A problem-driven enterprise operation knowledge co-creation platform and method

Abstract

The application discloses a problem-driven enterprise operation knowledge co-creation platform and method, and belongs to the technical field of enterprise data intelligence and organization knowledge management. The platform comprises a data source semantic retrieval service module, an AI analysis framework generator, an AI-assisted programmable task generator, a double-track auditing engine, a plug-in report life cycle management system, a dynamic access control module based on data source permission inheritance, and a report and knowledge asset tagging organization and sharing service module. The method of the application converts enterprise technical assets into a business-understandable semantic knowledge base through the data source semantic retrieval service module, automatically disassembles the problems input by employees using AI, generates executable code and report fragments, and after being audited by business and technology double tracks, the executable code and report fragments are used as reusable plug-ins to be deposited as organization knowledge assets. The application reduces the technical threshold for non-technical personnel to use enterprise data assets, and realizes end-to-end automatic generation from natural language problems to executable analysis reports.

Description

Technical Field

[0001] This invention belongs to the field of enterprise data intelligence and organizational knowledge management technology, and in particular relates to a problem-driven enterprise operation knowledge co-creation platform and method based on artificial intelligence (AI). Background Technology

[0002] Currently, enterprise operations, maintenance, and auditing departments generally rely on customized statistical analysis platforms for decision support. However, such platforms have three major drawbacks:

[0003] First, the development cycle is long and lacks flexibility. Analytical functions need to be planned jointly by core technical personnel and business stakeholders at the beginning of the project. Analytical methods not included in the current project phase must wait for the next development cycle, resulting in serious delays in problem discovery and resolution.

[0004] Secondly, report generation is highly dependent on manual processes. Although the platform provides basic statistical tools, the specific analytical logic, indicator definitions, and visualization presentations still need to be manually written by analysts. Business personnel often lack data comprehension skills when raising questions, resulting in a lengthy and inefficient process of clarifying requirements, and sometimes even leading to abandoning the analysis due to excessive communication costs.

[0005] Third, organizational knowledge is difficult to retain. Intense competition for positions within companies discourages employees from proactively sharing their analytical experience without clear incentive mechanisms. Furthermore, high job turnover means that individual knowledge is often lost when employees leave, failing to be transformed into organizational assets.

[0006] In recent years, although AI technology has been introduced into the field of data analysis, enterprises' underlying data capabilities, such as APIs (Application Programming Interfaces), data table structures, and field semantics, remain "silent assets" for technology departments, while business departments lack secure, compliant, and accessible channels for their use. Existing business intelligence (BI) tools, such as Tableau and Power BI, only support predefined reports and cannot support dynamic analysis and exploration starting from business problems.

[0007] Therefore, there is an urgent need for a new type of technology platform that starts with problems and aims at knowledge co-creation, enabling employees within enterprises to independently complete the entire process from problem identification to report generation with the assistance of AI, and to solidify effective analytical logic into reusable, auditable, and optimizable organizational knowledge assets. Summary of the Invention

[0008] This invention aims to solve the above-mentioned technical problems and provide a problem-driven enterprise operation knowledge co-creation platform and method to achieve the following objectives: reduce the technical threshold for non-technical personnel to use enterprise data assets; realize end-to-end automated generation from natural language questions to executable analysis reports; build a dual-track knowledge accumulation mechanism of "business-led and technology-supported"; and modularize and assetize the analysis logic to support cross-scenario reuse and continuous optimization.

[0009] The problem-driven enterprise operation knowledge co-creation platform provided by this invention includes seven functional modules: a data source semantic retrieval service module, an AI analysis framework generator, an AI-assisted programmable task generator, a dual-track audit engine, a plug-in report lifecycle management system, a dynamic access control module based on data source permission inheritance, and a report and knowledge asset tagging organization and sharing service module.

[0010] The data source semantic retrieval service module constructs a semantic knowledge base for the enterprise's data sources. This semantic knowledge base registers the enterprise's internal API interface configuration information and metadata of data sources in the data governance system, and constructs a vector index library to support semantic similarity retrieval. The data source semantic retrieval service module receives natural language questions input by users and returns a list of matching API interfaces and a set of recommended fields.

[0011] The AI ​​analysis framework generator automatically breaks down user-input natural language questions into a structured, multi-dimensional analysis framework. This framework includes multiple analysis dimensions, allowing users to modify these dimensions and simultaneously generate the overall report structure and a list of data processing tasks based on individual metrics. Users can customize data processing plugins through the AI ​​analysis framework generator, defining structured prompts for each task. When a user clicks on a data processing task, the AI-assisted programmable task generator generates and tests executable code. After verification, the executable code is submitted to the data processing plugin. The AI ​​analysis framework generator provides customization of data processing result display templates and report editing and previewing. It automatically applies user permission context, returning null values ​​or placeholders for analysis dimension fields, corresponding executable code, and data processing results that the user does not have permission to view. The AI ​​analysis framework generator submits the report fragments generated after executing the data processing tasks to a dual-track review engine.

[0012] The dual-track review engine executes a dual business and technical review process. The business review process involves the business department evaluating the business value, applicable scenarios, and access permissions of the report fragments, while the technical review process involves the technical department reviewing the security, performance, and compliance of the code. Only after both review processes are passed is the report fragment placed in the release queue; otherwise, it is returned to the AI ​​analysis framework generator for further editing.

[0013] The plug-in-based report lifecycle management system manages report fragments as reusable assets, providing lifecycle management services. Report fragments are categorized into three types: reports, report segments, and visualization components, stored in a non-relational database. Each report fragment is bound to an application scenario tag and associated with a codebase project, version number, test logs, audit records, and permission settings. The lifecycle management service is responsible for the periodic execution of data processing plug-ins or visualization component display templates. After a data processing plug-in is first invoked and performs calculations, the output result, along with the invocation parameters, is stored in the non-relational database. Subsequent requests with the same invocation parameters preferentially retrieve the stored results from this non-relational database.

[0014] The dynamic access control module based on data source permission inheritance sets up a permission inheritance mechanism, automatically inherits data interface permissions, follows the principle of tightening permissions, and supports dynamic authorization of reports after review.

[0015] The report and knowledge asset tagging organization and sharing service module structures and binds multi-dimensional tags to the generated report fragments and publishes them on demand.

[0016] Accordingly, based on the aforementioned knowledge co-creation platform, this invention provides a problem-driven enterprise operation knowledge co-creation method, comprising the following steps:

[0017] Step 1: Receive natural language questions input by employees within the enterprise;

[0018] Step 2: The AI ​​analysis framework generator automatically decomposes the problem into a multi-dimensional analysis framework based on the enterprise's pre-built data source semantic knowledge base. Each analysis dimension in the framework includes a task description, measurement indicators, and required data interfaces.

[0019] Step 3: Receive employee modifications to the analysis dimensions in the multidimensional analysis framework to form a customized analysis framework;

[0020] Step 4: The AI ​​analysis framework generator generates data processing tasks and prompts for each indicator. The AI-assisted programmable task generator generates data processing code based on the prompts, executes the data processing code in the test environment, and returns the results for employees to verify.

[0021] Step 5: Employees define the data display method and text description to form a report segment. If the data is not accessible to them, a placeholder image will be displayed during the preview.

[0022] Step 6: The report snippet is submitted to the dual-track review engine, where a dual business and technical review process is executed; employees add tags to the report snippet, and the business department assesses whether to expand the report's visibility and applicable scenarios; the tags include organizational tags, cycle tags, scenario tags, functional tags, and industry tags;

[0023] Step 7: Incorporate the approved tagged report snippets and data processing plugins into the organization's knowledge base to support subsequent recommendations and reuse;

[0024] Step 8: When a new problem is broken down and a historically similar analysis dimension is identified, the system automatically recommends data processing plugins for reuse.

[0025] Compared with the prior art, the advantages and positive effects of the present invention are as follows:

[0026] (1) The problem-driven enterprise operation knowledge co-creation platform and method provided by this invention supports non-technical employees within enterprises, such as operations, marketing, and product personnel, to independently initiate analysis needs, construct analysis logic, and generate visual reports. This enables employees within the enterprise to independently complete the entire process from problem identification to report generation with AI assistance. The platform and method of this invention solve the problem that report generation within enterprises is highly dependent on manual labor and that organizational knowledge is difficult to accumulate. It realizes the accumulation of effective analysis assets into reusable organizational knowledge assets through a dual-track review mechanism of business and technology.

[0027] (2) The enterprise operation knowledge co-creation platform and method of the present invention focuses on the issue of data usage permissions within the enterprise, and provides a dual business and technical review mechanism for the data processing process, which is particularly suitable for application scenarios with strict requirements for the accuracy of data indicators. Using the platform of the present invention, even if enterprise employees raise low-value questions, the platform can still output high-value, reusable report fragments and data processing plugins, and the data processing plugins are periodically triggered to update reports through lifecycle management, avoiding serious delays in problem discovery and resolution. Attached Figure Description

[0028] Figure 1 This is an overall architecture diagram of the problem-driven enterprise operation knowledge co-creation platform according to an embodiment of the present invention;

[0029] Figure 2 This is a flowchart of an implementation of the problem-driven enterprise operation knowledge co-creation method according to an embodiment of the present invention;

[0030] Figure 3 This is a schematic diagram of the state machine for dual-track review and dynamic permission control in an embodiment of the present invention;

[0031] Figure 4 This is a schematic diagram of the report fragment tagging storage and sharing mechanism according to an embodiment of the present invention. Detailed Implementation

[0032] The technical solution of the present invention will be described in detail below with reference to the accompanying drawings and embodiments.

[0033] like Figure 1As shown, the problem-driven enterprise operation knowledge co-creation platform of this invention includes seven functional modules: data source semantic retrieval service module, AI analysis framework generator, AI-assisted programmable task generator, dual-track audit engine, plug-in report lifecycle management system, dynamic access control module based on data source permission inheritance, and report and knowledge asset tagging organization and sharing service module.

[0034] The function of the data source semantic retrieval service module is to transform the data assets accumulated by the technical department into a semantic knowledge base that business personnel can understand and query, supporting automatic mapping from natural language to data interfaces. The implementation method of the data source semantic retrieval service module in this embodiment is as follows:

[0035] (11) Register the metadata of the enterprise data governance system, including data table name, field name, field type, business meaning label and field-level access permissions;

[0036] (12) API interface configuration information for accessing the API gateway, including interface address URL, request parameters, return fields and interface call permissions;

[0037] (13) A semantic understanding model is constructed based on the Transformer architecture and trained using over 10,000 samples jointly annotated by business and technical experts. Each core data interface corresponds to at least 10 different analytical application scenarios / business scenarios. A vector index library is also constructed, which supports continuous optimization through user retrieval logs and feedback to support efficient semantic similarity retrieval. The samples jointly annotated by business and technical experts refer to the extraction of metadata from API interfaces and enterprise data governance systems, standardization of metadata fields, annotation of scenario descriptions, and expert review of the annotations. The standardized fields and scenarios can be vectorized and stored in the vector index library for semantic similarity retrieval of input questions.

[0038] (14) Further, in order to enhance the model’s ability to understand business scenarios, an enterprise-level domain knowledge graph can be constructed. The knowledge graph automatically extracts entities and relationships based on internal operation documents, analysis reports and interface metadata. The knowledge graph is then distilled using a large language model to generate structured scenario-field association rules. After manual review, a scenario classification dictionary is formed to expand the business semantics of the original fields and improve the recall and accuracy of natural language queries.

[0039] (15) Optionally, when an enterprise has already deployed an Elasticsearch (ES) system, the registered data source metadata, API configuration information, and labeled sample library are injected into ES to build a dual-engine retrieval architecture:

[0040] The first engine: Based on structured text such as field names, business tags, interface names, and application scenarios, it achieves high-precision keyword matching through ES's full-text search capabilities;

[0041] The second engine, based on vector representations generated by a semantic understanding model, utilizes ES vector retrieval plugins, such as kNN or dense vectors, to achieve semantic similarity matching for natural language queries. The dual engines support result fusion and ranking, and continuously optimize the quality of the semantic vector library through user feedback and retrieval logs, thereby accelerating the rapid deployment and iteration of the system on the existing technology stack.

[0042] (16) Supports context-aware semantic extensions, such as automatically recommending related fields such as "login_freq" (login frequency) and "interaction_count" (interaction count) when "user activity level" is identified.

[0043] The data source semantic retrieval service module receives natural language questions input by users, responds to natural language queries, and returns a list of one or more matching API interfaces and their recommended set of fields.

[0044] The AI ​​analysis framework generator's function is to automatically break down user-inputted fuzzy business problems into structured multidimensional analysis frameworks through human-computer collaborative interaction, and generate corresponding executable data processing tasks and visualization templates. The AI ​​analysis framework generator's functionality is implemented through steps 21-24 below.

[0045] (21) Generate the initial analysis framework, including: using a fine-tuned Large Language Model (LLM), receiving natural language questions as input, and outputting a list of analysis dimensions in JSON format; each dimension includes: dimension name, dimension description, sub-dimension description, key analysis indicators, analysis task description, indicator calculation logic, recommended data source and recommended display method; generate two parts simultaneously: (1) the overall structure framework of the report; (2) a list of data processing sub-tasks based on a single indicator.

[0046] (22) Customized data processing plugins include: for each data processing subtask, business personnel define structured prompts with AI assistance, specifying: input fields, processing logic (such as aggregation, filtering, association), and output format; the system's AI-assisted programmable task generator automatically generates data processing code that conforms to syntax specifications based on the prompts, such as Python / Pandas scripts; automatically applies the current user's permission context, returning null values ​​or placeholders for analysis dimension fields that the user does not have permission to view; to improve reusability, the plugin supports multiple output field designs; after approval, the data processing plugin is registered to the data source semantic retrieval service module, becoming a standardized data service that can be called by other analysis tasks.

[0047] (23) Customized display templates, including: business personnel can define multiple visualization display templates for the same data processing result, such as line charts, heat maps, and text summaries; with AI assistance, by filling in structured prompts, including chart types, threshold rules, and text description templates, display component code that conforms to front-end rendering specifications is generated; the display templates are decoupled from the data processing plugins and support independent updates and combined use; when the generated code is executed during testing, the current user's permission context is automatically applied, and the code and data processing results of the data processing tasks corresponding to fields that the user does not have permission to view are returned as null values ​​or placeholders.

[0048] (24) Visual report editing and preview, including: providing a graphical interface that allows business personnel to add or delete dimensions, adjust the order of dimensions, rearrange the order of core indicator paragraphs, and preview the report effect in real time; when the generated code is executed in the test, the current user's permission context is automatically applied, and fields without permissions return null values ​​or placeholders.

[0049] The AI-assisted programmable task generator functions to transform structured data processing tasks into executable code, providing an integrated interactive interface for AI programming startup, code preview, execution testing, and progress feedback, and supporting business personnel in verifying and confirming the generated logic. The AI-assisted programmable task generator function is implemented as follows: After business personnel select a data processing task from the analysis framework, they can click "Generate Code" to enter this module; it calls a large-scale domestic open-source code model, finely tuned for enterprise operations, such as Qwen or DeepSeek-Coder, to generate backend data processing scripts (Python / Pandas) and frontend visualization component code (such as ECharts configuration or React components) that conform to executable syntax specifications; it displays the generated code content in real time, providing operation options such as "regenerate," "manually edit," and "save draft"; it supports one-click execution of the code in the test environment, with the interface synchronously displaying execution progress, resource consumption, return results, and error logs; it calls anonymized data samples in test mode through the API gateway to verify the correctness of the code output; when the generated code is executed in the test, it automatically applies the current user's permission context, returning null values ​​or placeholders for fields without permissions; after business personnel confirm that it is correct, they can submit the code as a formal data processing plugin and enter a dual-track review process.

[0050] The dual-track review engine ensures the quality and security of knowledge accumulation. This is achieved through the following steps (41-43):

[0051] (41) Business review process: Through the review design of the organizational structure, the business value, applicable scenarios and access permissions of the evaluation report segments are assessed, such as "visible only to the growth team";

[0052] (42) Technical review process: Representatives from the data governance team review the code for SQL injection risks, resource consumption, and whether it can be optimized into a new materialized view;

[0053] (43) Trigger the state machine based on the review results: If any process is rejected, return to the AI ​​analysis framework generator for further editing; if both processes are approved, enter the release queue.

[0054] The plug-in report lifecycle management system manages report fragments as reusable assets, providing lifecycle management services. These lifecycle management services are responsible for the periodic execution of data processing plug-ins or display templates. Implementation includes: report fragments are categorized by type: reports, report fragments, and visualization components; support for multi-level tags binding to application scenarios, such as "Organization / Growth Center / Retention Analysis" and "Scenario / Major Promotion Review"; storage in a high-concurrency read / write-supporting non-relational database, such as MongoDB, associated with code repository projects, version numbers, test logs, audit records, and permission settings; and support for version control, canary releases, and usage popularity statistics. Optionally, the system is configured with a result storage mechanism: when a data processing plug-in is called for the first time, calculations are performed and the output results, along with the call parameters, are stored in the non-relational database; for subsequent requests with the same call parameters—whether from the generation process of other analysis reports or rendering requests from end users viewing already generated reports—the system prioritizes reading the stored results from the non-relational database to avoid redundant calculation tasks.

[0055] The function of the dynamic access control module based on data source permission inheritance is to support full participation in problem raising and report construction while ensuring enterprise data security. This is achieved by automatically inheriting data source interface permissions, adhering to permission tightening principles, and supporting dynamic authorization after review, thus unifying "open co-creation" and "security compliance." The implementation of the dynamic access control module includes the following steps 61-64, such as... Figure 3 As shown.

[0056] (61) Permission inheritance mechanism: All data interfaces that the analysis tasks rely on are from the API interfaces registered by the semantic retrieval service module of the data source and the metadata of the source data, which include field-level and interface-level access permissions; the system automatically calculates the minimum permission set for each indicator in the report: if an indicator integrates multiple data interfaces, its permission is the intersection of the permissions of each interface, that is, the "tightening only" principle.

[0057] (62) Restricted preview during the creation stage: Any employee can initiate questions and build report frameworks, customize plugins and templates; during the preview or generation process, if a user does not have permission to access a certain data interface, the system will automatically replace it with a placeholder image in the corresponding display position and prompt "You do not have permission to view this data, which has been submitted to the organization for review".

[0058] (63) Review-driven permission expansion and knowledge evolution: After the report is submitted for dual review, the business manager can assess whether to expand the visibility scope, such as "visible to the whole company", and the technical manager confirms the compliance of data use; after the review is approved, the system binds an explicit access policy to the report fragment, covering the original interface permissions, and realizes "organizational authorization > personal permissions"; at the same time, the report fragment opens version iteration permissions: any employee with the required data access permissions can create an enhanced version based on the original issue, and the system automatically retains the version lineage and associates it with the same analysis topic, supporting the continuous accumulation and optimization of organizational knowledge.

[0059] (64) Binding of permissions and knowledge assets: The access control policies of the final knowledge plugins and reports are stored along with the tags to ensure that permission verification is automatically performed when sharing and reusing.

[0060] The report and knowledge asset tagging organization and sharing service module, namely Figure 1 The tag-based sharing service module's functions include: structuring generated reports, display snippets, and complete reports; binding multi-dimensional tags; and on-demand publishing. It supports multi-entry sharing based on organizational paths, topic interests, and time series, and provides third-party referencing and collaboration interfaces. The implementation methods of the tag-based organization and sharing service module include:

[0061] (71) During the report release phase, multi-dimensional tags are automatically assigned by business personnel or lifecycle management services, including:

[0062] Organizational tags, such as "Growth Center" and "User Operations Group";

[0063] Periodic tags, such as "monthly report" and "weekly report";

[0064] Context-based tags, such as: "Major Sales Review" and "Anomaly Investigation";

[0065] Functional tags, such as "Retention Analysis" and "Cost Monitoring";

[0066] Industry / sector tags, such as "energy storage" and "battery swapping".

[0067] (72) Based on the above tags, the working groups within the department uniformly formulate hierarchical display directory paths, such as / Growth Center / Big Promotion Review / Weekly Report. The system automatically organizes the report content and supports permission inheritance and dynamic aggregation.

[0068] (73) The report can be automatically generated periodically by the lifecycle management service module (e.g., 3 am daily) or triggered on demand (e.g., by clicking “Generate the latest version”).

[0069] (74) The published report can be accessed through the following paths:

[0070] Organizational path navigation: Browse level by level according to the enterprise structure;

[0071] Tag subscription path: Automatically aggregate related content based on tags that employees follow, such as "#user churn";

[0072] Timeline path: Displays the latest reports in reverse chronological order of their generation time.

[0073] (75) Provide standardized APIs and embedded components, such as iframes or Web Components, to allow third-party systems, such as DingTalk, Lark, and internal Wikis, to reference report snippets and overlay collaborative functions such as comments, @ reminders, and voting to form a closed loop of discussion around the data.

[0074] (76) When a report is displayed in an organizational path or tag subscription stream, the system verifies the user’s access rights to the report and report fragments in real time, and unauthorized content will not be rendered.

[0075] Figure 1 The system demonstrates how a business employee initiates a natural language question, which is then broken down into a multi-dimensional analysis task list by an AI analysis framework generator (enabled by a domestically developed large-scale model). Each task is then converted into executable code by an AI-assisted programmable task generator (also enabled by a domestically developed large-scale model). The code and generated report snippets are submitted to a dual-track review engine for business and technical review. Upon approval, the report is managed centrally by a plug-in report lifecycle management system, which supports automatic execution of data processing plug-ins to generate the latest report at preset intervals or on demand. Access permissions for reports and report snippets are constrained by a dynamic access control module based on API gateway metadata and a permission policy library, adhering to a principle of tightening permissions. Finally, a permission-tagged shared service module (enabled by a domestically developed large-scale model) enables content aggregation and display based on multi-dimensional tags such as organization, scenario, and cycle, and allows for reference by third-party systems. The system also supports an optional result caching mechanism to improve the reuse efficiency of high-frequency queries.

[0076] like Figure 2 As shown, the problem-driven enterprise operation knowledge co-creation method of this invention includes steps S100-S800. This embodiment analyzes the "reasons for the recent decline in content publishing volume".

[0077] S100: Employee input of natural language problem. In this example, an operator inputs on the platform: "Why has the number of content posts decreased in the past week?"

[0078] S200: The AI ​​analysis framework generator is based on the semantic knowledge base of enterprise data sources and automatically breaks down problems into multi-dimensional analysis dimensions.

[0079] The AI ​​analytics framework generator calls the semantic understanding model and outputs the following dimensions, described in JSON format:

[0080]

[0081] In this example, based on the current input question, the LLM outputs two dimensions: "time trend" and "user segmentation." The task description for each dimension is in the `task` field, the `metric` field records the metrics that need to be measured, and the required data interfaces are recorded in the `dataSource` field. `content_publish_log` represents the content publishing log, and `user_register_log` represents the user registration log.

[0082] S300: Employee tailoring framework to form a customized analysis path.

[0083] In this embodiment, the operators considered "external activity impact" to be more important, so they removed the "user segmentation" dimension and added a "system announcement association" dimension. The system announcement association dimension requires access to marketing log data, specifically the `marketing_campaign_log` interface. However, since users do not have permission to access the `marketing_campaign_log` interface, the recommended data source field for the newly added dimension is displayed as a placeholder image in the preview interface, with the message: "You do not have permission to view this data; it has been submitted for organizational review."

[0084] S400: The AI-assisted programmable task generator transforms data processing tasks of various dimensions into programmable tasks, generating and testing data processing code. The data processing code is executed in a test environment, and the results are returned for employee verification.

[0085] In this embodiment, the system generates structured prompts for each dimension; it calls the finely tuned Tongyi Qianwen code model to generate a Python / Pandas script; when executed in the test environment, the marketing_campaign_log related fields return null values, while the other dimensions are calculated normally; the operations staff confirms that the trend analysis results are correct and submits a report fragment for review.

[0086] S500: Employees define visualization styles and text descriptions to create report segments. If the report contains data that employees do not have access to, a placeholder image will be displayed during the preview, and the report can still be submitted for dual review.

[0087] In this embodiment, employees add text descriptions, select a line chart to display the time trend, and set a threshold of "decrease > 15% highlighted in red"; add a text description template: "If the number of posts is below the threshold for three consecutive days, it is recommended to check the backlog of content review queue"; and create an additional "weekly summary" text template for the same data result for mobile push.

[0088] S600: Submit report fragments to a dual-track review process: business assessment of value and permissions, and technical review of code security and performance.

[0089] After configuring basic attribute information, employees submit their applications for review. The configured information includes tag information as follows:

[0090] (1) Organizational: / Growth Center / Content Operations Group; (2) Periodic: Weekly Report 2025W48;

[0091] (3) Scenario-based: anomaly investigation; (4) Function-based: monitoring release; (5) Industry-based.

[0092] Business review: After evaluation by the Operations Director, the visibility of the report is set to "visible to the whole company", and the scene tag "#contentoperations" can be added.

[0093] Technical review will be conducted as follows:

[0094] (1) The data engineer confirmed that the code has no security risks and agreed to this release;

[0095] (2) At the same time, record the high-frequency query logic involved in the generation of this report by the data interface, such as the high-frequency query logic of content_publish_log, materialize it into a new view, and provide a data processing plugin based on daily, weekly and monthly statistical methods;

[0096] (3) After the API gateway and data source semantic retrieval service are available, the data interface will notify the owner and user of the content_publish_log interface to upgrade the data processing plugin triggered by interface optimization.

[0097] (4) After the review is approved, the system binds an explicit access policy to the report fragment, overriding the original interface permissions.

[0098] S700: Once approved, it will be included in the organization's knowledge base as a tagged, reusable plugin / report fragment, supporting subsequent recommendations and reuse.

[0099] In this step, the lifecycle management service is responsible for the periodic execution of plugins or templates; the data results, fragments, and reports are stored in a knowledge base in an unstructured database, including information such as permissions, tags, and the IDs of referenced plugins or templates; based on the report's tags, the system pushes new data messages to the tag's subscribers. Content strategy managers who follow #AnomalyInvestigation on the "Tag Subscription" page will automatically receive the report push; if they find insufficient analysis depth, they can click "Create Enhanced Version" to add the "Impact of Major Promotional Activities" dimension, forming version V2, and the system retains the version hierarchy (V1←V2); the report is embedded in the Lark knowledge base, and team members can directly @data analysts in an iframe to discuss root causes; when viewing charts, the system reads the cached report fragments from the unstructured database.

[0100] S800: When a new problem is broken down by AI and a historically similar analysis dimension is identified, the system automatically recommends data processing plugins for reuse.

[0101] When users enter keywords such as "content publishing" or "publishing behavior" in the search box or analysis interface, the system automatically recommends three types of resources based on semantic matching: (1) stored highly relevant report snippets; (2) a shortcut to "generate a complete report"; and (3) a tool to "create a new report snippet". If a user chooses to start a new report generation process, they can manually trigger the AI ​​analysis framework generator to reuse existing logic or build new dimensions. During the report viewing stage, if multiple users request the "publishing trend" chart with the same parameters, the system prioritizes reading the stored results from the unstructured database to avoid repeatedly executing the data processing plugin. When creating a new analysis report that includes existing indicators such as "comparing the average daily publishing volume of the last 7 days with the previous 7 days", the data source semantic retrieval service module returns the corresponding data processing plugin as a high-priority candidate, and the user can directly select the plugin without redefining the logic or performing calculations.

[0102] Figure 3 The report segment demonstrates the status transition logic from draft, submission, business review, technical review to release or rejection. Through dual-track review and dynamic permission control, it implements a mechanism for tightening permissions, covering organizational authorization, and version iteration.

[0103] Figure 4 It demonstrates how a report snippet can be bound to multi-dimensional tags such as organization, period, and scenario, and can be accessed through three paths: organization directory, tag subscription, and time stream. It also supports being referenced by third-party systems via API or embedded components and can be overlaid with collaborative functions such as comments and @ reminders.

[0104] Compared with traditional operation and management platforms and general BI / reporting platforms, the technical solution of this invention has the following advantages, as shown in Table 1.

[0105] Table 1. Comparison of the method of this invention with traditional operation and management platforms and general BI / reporting platforms.

[0106]

[0107] Through practical testing and verification, this invention brings the following significant technical effects in enterprise operation scenarios:

[0108] (1) The efficiency of analysis has been greatly improved: the report generation time has been shortened from monthly or weekly in the traditional model to within one to two days. Business personnel can independently complete the entire process from raising a problem to a visual report without waiting for IT scheduling. However, because there is still an audit mechanism, some process time is still required.

[0109] (2) Effective knowledge accumulation and reuse: More than 85% of effective analysis logic is solidified into reusable data processing plugins or report fragments and automatically archived through a tagged directory; while sharing reports, the original analysis framework and the tailored framework of the report can be widely exchanged; when similar analysis dimensions reappear, the system recommends reusing the plugin; when multiple reports use the same plugin, the data processing results can be reused to avoid repeated analysis and calculation.

[0110] (3) Non-technical staff are deeply involved in data analysis: More than 90% of employees in non-technical positions such as operations and marketing can independently complete basic to intermediate analysis tasks with the assistance of AI, significantly reducing the reliance on professional analysts.

[0111] (4) Release technical resources and focus on high-value work: Data engineers are freed from 60% of requirement communication and simple script development, and instead devote themselves to high-level tasks such as data architecture optimization, materialized view design and API governance.

[0112] (5) Achieve a secure and controllable knowledge co-creation mechanism: Through permission inheritance + tightening principle + audit-driven authorization, ensure that any employee can raise questions and build reports, while sensitive data is always under control; unauthorized fields are automatically replaced with placeholder images, which protects privacy and does not block the flow of questions.

[0113] (6) Support for continuous knowledge evolution and collaborative optimization: The published report fragments are open to multiple users for version iteration. Senior employees can expand the analysis dimensions based on the original problem. The system automatically retains the version lineage, forming an organizational learning closed loop of "problem → insight → optimization".

[0114] (7) Improve the efficiency of knowledge discovery and sharing: (1) Based on the three entry points of organizational path, tag subscription and time stream, employees can efficiently discover relevant reports; by embedding components, reports can be introduced into collaboration platforms such as DingTalk and Lark, supporting interactions such as comments and @ reminders, and promoting data-driven decision-making consensus. (2) At the same time, based on the permission design of report fragments, it avoids 1% of restricted content leading to 100% knowledge sharing.

[0115] (8) System performance and resource utilization optimization: Through the optional result storage mechanism, the stored results are directly returned for query requests with the same parameters, reducing the overhead of repeated calculations by more than 40% and reducing the database load.

[0116] (9) Compatible with existing enterprise technology stack, accelerating implementation: In an environment where Elasticsearch has been deployed, a dual-engine retrieval architecture of keywords + semantic vectors can be quickly built without the need for a complete overhaul, achieving a smooth upgrade.

[0117] Except for the technical features described in the specification, all other technologies are known to those skilled in the art. Descriptions of well-known components and technologies are omitted in this invention to avoid redundancy and unnecessary limitation. The embodiments described above do not represent all embodiments consistent with this application. Various modifications or variations that can be made by those skilled in the art without creative effort based on the technical solutions of this invention are still within the protection scope of this invention.

Claims

1. A problem-driven knowledge co-creation platform for enterprise operations, characterized in that, The platform includes seven functional modules: a data source semantic retrieval service module, an AI analysis framework generator, an AI-assisted programmable task generator, a dual-track audit engine, a plug-in report lifecycle management system, a dynamic access control module based on data source permission inheritance, and a report and knowledge asset tagging organization and sharing service module. The data source semantic retrieval service module constructs a data source semantic knowledge base for the enterprise. This semantic knowledge base registers the enterprise's internal API interface configuration information and metadata of data sources in the data governance system, and constructs a vector index library to support semantic similarity retrieval. The data source semantic retrieval service module receives natural language questions input by the user and returns a list of matching API interfaces and a set of recommended fields. The AI ​​analysis framework generator automatically breaks down the natural language questions input by the user into a structured multidimensional analysis framework. This framework contains multiple analysis dimensions, allowing users to modify the analysis dimensions within the framework. It also generates the overall structure of the report and a list of data processing tasks based on a single indicator. Users can customize data processing plugins through the AI ​​analysis framework generator and define structured prompts for each data processing task. When a user clicks on a data processing task, the AI-assisted programmable task generator generates executable code for the task and tests the code. After confirming that it is correct, the executable code is submitted to the data processing plugin. The AI ​​analysis framework generator provides customization of the display template for the data processing results, as well as the editing and previewing of the report. The AI ​​analysis framework generator automatically applies the user's permission context, returning null values ​​or placeholders for analysis dimension fields that the user does not have permission to view, as well as the corresponding executable code and data processing results. The AI ​​analysis framework generator submits the report fragment generated after executing the data processing task to the dual-track review engine. The dual-track review engine executes a dual business and technical review process. The business review process involves the business department evaluating the business value, applicable scenarios, and access permissions of the report fragments, while the technical review process involves the technical department reviewing the security, performance, and compliance of the code. After both review processes are passed, the report fragments are placed in the release queue; otherwise, they are returned to the AI ​​analysis framework generator for further editing. The plug-in-based report lifecycle management system manages report fragments as reusable assets, providing lifecycle management services. Report fragments are categorized into three types: reports, report segments, and visualization components, stored in a non-relational database. Application scenario tags are bound to each report fragment, and they are associated with code repository projects, version numbers, test logs, audit records, and permission settings. After the data processing plugin is first invoked and performs calculations, the output result, along with the invocation parameters, is stored in the non-relational database. Subsequent requests with the same invocation parameters preferentially retrieve the stored results from the non-relational database. The lifecycle management service is responsible for the periodic execution of the data processing plugin or visualization component display template. The dynamic access control module based on data source permission inheritance sets up a permission inheritance mechanism, automatically inherits data interface permissions, follows the principle of permission tightening, and supports dynamic authorization of reports after review. The report and knowledge asset tagging organization and sharing service module performs structured organization, multi-dimensional tag binding, and on-demand publishing of the generated report fragments.

2. The problem-driven enterprise operation knowledge co-creation platform according to claim 1, characterized in that, The data source semantic retrieval service module is built on the Transformer architecture to construct a semantic understanding model. It is trained using samples jointly annotated by business and technical experts. The annotated samples refer to the metadata of the API interface and data source, as well as the annotated application or business scenario description. The vector index library stores the standardized fields of the API interface and metadata and the scenario in a vectorized manner for semantic similarity retrieval of the input question.

3. The problem-driven enterprise operation knowledge co-creation platform according to claim 2, characterized in that, The aforementioned data source semantic retrieval service module constructs an enterprise-level domain knowledge graph. The knowledge graph automatically extracts entities and relationships based on internal enterprise operation documents, analysis reports, API interfaces, and interface metadata. It performs context distillation on the knowledge graph through a large language model to generate structured scenario-field association rules, which are then manually reviewed to form a scenario classification dictionary.

4. The problem-driven enterprise operation knowledge co-creation platform according to claim 2, characterized in that, The aforementioned data source semantic retrieval service module, when an enterprise has already deployed an Elasticsearch system, injects the registered API interface configuration information, data source metadata, and labeled sample library into the Elasticsearch system to build a dual-engine retrieval architecture. The first engine performs keyword matching based on full-text search of field names, business tags, interface names, and / or scenarios; The second engine performs semantic similarity matching of natural language based on semantic vector representation; The search results from the dual-engine retrieval architecture are fused and sorted before being used to respond to natural language queries.

5. The problem-driven enterprise operation knowledge co-creation platform according to claim 1, characterized in that, The AI ​​analysis framework generator allows users to modify analysis dimensions. Each analysis dimension includes a dimension name, dimension description, sub-dimension description, key analysis indicators, analysis task description, indicator calculation logic, recommended data source, and recommended display method. The AI ​​analysis framework generator also allows users to define structured prompts for each data processing task. The prompts include input fields, processing logic, and output format.

6. The problem-driven enterprise operation knowledge co-creation platform according to claim 1, characterized in that, The AI-assisted programmable task generator calls open-source code models to generate backend data processing scripts and frontend visualization component code, providing an interactive interface for code preview, execution testing, and progress feedback. When executing tests on the code, it automatically applies the current user's permission context, returning null values ​​or placeholders for fields that the user does not have permission to view. After confirming that the code is correct, it submits the code to the data processing plugin.

7. The problem-driven enterprise operation knowledge co-creation platform according to claim 1, characterized in that, The dynamic access control module based on data source permission inheritance includes: (61) Set up a permission inheritance mechanism: All data interfaces that the analysis tasks rely on are from the metadata registered by the data source semantic retrieval service module, including field-level access permissions and interface-level access permissions; the dynamic access control module automatically calculates the minimum permission set for each indicator in the report. If an indicator integrates multiple data interfaces, then the permission of the indicator is the intersection of the permissions of each data interface. (62) Support all employees to initiate the creation of questions and reports. During the report preview or generation process, if a user does not have permission to access a certain data interface, the display position of the corresponding data will be automatically replaced with a placeholder image, and the user will be prompted that they do not have permission to view the data. (63) Support dynamic authorization of reports after review: After a report fragment is submitted to the dual-track review engine, when the business department evaluates to expand the visibility of the report and the technical department confirms that the data use is compliant, an explicit access policy is bound to the report fragment to cover the original interface permissions. At the same time, version iteration permissions are opened for the report fragment, allowing users with the required data access permissions to create enhanced versions based on the original issue, automatically retaining the version lineage and associating it with the same analysis topic. (64) Bind permissions to report fragments for storage, and automatically perform permission verification when the report fragment is shared or reused.

8. The problem-driven enterprise operation knowledge co-creation platform according to claim 1, characterized in that, The implementation of the aforementioned report and knowledge asset tagging organization and sharing service module includes: (71) During the report release phase, business personnel or lifecycle management services automatically assign multi-dimensional tags to report segments, including organizational tags, cycle tags, scenario tags, functional tags, and industry or field tags. (72) Develop hierarchical display directory paths based on multidimensional tags and automatically organize report content; (73) Report segments are automatically generated periodically by the lifecycle management service, or triggered by the user on demand; (74) The access paths for the published report include the organization path, tag subscription path, and time stream path; (75) Provides standardized APIs and embedded components to support data referencing and collaborative integration with third parties; (76) When a report fragment is displayed in the organization path or tag subscription path, the user's access rights to the report fragment are verified in real time, and unauthorized content is not rendered.

9. A problem-driven method for co-creating enterprise operational knowledge, implemented based on the enterprise operational knowledge co-creation platform described in any one of claims 1 to 8, characterized in that, The method includes the following steps: Step 1: Receive natural language questions input by employees within the enterprise; Step 2: The AI ​​analysis framework generator breaks down the problem into a multi-dimensional analysis framework based on the semantic knowledge base of the data source. Each analysis dimension in the framework includes a task description, measurement indicators, and required data interfaces. Step 3: Receive feedback from employees regarding the addition or removal of analytical dimensions in the multidimensional analysis framework; Step 4: The AI ​​analysis framework generator generates data processing tasks and prompts for each indicator. The AI-assisted programmable task generator generates data processing code based on the prompts, executes the data processing code in the test environment, and returns the results for employees to verify. Step 5: Employees define the data display method and text description to form a report segment. If the data is not accessible to them, a placeholder image will be displayed during the preview. Step 6: The report fragment is submitted to the dual-track review engine to perform a dual business and technical review process; Employees add tags to report segments, and business departments assess whether to expand the visibility and applicable scenarios of the reports; the tags include organizational tags, cycle tags, scenario tags, functional tags, and industry tags; Step 7: Store the approved tagged report snippets and data processing plugins; Step 8: When a new problem is broken down and historically similar analysis dimensions are identified, the automatic data processing plugin is reused.

10. The method according to claim 9, characterized in that, In step 6, during the technical review, the technical department reviews the security of the code and the compliance of the data. If it confirms that there are no security risks in the code, it agrees to publish the report. At the same time, it records the high-frequency query logic and data processing plugins involved in the generation of this report by the data interface, and notifies the owner and user of the data interface to upgrade the data processing plugins triggered by interface optimization.

Images