] 1. The present invention provides a method for implementing data analysis based on a large language model intelligent agent, which uses a large model intelligent agent to perform semantic analysis on natural language text, quickly and accurately understand the user's real-time query needs, and improve the interactive experience.

[0020] 2. The present invention provides a method for implementing data analysis based on a large language model intelligent agent. By combining the business indicator system, the large model is fine-tuned to accurately understand and identify the semantics of the business field and meet specific business needs.

[0021] 3. The present invention provides a method for implementing data analysis based on a large language model intelligent agent. Through a task planning engine, the query target is converted into task steps, thereby realizing automatic planning and execution of tasks and improving efficiency and accuracy. DETAILED DESCRIPTION

[0022] In order to make the purpose, technical scheme and advantages of the present invention clearer, the present invention is described below by specific examples. However, it should be understood that these descriptions are only exemplary and are not intended to limit the scope of the present invention. In addition, in the following description, known knots are omitted.

[0023] A method for implementing data analysis based on a large language model agent, the method specifically comprises the following steps:

[0024] S1, constructing an agent model, which includes a business indicator system module, a task planning module and an execution module. The business system module classifies different types of indicators according to different business needs and business scenarios. The task planning module is used to plan and execute received tasks, and the execution module is used to execute the planned tasks and output the execution results.

[0025] S2, the user inputs real-time query requirements to the constructed intelligent agent model, and the intelligent agent model performs semantic analysis on the input query requirements based on the large language model to obtain the query target;

[0026] By using large model agents to perform semantic analysis on natural language texts, we can quickly and accurately understand users' real-time query needs and improve the interactive experience.

[0027] S3, inputting the obtained query target into the task planning module, and the task planning module performs task planning according to the query target and generates a task query result;

[0028] S4, the execution module visualizes the generated task query results and displays the visualized query results on the display screen to complete data analysis and presentation.

[0029] The business indicator system module includes indicator classification, indicator definition and indicator association. Indicator classification is to classify indicators into different categories according to different business needs and application scenarios. Indicator definition is to define each indicator and determine the formula and definition of the meaning and calculation method of each indicator; indicator association is the relationship and influence between different indicators.

[0030] By combining the business indicator system, the big model can be fine-tuned to accurately understand and identify the semantics of the business domain and meet specific business needs.

[0031] The business indicator system module is specifically as follows: after the intelligent agent model receives the user's query requirements, the business indicator system determines the original data that needs to be obtained based on the query results obtained and the indicator definition, and obtains data related to the business indicators from the data source; uses the calculation formula defined in the indicator definition, processes the data according to the calculation formula, processes the collected data, and calculates the specific values ​​of each business indicator; based on the indicator correlation relationship, analyzes related indicators to generate analysis results and insights that conform to business logic.

[0032] The data sources include databases, ERP systems and CRM systems.

[0033] The task planning module plans the query requirements specifically by: clarifying the task objectives, decomposing the task into multiple subtasks according to the task objectives, defining the relationship between each subtask, including but not limited to forward and backward dependencies and parallel execution, and processing each subtask separately to complete the task planning.

[0034] The intelligent agent model is based on the business logic of the business system module and performs task planning through the task planning module. At the same time, the task planning module processes and outputs the results after task planning. The task planning module evaluates the output results and sends them to the execution module for visual display.

[0035] Example

[0036] The intelligent body data large-screen system device can be a hardware and software integrated machine, or it can be a software system installed on a computer, with the help of the computer's microphone, speaker, display hardware and software, and mouse and keyboard input devices; users input real-time query requirements through voice or text, and the intelligent body performs semantic analysis on the converted or input text.

[0037] Big model knowledge guidance based on business indicator system

[0038] The agent contains a set of business indicator systems, which are targeted at the business domain of the application and include key business indicators and related terms. The business indicator system includes indicator classification, indicator definition and indicator association.

[0039] The framework format of the indicator system is: primary classification, secondary classification, and tertiary classification;

[0040] Indicator name: indicator description, calculation formula, data source, update frequency, application scenario, and warning value;

[0041] Indicator correlation: causal relationship, synergistic relationship, and constraint relationship;

[0042] The indicator system divides indicators into different categories according to different business needs and application scenarios. The following takes financial indicators as an example:

[0043] First-level classification: financial indicators

[0044] Secondary classification: Profitability indicators

[0045] ·Third level classification: gross profit margin

[0046] ·Third-level classification: Net profit margin

[0047] ·Third level classification: operating profit margin

[0048] ·Third-level classification: Return on assets (ROA)

[0049] ·Third level classification: Return on equity (ROE)

[0050] Secondary classification: operational efficiency indicators

[0051] ·Third level classification: Inventory turnover rate

[0052] ·Level 3 classification: Accounts receivable turnover rate

[0053] ·Third-level classification: total asset turnover rate

[0054] ·Third-level classification: Fixed asset turnover rate

[0055] Secondary classification: Debt-paying capacity indicators

[0056] ·Level 3 classification: current ratio

[0057] ·Level 3 classification: quick ratio

[0058] ·Third-level classification: debt-to-asset ratio

[0059] ·Three-level classification: interest coverage ratio

[0060] Secondary classification: Growth capability indicators

[0061] ·Third-level classification: operating income growth rate

[0062] ·Third-level classification: Net profit growth rate

[0063] ·Third-level classification: total asset growth rate

[0064] Secondary classification: cash flow indicators.

[0065] ·Level 3 classification: cash flow from operating activities

[0066] · Level 3 classification: free cash flow

[0067] ·Third level classification: cash flow ratio

[0068] Indicator definition: Define each indicator in detail to ensure that all relevant personnel have a unified understanding of the meaning and calculation method of the indicator. Take "gross profit margin" as an example:

[0069] Indicator name: Gross profit margin

[0070] ·Indicator description: Indicators that reflect the profitability and sales pricing level of enterprise products

[0071] Calculation formula: (operating income - operating costs) / operating income * 100%

[0072] Data source: Company income statement

[0073] Update frequency: quarterly / annual

[0074] Application scenario: Evaluate product pricing strategy and analyze cost control effects

[0075] Warning value: 20% below the industry average;

[0076] Indicator correlation: describes the relationship and impact between different indicators, including:

[0077] Causal relationship: such as "gross profit margin increases → net profit margin may increase (if other expenses remain unchanged)", "inventory turnover rate increases → may lead to an increase in total asset turnover rate", "operating income growth rate increases → may lead to an increase in net profit growth rate";

[0078] Synergistic relationship: such as "accounts receivable turnover rate and operating income growth rate: the joint increase of the two may indicate that the company's collection ability is improving while maintaining good growth", "return on assets (ROA) and return on equity (ROE): the two are usually positively correlated and jointly reflect the company's overall profitability", "current ratio and quick ratio: the two usually change in the same direction and jointly reflect the company's short-term debt repayment ability";

[0079] Constraint relationships: such as "gross profit margin and net profit margin: net profit margin is usually lower than gross profit margin, and the gap between the two reflects the impact of operating expenses", "asset-liability ratio and return on equity (ROE): a higher asset-liability ratio may push up ROE, but also increase financial risks", "operating cash flow and net profit: in the long run, operating cash flow should maintain a certain ratio with net profit, otherwise there may be profit quality problems."

[0080] The process of using the business indicator system:

[0081] 1. The big model receives user queries, analyzes semantics and identifies a set of key business indicators to measure and manage business performance.

[0082] 2. According to the definition of indicators, determine the original data that needs to be obtained, and use RPA to obtain data related to business indicators from various data sources (such as databases, ERP systems, CRM systems, etc.).

[0083] 3. Use the defined calculation formula to process the data according to the calculation formula, process the collected data, and calculate the specific values ​​of various business indicators.

[0084] 4. Analyze relevant indicators based on indicator correlation and generate analysis results and insights that conform to business logic.

[0085] 5. Use data visualization tools (such as charts, dashboards, etc.) to present the calculation results to relevant decision makers in an intuitive form to support business decision-making and performance management;

[0086] The big model is optimized for business text data. It can understand and identify the semantics of the business domain. The big model has been fine-tuned and adapted to the business indicator system and context. The big model's understanding and accuracy of business needs can meet the requirements.

[0087] The fine-tuning process of the big model introduces constraints and rules in the business indicator system, and the semantic representation of model learning is constrained by the system rules. The big model is based on the definition and association of the business indicator system, and the parameters and weights of the model learning are adjusted to conform to the business logic; the intelligent agent containing the business indicator system uses a big model that can accurately understand the semantics of natural language text and analyze the user's real-time customized query needs.

[0088] Intelligent task decomposition and step dynamic planning and scheduling of agents

[0089] The intelligent agent inputs the query target obtained by the large model analysis into the task scheduling engine, and performs intelligent task decomposition and dynamic planning and scheduling of steps to achieve the target; for example, the intelligent scheduling engine first decomposes the target generated based on the user chart into three subtasks: data query, processing, and result generation, and then decomposes and dynamically schedules each subtask again.

[0090] The intelligent decomposition method of tasks is as follows:

[0091] Task decomposition is the process of gradually breaking down complex tasks into smaller, more manageable subtasks, which can help intelligent agents better understand the task logic, gradually reason and complete the task.

[0092] 1. Clarify the task objectives; clarify the overall objectives of the task and ensure a clear understanding of the final output of the task.

[0093] 2. Break down the big task into small tasks. Analyze the current situation and task objectives, consider possible courses of action, and predict the possible results of each course of action. Through this process, the overall task is broken down into 5 small, independent subtasks. Each subtask should have a clear goal and behavioral role description, and be relatively independent so that it can be handled separately.

[0094] 3. Create steps for each subtask. Similar to the above method, further refine each subtask into 5 steps, clarify its step goals and behavioral role descriptions, and clarify the execution order of the steps. Each step should be specific and actionable to ensure that the task can proceed smoothly. Each step corresponds to a task API of the robotic process automation.

[0095] 4. Define the relationship between subtasks. Clarify the relationship between subtasks to ensure that tasks are executed in the correct order. Relationships include but are not limited to dependencies, parallel execution, etc.

[0096] Step Dynamic Planning Scheduling has two functional modules, including:

[0097] Reasoning module: responsible for analyzing the progress of the current task, evaluating the execution results, identifying potential problems and generating adjustment suggestions.

[0098] Action module: responsible for executing the specific task steps and making adjustments based on the suggestions of the reasoning module.

[0099] Steps Dynamic planning scheduling method is as follows:

[0100] 1. Task step execution: The action module executes the task steps in the initial plan in sequence, observing and recording the results of the actions. After each step is completed, the reasoning module evaluates and determines whether the next step of execution needs to be adjusted.

[0101] 2. Result evaluation: The reasoning module analyzes the execution results and uses predefined evaluation criteria (such as relevance, accuracy, completeness, etc.) to determine whether there are problems or room for improvement. If problems are found or improvements are needed, the reasoning module generates adjustment suggestions and feeds them back to the action module.

[0102] 3. Dynamic adjustment: The action module adjusts the task steps according to the suggestions of the reasoning module. The adjustments may include re-planning the steps, splitting the steps into smaller subtasks, and adjusting the execution priority strategy. If new steps are generated: Based on the observation results, new steps may be generated. The new steps are inserted into the task queue and reordered. If steps are skipped or merged: If some steps become unnecessary, they can be skipped. If optimization is found, related steps can be merged.

[0103] The execution result evaluation method is as follows:

[0104] Evaluation Framework

[0105] a) Define the evaluation dimensions: accuracy: whether the results meet expectations; completeness: whether all necessary information is included; relevance: how relevant the results are to the task objectives; efficiency: the time and resources required to complete the task;

[0106] b) Scoring mechanism: Set a scoring standard of 0-10 for each dimension, assign weights to different dimensions (default is equal distribution), and calculate the weighted average score;

[0107] c) Threshold setting: set a minimum acceptable overall score (e.g. 7 / 10) and set individual minimum scores for key dimensions (e.g. accuracy no less than 8 / 10);

[0108] Evaluation process:

[0109] a) Automatic evaluation: Use predefined rules and standards to conduct preliminary evaluation. Use large models to analyze the results and give preliminary scores;

[0110] b) Comparative analysis: Compare the results with historical data or preset benchmarks to detect outliers or inconsistencies;

[0111] c) Semantic understanding: Use a large model to analyze the semantic content of the results and assess whether the results meet the task objectives and context requirements;

[0112] d) Comprehensive evaluation: Combine the results of automatic evaluation, comparative analysis and semantic understanding to generate the final evaluation report and score;

[0113] e) Task decomposition and fine-tuning: If a step has a low score, consider further decomposing it into smaller subtasks. Steps with high scores can be used as templates for decomposing similar tasks;

[0114] f) Task re-decomposition: If the score of a single step is below the threshold, re-planning is triggered immediately;

[0115] g) Dynamic scheduling adjustment: Adjust task priority based on the score: high-scoring tasks may be given higher priority. Low-scoring tasks may need to be re-executed or alternative solutions may be sought.

[0116] h) Periodic review: Regularly (e.g., after each N steps) conduct an overall task evaluation and decide whether the task strategy needs to be significantly adjusted based on the cumulative score;

[0117] i) Termination conditions: set clear task completion criteria and check whether the termination conditions are met after each step;

[0118] j) Failure handling: define the maximum number of attempts or time limit. If failures continue, you can request human handling and adjust the evaluation criteria and weights based on human feedback;

[0119] k) Analysis after task completion: After all tasks are completed, a comprehensive evaluation is conducted again based on the scoring dimensions.

[0120] Example 2

[0121] Query the top three best-selling products in the past three months and their sales trends;

[0122] 1. Receive and analyze users’ real-time query requirements

[0123] The user enters a query by voice in front of the data analysis screen: "What are the three best-selling products in the past three months? What is the monthly sales trend this year?"

[0124] The intelligent agent performs semantic analysis to understand user needs involving two aspects: sales ranking and sales trend.

[0125] 2. Data acquisition and processing process of the intelligent agent:

[0126] The query target is passed into the task planning engine and broken down into multiple task steps;

[0127] The agent automatically writes code to call the API of the company's sales database to obtain the sales data of each product in the last three months;

[0128] The execution engine executes the task steps step by step according to the task plan:

[0129] 1) Get the sales data for the last three months.

[0130] 2) Add up the sales of each product to get the total sales.

[0131] 3) Sort the total sales and determine the top three products in sales.

[0132] 4) Get the sales data of these three products every month this year.

[0133] 3. Generate real-time data analysis visualization results on the data screen

[0134] The result generation planning engine receives the sales data returned by the execution engine, and determines the visualization results to be generated based on the user's query requirements and the business indicator system;

[0135] The data analysis module generates corresponding charts according to the instructions:

[0136] First, generate a bar chart or table showing the top three products in terms of sales in the last three months.

[0137] Next, generate a line chart for each of the top three selling products, showing their sales trends for each month this year.

[0138] The visualization results are displayed in real time on the monitoring screen page for users to view and analyze intuitively.

[0139] result:

[0140] Users can learn in real time on the smart data analysis screen what the three best-selling products are in the past three months, as well as the monthly sales trends of these three products this year. Through the visualization results, users can intuitively compare the sales of different products, helping companies formulate sales strategies and optimize product portfolios.

[0141] Finally, it should be noted that the above embodiments are only used to illustrate the technical solution of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, ordinary technicians in the field should understand that the specific implementation methods of the invention can still be modified or some technical features can be replaced by equivalents without departing from the spirit of the technical solution of the present invention, which should be included in the scope of the technical solution for protection of the present invention.

Claims

1. A method for implementing data analysis based on a large language model agent, characterized in that: The method specifically comprises the following steps: S1, constructing an agent model, which includes a business indicator system module, a task planning module and an execution module. The business system module classifies different types of indicators according to different business needs and business scenarios. The task planning module is used to plan and execute received tasks, and the execution module is used to execute the planned tasks and output the execution results. S2, the user inputs real-time query requirements to the constructed intelligent agent model, and the intelligent agent model performs semantic analysis on the input query requirements based on the large language model to obtain the query target; S3, inputting the obtained query target into the task planning module, and the task planning module performs task planning according to the query target and generates a task query result; S4, the execution module visualizes the generated task query results and displays the visualized query results on the display screen to complete data analysis and presentation.

2. The method for implementing data analysis based on a large language model agent according to claim 1, characterized in that: The business indicator system module includes indicator classification, indicator definition and indicator association. Indicator classification is to classify indicators into different categories according to different business needs and application scenarios. Indicator definition is to define each indicator and determine the formula and definition of the meaning and calculation method of each indicator; indicator association is the relationship and influence between different indicators.

3. The method for implementing data analysis based on a large language model agent according to claim 2, characterized in that: The business indicator system module is specifically as follows: after the agent model receives the user's query demand, the business indicator system determines the original data to be obtained according to the query results obtained and the indicator definition, and obtains data related to the business indicator from the data source; Use the calculation formula defined in the indicator definition, process the data according to the calculation formula, process the collected data, and calculate the specific values ​​of various business indicators; Analyze related indicators based on indicator correlation to generate analysis results and insights that conform to business logic.

4. The method for implementing data analysis based on a large language model agent according to claim 3, characterized in that: The data sources include databases, ERP systems and CRM systems.

5. The method for implementing data analysis based on a large language model agent according to claim 1, characterized in that: The task planning module plans the query requirements specifically by: clarifying the task objectives, decomposing the task into multiple subtasks according to the task objectives, defining the relationship between each subtask, including but not limited to forward and backward dependencies and parallel execution, and processing each subtask separately to complete the task planning.

6. The method for implementing data analysis based on a large language model agent according to claim 5, characterized in that: The intelligent agent model is based on the business logic of the business system module and performs task planning through the task planning module. At the same time, the task planning module processes and outputs the results after task planning. The task planning module evaluates the output results and sends them to the execution module for visual display.