Multi-agent collaborative data visualization analysis method, equipment and medium

Abstract

The embodiment of the invention discloses a multi-agent collaborative data visualization analysis method and device and a medium, and relates to the technical field of visualization analys.The method comprises the steps that a natural language analysis request input by a user is received, task splitting is conducted on the natural language analysis request through a preset task planning agent, and the task splitting result is obtained; generating corresponding task plan information, obtaining real-time operation data of a preset functional agent cluster, carrying out agent distribution on the plurality of sub-tasks based on the real-time operation data and the task plan information, determining a target functional agent corresponding to each sub-task, and sending the target functional agent to a server; the target function agent comprises any one of a data analysis agent and a visual display agent; and executing the corresponding data analysis subtask through the data analysis agent to obtain a data analysis result, performing agent cooperative verification on the data analysis result, and after the verification is passed, generating visual analysis data corresponding to the data analysis result through the visual display agent.

Description

Technical Field

[0001] This specification relates to the field of visualization analysis technology, and in particular to a method, device, and medium for multi-agent collaborative data visualization analysis. Background Art

[0002] In the field of data analysis and visualization, data analysis visualization is achieved through business intelligence tools and automated reporting systems. Business intelligence tools utilize an "ETL+OLAP+Visualization" architecture for fixed-mode analysis, while automated reporting systems generate formatted output based on pre-set scripts. However, business intelligence tools rely on manual intervention, requiring users to manually determine analysis dimensions, select analysis methods, and visualization formats, increasing the user barrier to entry. Automated reporting systems also have fixed analysis dimensions, and most systems pre-set fixed analysis dimensions, making it difficult to flexibly combine them for multi-dimensional analysis based on needs.

[0003] In addition to the aforementioned approaches, with the development of large-scale model technology, existing techniques utilize a single agent for automated analysis. This agent coordinates and controls the entire process, using a core algorithm module to manage all analytical processes, including data collection, processing, and visualization. However, relying on a single agent for the entire process poses challenges such as an inability to directly understand the user's natural language and generating unexpected results. Furthermore, as analytical requirements increase, the complexity of a single agent system grows exponentially, making maintenance and upgrades difficult. Therefore, traditional data analysis visualization methods rely on manual intervention, cannot meet users' multi-dimensional analysis needs, and analytical efficiency needs to be improved. Summary of the Invention

[0004] One or more embodiments of this specification provide a multi-agent collaborative data visualization analysis method, device, and medium for solving the following technical problems: In traditional data analysis visualization methods, manual intervention is relied upon, which cannot meet the user's multi-dimensional analysis needs, and the analysis efficiency needs to be improved.

[0005] One or more embodiments of this specification adopt the following technical solutions:

[0006] One or more embodiments of the present specification provide a multi-agent collaborative data visualization analysis method, the method comprising: receiving a natural language analysis request input by a user, performing task decomposition on the natural language analysis request through a preset task planning agent, and generating corresponding task planning information, wherein the task planning information includes multiple subtasks and task attribute information corresponding to each subtask; obtaining real-time operation data of a preset functional agent cluster, and performing agent allocation on the multiple subtasks based on the real-time operation data and the task planning information, and determining a target functional agent corresponding to each subtask, wherein the target functional agent includes any one of a data analysis agent and a visualization display agent; executing corresponding data analysis subtasks through the data analysis agent to obtain data analysis results, performing agent collaborative verification on the data analysis results, and after the verification passes, generating visualization analysis data corresponding to the data analysis results through a visualization display agent.

[0007] One or more embodiments of this specification provide a multi-agent collaborative data visualization analysis device, including:

[0008] at least one processor; and,

[0009] a memory communicatively connected to the at least one processor; wherein,

[0010] The memory stores instructions that can be executed by the at least one processor. The instructions are executed by the at least one processor to enable the at least one processor to perform the above method.

[0011] One or more embodiments of this specification provide a non-volatile computer storage medium storing computer-executable instructions, wherein the computer-executable instructions are configured to execute the above method.

[0012] At least one of the above-mentioned technical solutions adopted in the embodiments of this specification can achieve the following beneficial effects: Through the technical solutions of the embodiments of this specification, through the semantic completion of the task planning agent and the dimension decomposition driven by the knowledge graph, the high threshold problem of the traditional solution relying on manual experience to define the analysis dimension is overcome, and the vague natural language request is converted into a task plan, which significantly reduces the complexity of user operations and improves the accuracy of demand understanding; the dynamic agent allocation mechanism based on real-time operation data solves the problem of capability mismatch and resource idleness caused by static rules or polling strategies, and optimizes the overall utilization of the cluster through the precise matching of task characteristics and agent capabilities; an end-to-end trusted closed loop is constructed through the collaborative verification of data lineage markers, the data source fingerprint ensures the timeliness of the original data, the execution agent fingerprint verifies the consistency of the calculation process, and the visualization agent fingerprint predicts the compatibility of the results, avoiding the risks existing in the traditional single result review method; through the intelligent matching of the number of dimension fields, indicator type and time series markers, it replaces the fixed output method of manual specification or rule binding, and realizes the intelligent expression of cross-dimensional analysis results. BRIEF DESCRIPTION OF THE DRAWINGS

[0013] 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 the embodiments or the description of the prior art. Obviously, the drawings described below are only some of the embodiments described in this specification. For those skilled in the art, other drawings can be obtained based on these drawings without inventive work. In the drawings:

[0014] Figure 1 A flowchart of a multi-agent collaborative data visualization analysis method provided in an embodiment of this specification;

[0015] Figure 2 A structural diagram of a multi-agent collaborative data visualization analysis device provided in an embodiment of this specification. DETAILED DESCRIPTION

[0016] To help those skilled in the art better understand the technical solutions in this specification, the following will provide a clear and complete description of the technical solutions in the embodiments of this specification, in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of this specification, not all of them. All other embodiments obtained by those skilled in the art based on the embodiments of this specification without creative work should fall within the scope of protection of this specification.

[0017] The embodiments of this specification provide a data visualization analysis method for multi-agent collaboration. It should be noted that the execution subject in the embodiments of this specification can be a server or any device with data processing capabilities. Figure 1 This is a flow chart of a multi-agent collaborative data visualization analysis method provided in the embodiment of this specification, such as Figure 1 As shown, it mainly includes the following steps:

[0018] Step S101: Receive a natural language analysis request input by a user, perform task decomposition on the natural language analysis request through a preset task planning agent, and generate corresponding task planning information.

[0019] The task plan information includes multiple subtasks and task attribute information corresponding to each subtask;

[0020] In data analysis and visualization scenarios, user natural language requests often suffer from issues such as ambiguity, missing dimensions, and unspoken implicit requirements. For example, when a user requests analysis of sales anomalies, key elements such as the time range, regional granularity, and anomaly determination criteria are not specified. Direct execution of the request can result in analytical results that deviate from actual requirements.

[0021] In one embodiment of this specification, deep semantic completion and structured decomposition are performed through a task planning agent. The ambiguity of natural language can cause drift in analysis targets, necessitating semantic disambiguation based on historical user behavior and industry knowledge. Complex analysis requests imply multi-dimensional requirements, such as simultaneously examining product lines, channels, and time periods. These requirements must be decoupled into independent subtasks for parallel processing. Furthermore, logical dependencies between tasks, such as the need to complete total sales calculations before performing regional share analysis, necessitate establishing a clear task scheduling sequence.

[0022] Through the preset task planning agent, the natural language analysis request is decomposed into tasks to generate corresponding task planning information, specifically including: calling the large language model through the task planning agent, performing semantic completion on the natural language analysis request based on the pre-acquired user historical interaction data and predefined industry knowledge data, and generating user intention completion information; according to the task planning agent, the user intention completion information is decomposed into tasks to determine multiple subtasks and corresponding task attribute information, wherein the task attribute information includes task type and task dependency.

[0023] In one embodiment of the present specification, the task planning agent calls a large language model, loads a user historical interaction dataset containing all the user's operation records and result correction behaviors in the past three months, and loads a pre-built industry knowledge graph, such as the association rules of sales-inventory turnover-promotional activities in the retail industry. The natural language request, the user historical interaction dataset, and the industry knowledge graph are input into the large language model, and the attention mechanism is used to identify key information gaps. For example, when the user enters "compare store performance", the model automatically retrieves the user's history records and finds that 90% of its queries involve "stores in East China". At the same time, it extracts the rule of "performance must include sales per square meter index" from the industry knowledge base, and finally completes it to "compare the sales and sales per square meter of stores in East China in the past three months." A confidence threshold is set in the completion process, and a follow-up interaction is generated when the credibility of the completion element is insufficient, such as "please specify the comparison time period."

[0024] Based on the semantic completion results, perform task decomposition and attribute labeling. Use named entity recognition technology to extract core analysis objects, such as "sales volume" " Store " Quarterly", matching associated dimensions in the pre-built system knowledge graph. Taking the "sales" entity as an example, analysis axes such as time dimension, space dimension, and product dimension are automatically associated in the retail knowledge graph. After generating independent dimension subtasks for each analysis dimension, they are decomposed into atomic operations according to functional requirements. For example, "time dimension analysis" is decomposed into "extracting sales data for the past 12 months" and "calculating monthly year-on-year growth", which correspond to data acquisition subtasks and calculation subtasks respectively. In this process, task type labels are marked, data operation types are marked as data analysis types, and result integration types are marked as visualization types. The logical relationship between subtasks is parsed through the directed graph engine. When subtask B depends on the output of task A, such as "regional share analysis" requires the "total sales calculation" result, a dependency link from A to B is automatically established; tasks without logical coupling, such as "clothing category analysis" and "home appliance category analysis", are marked as parallel tasks. Finally, a task plan tree with attribute labels is output.

[0025] Through dynamic semantic completion and knowledge graph-driven task decomposition, user portraits and industry knowledge are integrated to achieve accurate restoration of intent, thereby improving the success rate of converting fuzzy requests into executable instructions; the automatic dimension association mechanism based on the knowledge graph shortens the access time of new business dimensions, and through explicit modeling of task dependencies, supports automatic load balancing of parallel tasks and optimal scheduling of serial tasks, thereby improving overall processing efficiency in complex multi-dimensional analysis scenarios.

[0026] The user intention completion information is task-decomposed to determine multiple subtasks and corresponding task attribute information, specifically including: performing intent recognition based on the user intention completion information to determine multiple corresponding target analysis entities, and performing entity matching and association relationship query in a pre-built system knowledge graph based on the target analysis entity to determine the association analysis dimension and at least one corresponding data source information corresponding to each target analysis entity; according to the multiple association analysis dimensions, determining the dimensional subtask corresponding to each association analysis dimension, performing task function decomposition on each of the dimensional subtasks, and determining multiple subtasks corresponding to each dimensional subtask; performing logical association analysis on the multiple subtasks to determine the task dependency relationship between the multiple subtasks, and determining the task type of each subtask based on the task function of each subtask, wherein the task type includes a data analysis type and a visualization display type, and the data analysis type includes any one or more of trend analysis, distribution statistical analysis, predictive analysis, comparative analysis, and attribution analysis.

[0027] In complex data analysis scenarios, user-intended information completion still faces the deep-seated problem of multiple entities intertwined and dimension coupling. For example, a completion request such as "Analyze quarterly smartphone sales in East China and compare them with competing products" includes four entities: region, product, time, and competitor. Generating a query directly without separating entity-associated dimensions will lead to dimensional confusion. For example, smartphones require associated specifications, while competitors require associated market share. Furthermore, different entities correspond to independent data sources: sales data resides in the ERP system, while competitor data requires crawling public market reports, creating data source conflicts.

[0028] In one embodiment of this specification, task decomposition is achieved through entity recognition and knowledge graph matching based on user intent to complete information. First, a domain-enhanced named entity recognition model is used to distinguish between core analysis entities and modified entities. For example, the smartphone in the example is the core entity, and the East China region, quarter, etc. are modified entities. For the abstract entity of "competitor", it is expanded into a specific entity set through the industry knowledge base.

[0029] A system knowledge graph is pre-built to extract structured and semi-structured data from heterogeneous data sources, including core enterprise databases (such as table schemas and primary and foreign key constraints in ERP and CRM systems), industry-standard documents, and public business reports. An automated schema mapping engine transforms the original data structure into a unified ontology model, identifies the business semantics of database table fields, and generates initial entity types. Industry documents are parsed to construct dimensional hierarchies, such as inheritance relationships between time dimensions, quarters, and months. Entity relationship triples from business reports, such as "promotional activity, impact, sales," are extracted to expand the business rule base. Records describing the same entity in multi-source data are aligned, calculating the Euclidean distance for numerical attributes and the semantic similarity for textual attributes. When the matching degree exceeds a preset threshold, records are merged into a single entity node. For example, if the price difference between different systems for the same product exceeds a preset threshold, records that cannot be merged are arbitrated based on data freshness and source authority weights, such as a weight of 0.7 for ERP data and 0.3 for crawler data. Relationship completion is achieved using a rule engine and graph neural networks. Business rules are applied to infer implicit relationships. For example, the "Purchase mobile phone accessories" event triggers the generation of the relationship <mobile phone model, has_accessory, accessory model>. Graph neural networks are also used to predict potential associations. For example, based on the co-occurrence pattern of "coffee machine" and "coffee beans," a complementary product relationship can be established.

[0030] In the pre-built system knowledge graph, the associated edges are traversed using entities as nodes. For example, the "smartphone" node is associated with specification dimensions such as screen size and processor model, market dimensions such as regional sales ranking and price range, and time series dimensions such as quarterly growth rate and year-over-year growth rate. Each associated dimension is automatically bound to a data source, such as mapping the specification dimension to a product database and the market dimension to a third-party data API. Executable task units are generated based on the associated dimensions, with independent subtasks generated for each associated dimension, and the corresponding dimensional subtasks are determined for each associated analysis dimension. For example, the time dimension creates a quarterly smartphone sales trend analysis, while the competitive dimension creates a comparative analysis of the market share of major brands.

[0031] Decompose the dimensional subtasks. First, generate a data extraction task based on the bound data source, such as pulling Q1-Q4 sales data from the ERP system. Then break it down by analysis type, for example, trend analysis requires moving average calculation. Data acquisition and calculation are marked as data analysis type, and the data analysis type is further broken down by analysis type to obtain any one or more of trend analysis, distribution statistics analysis, predictive analysis, comparative analysis, and attribution analysis. Tasks for integrating the results and generating charts are marked as visualization presentation type. Identify cross-task data input and output relationships. When task B requires the output fields of task A, such as when market share calculation requires obtaining sales figures for each brand, establish an A-B dependency edge. Shared resources are marked when concurrently accessing the same database table, and mutex locks are added to write tasks. Tasks without dependencies (such as sales analysis for East and North China) are marked as parallel groups. Serial sequences are constructed for continuous dependency chains, such as data cleaning, indicator calculation, and visualization. The final output is a task network with execution priorities and resource constraints.

[0032] Step S102: obtaining the real-time operation data of the preset functional agent cluster, and performing agent allocation for multiple subtasks based on the real-time operation data and task planning information, and determining the target functional agent corresponding to each subtask.

[0033] The target function agent includes any one of a data analysis agent and a visualization display agent;

[0034] Based on the real-time operation data and the task planning information, the multiple subtasks are assigned agents, and the target functional agent corresponding to each subtask is determined, specifically including: classifying the multiple subtasks according to the task type in the task attribute information, and determining the data analysis subtask of the data analysis type and the visualization subtask of the visualization display type; extracting task features of the data analysis subtasks, and determining the multidimensional task description features corresponding to each data analysis subtask, wherein the multidimensional task description features include the data analysis type and the data source type; performing agent matching in the functional agent cluster through the multidimensional task description features corresponding to each data analysis subtask and the real-time operation data, and determining the analysis functional agent corresponding to each data analysis subtask; and determining the visualization functional agent corresponding to each visualization subtask based on the analysis functional agent corresponding to each data analysis subtask and the task dependency in the task attribute information. According to the analysis function agent corresponding to each data analysis subtask and the task dependency relationship in the task attribute information, the visualization function agent corresponding to each visualization subtask is determined, specifically including: determining the upper-level data analysis subtask that the visualization subtask depends on according to the task dependency relationship; obtaining the output data features of the analysis function agent corresponding to the upper-level data analysis subtask, using the output data features as query features, matching in the function agent cluster, and determining the visualization function agent corresponding to the visualization subtask.

[0035] In a dynamic multitasking scenario, when the system simultaneously processes tasks like "forecasting mobile phone sales in the East China region for the next quarter" and "analyzing competitive product market share," the latter requires a time-series forecasting agent, while the latter requires a crawler analysis agent. Using static allocation rules can lead to timeouts for high-load agents like the SQL execution agent due to task backlogs, while specialized agents like the image recognition agent remain idle for extended periods. Furthermore, visualization subtasks are incompatible with pre-processed analysis results. For example, a time-series forecasting agent outputs a multivariate array, but a visualization agent that only supports single-dimensional pie chart rendering is incorrectly assigned. By implementing a dynamic allocation mechanism driven by real-time data and precisely matching task characteristics with agent capabilities, the system achieves a balanced balance between resource utilization and task adaptability.

[0036] In one embodiment of the present specification, first, based on the task type classification identifier in the task attribute information, the subtasks are divided into data analysis subtasks and visualization display subtasks; for the data analysis subtasks, the multidimensional task description features including the data analysis type and the data source type are extracted, and combined with the real-time operation data of the functional agent cluster, the real-time operation data here includes CPU load, memory usage and queue depth, the capability matching degree and load evaluation coefficient of each agent are calculated, and the agent with the best comprehensive score is selected as the analysis functional agent. When performing task assignment for the visualization display subtask, the upper-level data analysis subtask on which it depends is located according to the task dependency relationship, and the output data features of the analysis functional agent corresponding to the upper-level task are obtained. The output data features here include data structure, field type and data scale, which are used as query features to match the visualization functional agent with corresponding data processing and rendering capabilities in the functional agent cluster. Through dynamic capability matching and dependency-aware binding, the task execution efficiency and result output quality in a high-concurrency environment are ensured.

[0037] Specifically, subtasks are divided into data analysis and visualization display types based on task attribute information. For data analysis subtasks, multi-dimensional task description features are extracted. The computing model requirements are determined based on the task type label. For example, attribution analysis requires Shapley value computing capabilities, and trend analysis requires a time series prediction model. And identify the data source characteristics bound to the task, such as MongoDB queries require document processing agents, and Kafka stream processing requires window computing agents. Combined with various real-time operating data such as agent CPU / memory load, task queue length, and historical success rate, the capabilities of the agent are evaluated. For each task to be assigned, scan the functional agent cluster for agents with corresponding capabilities. For example, the attribution analysis task matches the agent registered with the Shapley algorithm, and calculate the comprehensive score of the available agents. The comprehensive score calculation method here can refer to the following formula: Score = Capability matching coefficient × 70% + (1-current load rate)

[0038] ×30%, selecting the agent with the highest overall score. A backup instance is activated when the highest-scoring agent's load exceeds 80%. Visualization subtasks are assigned using dependency-driven reverse matching, locating upstream data analysis subtasks based on task dependencies. For example, the "Market Share Comparison Chart" visualization task depends on the "Brand Sales Calculation" analysis task. Output metadata from the upstream analysis agent is obtained, including data structure, field type, and data size. For example, if the output is a JSON array containing {brand, sales, growth rate}, then the functional agent cluster is scanned for visualization agents that meet the criteria.

[0039] Through the above technical solutions, the system performance and output quality in complex data analysis scenarios are significantly improved through dynamic matching driven by task characteristics and collaborative binding based on dependency perception. Compared with the traditional static allocation solution, the real-time perception of the load status of the intelligent agent cluster and the precise matching of task characteristics effectively solves the processing bottleneck caused by resource mismatch, while ensuring timely response of high-priority tasks and optimizing cluster resource utilization efficiency. Secondly, at the task collaboration level, the reverse dependency binding mechanism of the visual intelligent agent completely overcomes the inherent defects of the separation of the front and back links, ensuring seamless connection between analysis results and visual rendering, greatly reducing the risk of execution interruption due to incompatible data formats, improving analysis timeliness, ensuring the accuracy of results, and meeting the analysis needs of multi-dimensional data analysis scenarios.

[0040] Step S103, execute the corresponding data analysis subtask through the data analysis agent to obtain the data analysis results, perform agent collaborative verification on the data analysis results, and after the verification is passed, generate visual analysis data corresponding to the data analysis results through the visual display agent.

[0041] In one embodiment of the present specification, corresponding data analysis subtasks are executed by corresponding data analysis agents, and the data analysis results are collaboratively verified by the agents.

[0042] The data analysis result is subjected to intelligent agent collaborative verification, specifically including: determining the data lineage marking information of the data analysis result, wherein the data lineage marking information includes the data source fingerprint information, the execution intelligent agent fingerprint information and the target input visualization function intelligent agent fingerprint information; based on the data lineage marking information, the data analysis result is subjected to multi-level verification to determine the verification result.

[0043] In distributed agent collaboration scenarios, data analysis results face multiple credibility risks. The diversity of data sources can lead to delays in underlying data updates. For example, offline data warehouses may not synchronize real-time orders, resulting in distorted analysis results. Algorithmic differences between different agent versions can lead to computational logic deviations. Furthermore, the rendering rules and constraints of visualization agents can easily lead to data structure incompatibility. The lack of a systematic verification mechanism can lead to a series of problems, including distorted results and rendering failures.

[0044] In one embodiment of the present specification, data lineage tagging information is first generated synchronously when the data analysis agent outputs results. The data lineage tagging information includes data source fingerprint information, execution agent fingerprint information, and target input visualization function agent fingerprint information. The data source fingerprint is generated by hashing the data source type, connection address, and data snapshot version, and is used to uniquely identify the data source; the execution agent fingerprint is generated based on the agent identifier, model version, and execution parameters, and is used to track the computing subject; and the target visualization function agent fingerprint is generated based on the target agent identifier and the set of chart rendering rules it supports, and is used to predict the compatibility of the results.

[0045] At the data traceability layer, the data source fingerprint is used to verify the data connection status and snapshot timeliness. If a data version lag is detected, an update process is triggered. At the processing and verification layer, the original parameters are retrieved based on the execution agent fingerprint, and the calculation is re-executed on some data samples. When the deviation between the new result and the original result exceeds the preset tolerance range, the verification is determined to have failed. At the visualization pre-check layer, the rendering rule set of the target visualization agent is obtained through its fingerprint. For example, if a heat map requires the input of a three-dimensional data matrix, the result data structure is verified to ensure that it meets the rule constraints. In scenarios where verification fails, the abnormal link is automatically isolated and intelligent weight distribution or parameter optimization is triggered. If the verification is successful, a verification certificate containing an integrity score is generated.

[0046] Through the above technical solutions, a cryptographic fingerprint system covering data sources, computing entities and output targets is built, making full-link traceability possible; a layered verification mechanism is established to achieve closed-loop verification from data freshness, computing consistency to rendering compatibility, making hidden risks explicit; a dynamic self-healing strategy is designed to accurately trigger the repair process according to the failure type, significantly reducing the need for manual intervention, and can meet the business needs of complex business scenarios.

[0047] Based on the data lineage marker information, the data analysis result is subjected to multi-level verification to determine the verification result, specifically including: through the data source fingerprint information, the data source of the data analysis result is traced back to verify the data source connection status and the timeliness of the data snapshot version; the corresponding agent execution information is obtained through the execution agent fingerprint information, and the result deviation rate is verified by re-executing the calculation on some data samples to perform a processing process verification on the data analysis result; the chart rendering rule set is obtained through the target visualization function agent fingerprint information to verify the compatibility of the data structure of the data analysis result with the chart rendering rule set to perform a visualization compatibility pre-verification on the data analysis result; when the data traceability verification, the processing process verification and the visualization compatibility pre-verification all meet the preset requirements, the verification result is determined to be a verification pass.

[0048] In one embodiment of this specification, when performing multi-level verification on the data analysis results, data provenance verification is first performed. Based on the data source fingerprint information, the metadata center is accessed to verify the data source connection status and snapshot timeliness. If the snapshot version is detected to be lagging behind the current business time window, such as when analyzing Q2 data and the snapshot is still at the end of Q1, the incremental update process is automatically triggered. If the data source connection is abnormal, such as the database service is unreachable, the backup data source is switched and the snapshot is reinitialized.

[0049] Next, the processing process is verified. The original execution environment configuration is retrieved from the registry using the agent's fingerprint information. A random sample of data, such as 10% of the original input data, is randomly selected and re-executed in a sandbox environment using the same parameters. The deviation distribution of the newly generated results is compared with the original results. If the deviation rate of key indicators exceeds a dynamic threshold, such as a deviation of >1% for total sales, the calculation logic is determined to be abnormal and a rollback of the agent version is triggered.

[0050] Finally, a visualization compatibility pre-check is performed, using the target visualization agent's fingerprint information to obtain its rendering rule constraint set. This verifies the compatibility of the data structure with the rules. First, the field dimension matches are checked. For example, heat maps require at least two-dimensional matrix data input. Next, field type compatibility is verified. For example, time series charts require time fields to be of the DateTime type. Finally, outlier distribution is detected. For example, pie charts require categorical fields to have ≤20 unique values. Incompatible fields are automatically converted, such as converting textual region names to geocodes, or optimization suggestions are generated.

[0051] When all three levels of verification pass, a verification certificate is generated. This certificate contains the data lineage integrity score and verification timestamp, serving as a pass for subsequent processes. Failure at any stage triggers a self-healing mechanism based on the error type: data issues trigger re-collection, computational deviations trigger intelligent weight distribution, and compatibility issues initiate rule negotiation.

[0052] By building a full-link traceability system anchored by cryptographic fingerprints, the data source can be verified, the computing subject can be located, and the output target can be predicted; a hierarchical and progressive verification path is designed to form a closed-loop verification network from data timeliness, computing consistency to rendering compatibility; in cross-domain analysis scenarios, such as joint dashboards integrating sales data and social media public opinion, the credibility of the results is significantly improved.

[0053] By visually displaying an intelligent agent, visual analysis data corresponding to the data analysis result is generated, specifically including: parsing the data structure characteristics of the data analysis result to extract the number of dimension fields, indicator field type and time series flag corresponding to the data analysis result; determining the chart type according to the number of dimension fields, the indicator field type and the time series flag, and generating visual analysis data corresponding to the data analysis result based on the chart type.

[0054] In dynamic multi-dimensional analysis scenarios, there may be a risk of incompatibility between data structure characteristics and chart types. Traditional systems use preset chart rules or manual specification methods. However, fixed mapping rules cannot adapt to data structure variations, and manual intervention disrupts the automated process.

[0055] In one embodiment of the present specification, the visualization display agent first analyzes the data structure characteristics of the data analysis results and identifies the number of dimension fields. The number and hierarchical relationship of statistical classification fields, such as the three-level regions of "province-city-district" are counted as 3 dimensions, and the core dimension (analysis target subject) and the modified dimension (screening condition) are distinguished; secondly, the indicator field type is determined, the statistical properties of the numerical field are identified, and the aggregation method is detected, such as the sum method, the average method, the counting method, etc. Finally, the time series mark detection is performed to verify the existence of ISO8601 format fields and their continuity, continuous day granularity or discrete quarterly points. Using the feature weight model, the time series mark triggers the timing chart, preferably a line chart or an area chart, and the single-dimensional classification field matches the comparison chart, such as a bar chart and a pie chart. The multi-dimensional field activates the associated chart, such as a heat map and a radar chart. Perform a discreteness analysis on the classification field, detect the magnitude of the numerical field, and optimize the time field granularity, such as automatically aggregating to the best display granularity according to the data density. When task dependencies indicate multi-dimensional relationships, such as the linkage between regional distribution and product line trends, a combined chart solution is generated, overlaying trend curves on a map and injecting interactive events so that users can click on a region to highlight the associated product line. Finally, based on the selected chart type, the rendering engine is invoked to generate a responsive visualization component.

[0056] Through the above technical solutions, adaptive decision-making driven by data structure characteristics can ensure that chart selection accurately matches analysis intent and data characteristics; through constraint-guided rendering optimization, the distribution and magnitude differences of discrete data can be automatically handled; and through task-dependency-aware composite chart generation, the collaborative expression of multi-perspective data can be achieved.

[0057] Through the technical solutions of the embodiments of this specification, through the semantic completion of the task planning agent and the dimension decomposition driven by the knowledge graph, the high threshold problem of the traditional solution relying on manual experience to define the analysis dimension is overcome, and the vague natural language requests are converted into task plans, which significantly reduces the complexity of user operations and improves the accuracy of demand understanding; the dynamic agent allocation mechanism based on real-time operation data solves the problem of capability mismatch and resource idleness caused by static rules or polling strategies, and optimizes the overall cluster utilization through the precise matching of task characteristics and agent capabilities; an end-to-end trusted closed loop is constructed through the collaborative verification of data lineage markers, the data source fingerprint ensures the timeliness of the original data, the execution agent fingerprint verifies the consistency of the calculation process, and the visualization agent fingerprint predicts the compatibility of the results, avoiding the risks existing in the traditional single result review method; through the intelligent matching of the number of dimension fields, indicator types and time series marks, it replaces the fixed output method of manual specification or rule binding, and realizes the intelligent expression of cross-dimensional analysis results.

[0058] The embodiments of this specification also provide a multi-agent collaborative data visualization analysis method, which receives natural language query requests input by users. The requirements describe the user's intention for data analysis, such as "Help me analyze the user orders of the current system from at least three dimensions." After receiving the user request, the question can also be rewritten based on the context to ensure that the problem description is clearer.

[0059] The task planning agent breaks down tasks according to user needs and assigns each task to a different agent. The role prompt defined by the task planning agent is similar to the following: " Understand each agent and its capabilities given below. Based on the user question, accurately grasp the user's intent and generate a task plan, ensuring that the plan can achieve the set goal through collaborative cooperation between agents without additional user assistance. You can add some restrictive prompts, such as the following: "Each step in the task plan should exist to advance the solution of the user's goal. Do not generate meaningless task steps. Ensure that the goal within each step is clear and complete. Each step should ideally use only one resource to complete a sub-goal. If the current goal can be decomposed into multiple sub-tasks of the same type, generate independent parallel tasks." Based on the user question, the generated task plan example is as follows:

[0060] ① Please count the changing trends of the current system user orders in different time periods, such as the changes in order quantity and order amount by day, week, and month - Data Analysis Agent

[0061] ② Please count the order quantity and amount distribution of users of different user types or different regions - Data Analysis Agent

[0062] ③ Please analyze the number and proportion of orders in each order status (such as pending payment, paid, shipped, completed, and canceled) in the current system - Data Analysis Agent

[0063] ④ Integrate the above analysis data and draw a picture to show it - visually display the intelligent agent

[0064] Assign the above tasks to different agents for execution.

[0065] The agent roles and definitions in this example are as follows:

[0066] The data analysis agent is responsible for extracting required fields from data sources and performing aggregate calculations and indicator analysis. Specifically, this involves connecting to various data sources, identifying relevant tables and fields within them based on the user's question, converting the user's question into SQL statements, and querying the relevant data. The SQLDatabaseSequentialChain or SQLDatabaseChain methods in langchain can be used to convert natural language into SQL statements and then execute database queries. The visualization agent is responsible for integrating the structured analysis results of multiple subtasks into a unified report structure for front-end presentation.

[0067] The task planning agent integrates the execution results of all the above tasks, obtains the results of the visualization agent, and displays them to the user. The display format includes but is not limited to pie charts, bar charts, line charts or area charts. The chart content covers different analysis dimensions (such as product category, user, month, etc.) to ensure that the multi-dimensional analysis requirements in the original analysis request are met.

[0068] Compared to existing technical solutions that rely on a single model or fixed template for data query and analysis, this invention introduces a multi-role agent collaboration mechanism. This decomposes natural language requests into several clearly structured and objective subtasks through a task planning agent, and distributes them in parallel to agents with specific capabilities for execution, significantly reducing manual intervention and improving overall task processing efficiency. The system can complete the entire process of data extraction, analysis, and charting without requiring further user input. Through contextual understanding and intent recognition mechanisms, the system performs semantic completion and question reconstruction on the user's original input, enabling the agent to more accurately understand the user's analysis goals, reducing task ambiguity, and effectively addressing the traditional approach's weak understanding of unstructured requirements.

[0069] By introducing collaborative agents such as task planning agents, data analysis agents, and visualization agents, this system forms a modular task management structure. Task planning logic can be dynamically adjusted based on specific needs, offering high flexibility and adaptability to diverse data analysis requirements and data source access, significantly improving the system's scalability and versatility in complex business environments. It supports analysis based on user needs, automatically matching data fields with statistical methods, and generating visualizations in a variety of display formats, including bar charts, line graphs, and pie charts. Compared to traditional visualization systems that only support fixed analysis dimensions, this approach more comprehensively covers the complex analysis requirements of user problems.

[0070] The embodiment of this specification also provides a multi-agent collaborative data visualization analysis device, such as Figure 2 As shown, the device includes: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions that can be executed by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to perform the above method.

[0071] The embodiments of this specification also provide a non-volatile computer storage medium storing computer executable instructions, wherein the computer executable instructions are configured to execute the above method.

[0072] The various embodiments in this specification are described in a progressive manner. Similar portions between the various embodiments can be referenced to each other, and each embodiment focuses on the differences from the other embodiments. In particular, the device, apparatus, and non-volatile computer storage medium embodiments are generally similar to the method embodiments, so their descriptions are relatively simplified. For relevant details, refer to the descriptions of the method embodiments.

[0073] The foregoing description of this specification describes specific embodiments. 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.

[0074] The devices and media provided in the embodiments of this specification correspond one-to-one to the methods. Therefore, the devices and media also have similar beneficial technical effects to their corresponding methods. Since the beneficial technical effects of the methods have been described in detail above, the beneficial technical effects of the devices and media will not be repeated here.

[0075] Those skilled in the art will appreciate that the embodiments of this specification may be provided as methods, systems, or computer program products. Therefore, this specification may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Furthermore, this specification 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.

[0076] This specification is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of this specification. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of processes and / or boxes in the flowchart and / or block diagram, 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 flowchart and / or block diagram. 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.

[0077] 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.

[0078] 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.

[0079] In a typical configuration, a computing device includes one or more processors (CPUs), input / output interfaces, network interfaces, and memory.

[0080] 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.

[0081] Computer-readable media includes 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 computer-readable media (transitory media), such as modulated data signals and carrier waves.

[0082] 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 list of elements includes not only those elements but also other elements not explicitly listed, or elements inherent to such process, method, commodity, or apparatus. In the absence of further limitations, the phrase "comprising a..." … ” The specified elements do not exclude the existence of other identical elements in the process, method, product or equipment that includes the elements.

[0083] The foregoing description is merely one or more embodiments of this specification and is not intended to limit this specification. It will be apparent to those skilled in the art that various modifications and variations may be made to one or more embodiments of this specification. Any modifications, equivalent substitutions, or improvements made within the spirit and principles of one or more embodiments of this specification are intended to be within the scope of the claims of this specification.

Claims

1. A multi-agent collaborative data visualization analysis method, characterized in that: The method comprises: Receive a natural language analysis request input by a user, perform task decomposition on the natural language analysis request through a preset task planning agent, and generate corresponding task planning information, wherein the task planning information includes multiple subtasks and task attribute information corresponding to each subtask; Acquire real-time operating data of a preset functional agent cluster, perform agent allocation on the multiple subtasks based on the real-time operating data and the task plan information, and determine a target functional agent corresponding to each subtask, wherein the target functional agent includes any one of a data analysis agent and a visualization display agent; The data analysis agent executes the corresponding data analysis subtask to obtain the data analysis results, and the data analysis results are collaboratively verified by the agent. After the verification is passed, the agent is visually displayed to generate visual analysis data corresponding to the data analysis results.

2. The data visualization analysis method for multi-agent collaboration according to claim 1 is characterized in that: The preset task planning agent decomposes the natural language analysis request into tasks and generates corresponding task planning information, specifically including: The task planning agent calls a large language model, performs semantic completion on the natural language analysis request based on pre-acquired user historical interaction data and pre-defined industry knowledge data, and generates user intent completion information; According to the task planning agent, the user intention completion information is task-decomposed to determine a plurality of subtasks and corresponding task attribute information, wherein the task attribute information includes task type and task dependency.

3. The data visualization analysis method for multi-agent collaboration according to claim 2 is characterized in that: Decomposing the user intention completion information into tasks to determine multiple subtasks and corresponding task attribute information, specifically including: Intent recognition is performed based on the user intention completion information to determine multiple corresponding target analysis entities, and entity matching and association relationship query are performed in a pre-built system knowledge graph based on the target analysis entities to determine the association analysis dimension and at least one corresponding data source information corresponding to each target analysis entity; According to the multiple correlation analysis dimensions, determine the dimension subtask corresponding to each correlation analysis dimension, perform task function decomposition on each dimension subtask, and determine multiple subtasks corresponding to each dimension subtask; Perform logical association analysis on the multiple subtasks to determine the task dependency relationship between the multiple subtasks, and determine the task type of each subtask based on the task function of each subtask, wherein the task type includes a data analysis type and a visualization display type, and the data analysis type includes any one or more of trend analysis, distribution statistical analysis, predictive analysis, comparative analysis, and attribution analysis.

4. The data visualization analysis method for multi-agent collaboration according to claim 1 is characterized in that: Based on the real-time operation data and the task plan information, the plurality of subtasks are assigned to agents, and a target functional agent corresponding to each subtask is determined, specifically including: Classifying the plurality of subtasks according to the task type in the task attribute information to determine a data analysis subtask of a data analysis type and a visualization subtask of a visualization display type; Performing task feature extraction on the data analysis subtasks to determine a multidimensional task description feature corresponding to each of the data analysis subtasks, wherein the multidimensional task description feature includes a data analysis type and a data source type; Through the multi-dimensional task description features corresponding to each of the data analysis subtasks and the real-time operation data, agent matching is performed in the functional agent cluster to determine the analysis functional agent corresponding to each of the data analysis subtasks; The visualization function agent corresponding to each visualization subtask is determined based on the analysis function agent corresponding to each data analysis subtask and the task dependency relationship in the task attribute information.

5. The method for data visualization analysis of multi-agent collaboration according to claim 4, characterized in that: Determining the visualization function agent corresponding to each visualization subtask according to the analysis function agent corresponding to each data analysis subtask and the task dependency relationship in the task attribute information specifically includes: Determining, based on the task dependency, the upper-layer data analysis subtask on which the visualization subtask depends; The output data features of the analysis functional agent corresponding to the upper-level data analysis subtask are obtained, and the output data features are used as query features to match the functional agent cluster to determine the visualization functional agent corresponding to the visualization subtask.

6. The data visualization analysis method for multi-agent collaboration according to claim 1 is characterized in that: Performing agent-coordinated verification on the data analysis results, specifically including: Determining data lineage tag information of the data analysis result, wherein the data lineage tag information includes data source fingerprint information, execution agent fingerprint information, and target input visualization function agent fingerprint information; According to the data lineage mark information, the data analysis results are subjected to multi-level verification to determine the verification results.

7. The data visualization analysis method for multi-agent collaboration according to claim 6 is characterized in that: Performing multi-level verification on the data analysis results based on the data lineage marker information to determine the verification results, specifically including: Using the data source fingerprint information, the data source of the data analysis result is traced and verified to determine the data source connection status and the timeliness of the data snapshot version; Obtaining corresponding agent execution information through the executing agent fingerprint information, re-calculating on some data samples, and verifying the result deviation rate to perform a processing check on the data analysis results; Obtaining a chart rendering rule set through the target visualization function agent fingerprint information, verifying the compatibility of the data structure of the data analysis result with the chart rendering rule set, and performing visualization compatibility pre-verification on the data analysis result; When the data traceability verification, the processing process verification and the visualization compatibility pre-verification all meet the preset requirements, the verification result is determined to be verification passed.

8. The method for data visualization analysis of multi-agent collaboration according to claim 1, characterized in that: Generate visual analysis data corresponding to the data analysis results by visually displaying the intelligent agent, specifically including: Analyzing the data structure characteristics of the data analysis results to extract the number of dimension fields, indicator field types, and time series flags corresponding to the data analysis results; A chart type is determined according to the number of dimension fields, the indicator field type, and the time series flag, so as to generate visual analysis data corresponding to the data analysis result based on the chart type.

9. A multi-agent collaborative data visualization analysis device, characterized in that: The device comprises: at least one processor; and, a memory communicatively connected to the at least one processor; wherein, The memory stores instructions that can be executed by the at least one processor. The instructions are executed by the at least one processor to enable the at least one processor to perform the method according to any one of claims 1 to 8.

10. A non-volatile computer storage medium storing computer executable instructions, characterized in that: The computer executable instructions are configured to execute the method according to any one of claims 1 to 8.

Images