Intelligent intelligence analysis method and system based on man-machine cooperation

Abstract

The invention discloses an intelligent intelligence analysis method and system based on man-machine collaboration, and aims to overcome the defects in the prior art, break the binary opposition of a full-automatic black box and a pure manual tool and construct a new intelligence analysis mode which can efficiently process mass data and can ensure that the analysis process is transparent and the result is traceable. Professional judgment and domain knowledge of intelligence analysts are deeply and seamlessly fused into an automatic data processing flow, and collaboration and enhancement of man-machine in the cognitive level are achieved instead of simple task division. The implicit knowledge generated in the analysis process is subjected to dominant and structured processing and is precipitated into reusable and evolvable mechanism-level knowledge assets, and'knowledge islands' are broken through.

Description

Technical Field

[0001] This invention belongs to the field of computer application technology, specifically relating to an intelligent intelligence analysis method and system based on human-computer collaboration. Background Technology

[0002] Currently, the field of intelligence analysis mainly relies on two types of tools: Fully automated "black box" analysis systems: These systems typically employ complex algorithms such as deep learning to process input data end-to-end and directly output analysis results (such as threat scoring and event prediction). Their advantages include high processing speed and the ability to handle massive amounts of data. However, their core drawback lies in their "black box" nature: the analysis process is opaque, the decision-making basis is difficult to interpret, and analysts cannot understand how the results were arrived at, nor can they trace or verify them. This "uninterpretability" severely undermines the credibility of the analysis results, and in intelligence analysis scenarios requiring a high degree of responsibility and rigor, analysts often hesitate to readily adopt their conclusions.

[0003] Manual analysis tools: These tools are represented by traditional databases, search engines, and document editors. Analysts need to manually collect, filter, read, label, and organize massive amounts of data, relying on personal experience and intellect for correlation analysis and reasoning. Their advantage is that the analysis process is completely controllable, allowing analysts to fully utilize their professional judgment. However, their disadvantages are also significant: extremely low efficiency, unable to adapt to the exponential growth of data volume in the modern information environment; the analysis process heavily relies on the analyst's individual abilities, making standardization and scalability difficult; and the analysis results are mostly unstructured reports, where the knowledge and insights are difficult to systematically preserve and reuse, forming "knowledge silos."

[0004] In summary, existing technologies suffer from the following three core pain points: Question 1: The conflict between efficiency and trust. Automated tools solve the efficiency problem but sacrifice trust, while manual tools guarantee trust but sacrifice efficiency. There is a seemingly insurmountable gap between the two.

[0005] Question 2: Human-machine disconnect hinders the integration of intelligence. In automated processes, analysts' professional and domain knowledge cannot be effectively integrated. Analysts can only intervene at the start (setting parameters) and end (viewing results) of the process, and cannot exert influence on key intermediate processing links (such as entity recognition and relationship judgment). This results in machine processing results often being disconnected from the analyst's actual cognition and judgment, generating a lot of "noise".

[0006] Question 3: Loss of knowledge assets, forming silos. During the analysis process, the valuable insights, corrected errors, and established connections generated by analysts—such as tacit knowledge—mostly remain in their personal minds or in the final report, failing to be structurally extracted and preserved. When analysts leave or embark on new tasks, these knowledge assets are lost, unable to form institutional-level, accumulative, and reusable knowledge wealth.

[0007] Therefore, the market urgently needs a new intelligence analysis method and system that can integrate the efficiency of machines with human wisdom, make the analysis process transparent and traceable, and transform the knowledge gained from the analysis process into sustainable institutional assets. Summary of the Invention

[0008] To overcome the shortcomings of existing technologies, this invention provides an intelligent intelligence analysis method and system based on human-machine collaboration. It aims to address the deficiencies in existing technologies, break down the binary opposition between "fully automated black boxes" and "purely manual tools," and construct a new intelligence analysis model that can efficiently process massive amounts of data while ensuring transparency and traceability of the analysis process. It deeply and seamlessly integrates the professional judgment and domain knowledge of intelligence analysts into the automated data processing flow, achieving human-machine collaboration and enhancement at the cognitive level, rather than simply dividing tasks. It makes the tacit knowledge generated during the analysis process explicit and structured, and precipitates it as reusable and evolvable institutional-level knowledge assets, breaking down "knowledge silos."

[0009] The technical solution adopted by this invention to solve its technical problem is as follows: Step 1: Task-driven data acquisition; An analysis task is created through the system interface; the task has a structured configuration; the system plans the data collection strategy based on the task configuration and automatically retrieves raw data from the specified multi-source heterogeneous data sources. Step 2: Automated assembly line processing; The system automatically processes the collected raw data in the background, removing advertisements and invalid information, and merging duplicate content; it uses NLP models to perform entity recognition, relation extraction, and event extraction to generate structured preliminary intelligence data, which exists in the form of triples or event frames. Step 3: Interactive knowledge graph exploration; The structured data generated in step 2 is visualized on an interactive interface in the form of a knowledge graph. Step 4: Knowledge confirmation and accumulation in human-machine collaboration; When exploring the map, any errors or omissions that may exist in the results automatically extracted by the system can be reviewed, corrected, or supplemented directly in the map interface. Step 5: Assist in generating the report; After completing the exploration and review to form the core judgment, the most critical entities, relationships and paths are selected in the knowledge graph; the system automatically generates a preliminary analysis report framework based on the selected content and the initial analysis task objectives.

[0010] A human-machine collaborative intelligent intelligence analysis system includes: Task Management Module: Used to receive, create and manage analysis tasks, define task keywords, data sources and analysis objectives, and serve as the entry point and controller for the entire analysis process; Data acquisition module: Connects to the task management module and automatically and selectively collects raw data from multiple heterogeneous data sources, including the Internet, internal databases, and file systems, according to the task configuration. Data Processing and AI Engine Module: Connected to the data acquisition module, it is responsible for cleaning and deduplicating the raw data, and calling NLP models to perform entity recognition, relationship extraction and event extraction to generate structured preliminary intelligence data; Knowledge Graph Storage and Query Module: Connected to the Data Processing and AI Engine Module, this module stores and manages two types of knowledge graphs: one is the "preliminary knowledge graph" generated by the AI ​​engine and awaiting review; the other is the "master knowledge graph" that has been confirmed and corrected and serves as an institutional asset. This module provides efficient graph query and update interfaces. Interactive visualization module: Connected to the data processing and AI engine module and the knowledge graph storage and query module respectively, it is the system's "human-computer interaction interface"; it is responsible for presenting knowledge graph data in a graphical way, and receiving interactive operations such as clicking, dragging, filtering, and path querying, and updating the view in real time. Human-machine collaborative review module: Connected with the interactive visualization module and the knowledge graph storage and query module, it provides an interactive interface for reviewing, correcting and supplementing knowledge, and transforms the analyst's operation instructions into update operations on the main knowledge graph, thereby realizing the accumulation of knowledge; Report generation module: Connected to the interactive visualization module and the knowledge graph storage and query module, it automatically integrates information and generates a preliminary report framework containing data references and graph fragments based on the key content selected by the analyst in the graph, for the analyst's final use.

[0011] Preferably, the task management module serves as the starting point and control center of the intelligent intelligence analysis system. It provides a structured and visual interface for creating, defining, initiating, monitoring, and managing the entire lifecycle of intelligence analysis tasks. This transforms vague analytical requirements into structured task instructions that are machine-understandable, executable, and traceable, ensuring that all subsequent stages of data collection, processing, analysis, and report generation are closely aligned with clearly defined objectives.

[0012] Preferably, the data acquisition module is the data entry point of the intelligent intelligence analysis system. It receives structured instructions from the task management module and is responsible for data capture and aggregation from multiple heterogeneous data sources, including the Internet, internal databases, and file systems. It transforms the analysis objectives into an executable data acquisition plan and transmits the raw data to the downstream data processing and AI engine modules in a standardized format, providing raw materials for subsequent intelligent analysis.

[0013] Preferably, the data processing and AI engine module is the core computing hub of the intelligent intelligence analysis system. It receives raw, heterogeneous, and unstructured data from the data acquisition module, and cleans, integrates, and analyzes it through a series of automated, pipeline-style processing steps, ultimately transforming it into structured intelligence elements that can be used for knowledge graph construction and visualization exploration. The data processing and AI engine module utilizes natural language processing and machine learning technologies to translate massive amounts of text information into machine-understandable and human-verifiable knowledge, laying a solid foundation for subsequent human-computer interaction and knowledge accumulation.

[0014] Preferably, the knowledge graph storage and query module is the "memory center" and "knowledge asset repository" of the intelligent intelligence analysis system. It is responsible for efficiently and reliably storing, managing and querying two types of core knowledge graphs: the "preliminary knowledge graph" automatically generated by the AI ​​engine and awaiting review, and the "main knowledge graph" formed after being confirmed, corrected and supplemented by analysts, which serves as the core asset of the organization.

[0015] Preferably, the interactive visualization module is the "human-computer interaction window" and "cognitive enhancement interface" of the "intelligent intelligence analysis system", which transforms the abstract and complex entity and relationship data stored in the knowledge graph into an intuitive and interactive graphical language and presents it to the analyst.

[0016] Preferably, the human-machine collaborative review module is the collaborative core and knowledge accumulation entry point of the intelligent intelligence analysis system. It is integrated into the interactive visualization module and provides a set of direct, efficient, and traceable interactive controls for reviewing, correcting, supplementing, and annotating the preliminary knowledge automatically generated by the AI ​​engine.

[0017] Preferably, the report generation module is the "value outlet and result converter" of the intelligent intelligence analysis system. It receives analyst selections from the interactive visualization module and, combined with structured knowledge from the knowledge graph storage and query module, automatically generates a preliminary report framework that is structurally complete, detailed in content, well-supported by evidence, and traceable.

[0018] Preferably, the raw data includes text and images; the entity recognition results include personal names, place names, organization names, and technical terms; the intelligence elements include entities, relationships, and events; and the preliminary knowledge automatically generated by the AI ​​engine includes entities, relationships, events, and attributes.

[0019] The beneficial effects of this invention are as follows: 1. Dual Enhancement in Efficiency and Depth: Automated data collection and processing frees analysts from tedious manual labor, allowing them to focus on high-value cognitive reasoning. Simultaneously, interactive knowledge graphs help analysts quickly uncover deep, non-explicit connections, achieving a depth and breadth of analysis far exceeding traditional methods.

[0020] 2. Significantly enhanced credibility and traceability of results: Due to the deep involvement and review of key knowledge links by analysts, the accuracy and credibility of the analysis results are guaranteed. Every conclusion in the report can be traced back to the original information through data citations and graphical pathways, meeting the extremely high requirements of rigor in intelligence analysis.

[0021] 3. A paradigm shift from "cognitive substitution" to "cognitive enhancement": This invention's system is positioned as an "intelligent co-pilot," its core being to enhance the analyst's cognitive abilities, rather than replacing their judgment. The analyst remains the leader in the analysis process, while the system is responsible for providing information, demonstrating connections, and reducing cognitive load; the two complement each other.

[0022] 4. A self-evolving institutional-level knowledge asset has been constructed: Through a "knowledge verification and accumulation" mechanism, analysts' valuable personal wisdom and experience are transformed into a structured, accumulative, and reusable institutional knowledge base. This "knowledge brain" will continuously grow and evolve with the completion of each analytical task, providing higher-quality prior knowledge for future analytical tasks and laying a solid data foundation for more advanced AI applications (such as intelligent inference and trend prediction), completely breaking down "knowledge silos." Attached Figure Description

[0023] Figure 1 This is a flowchart of the method of the present invention.

[0024] Figure 2 This is a structural block diagram of the system according to an embodiment of the present invention. Detailed Implementation

[0025] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0026] This invention provides an intelligent intelligence analysis method and system based on human-machine collaboration. Its core idea is to create an "Intelligent Co-pilot," which does not replace the analyst but rather acts as a powerful assistant, collaboratively completing intelligence analysis tasks within a closed loop of human-machine interaction.

[0027] The method of this invention forms a complete human-machine collaborative closed loop, such as... Figure 1 As shown: Step S101: Task-driven data acquisition; Analysts create a specific "analysis task" through the system interface. This task is not a vague instruction, but a structured configuration. Based on the task configuration, the system intelligently plans the data collection strategy and automatically and accurately retrieves raw data from specified multi-source heterogeneous data sources. This step transforms the process from "finding a needle in a haystack" to "following a map," improving the efficiency and relevance of data collection.

[0028] Step S102: Automated production line processing; The system automatically processes the collected raw data in the background. It removes advertisements and invalid information, and merges duplicate content. It utilizes NLP models for entity recognition (identifying names of people, places, organizations, technical terms, etc.), relation extraction (identifying relationships between entities, such as "Company A invests in Company B," "C is the CEO of D"), and event extraction (identifying specific types of events, such as "merger and acquisition events," "product launch events"). This generates structured preliminary intelligence data, typically in the form of triples (entity 1 - relation - entity 2) or event frames.

[0029] Step S103: Interactive knowledge graph exploration; The system visualizes the structured data generated in step S102 as a "knowledge graph" on an interactive interface. Analysts are no longer passively reading text lists, but actively exploring the graph.

[0030] Step S104: Knowledge confirmation and accumulation in human-machine collaboration; This is the core aspect of the invention. When analysts explore the graph, they may find that the results automatically extracted by the system may contain errors (such as misidentification or incorrect relationships) or omissions (such as undiscovered important associations). At this time, analysts can directly review, correct, or supplement the graph on the graph interface.

[0031] Step S105: Generating an auxiliary report; Once analysts have completed their exploration and review and formed core judgments, they can select the entities, relationships, and paths they deem most critical within the knowledge graph. Based on the selected content and the initial analysis task objectives, the system automatically generates a preliminary analysis report framework.

[0032] A human-machine collaborative intelligent intelligence analysis system, such as Figure 2 As shown, it includes the following interconnected modules: 1. Task Management Module: Used to receive, create and manage analyst analysis tasks, define task keywords, data sources and analysis objectives, and serve as the entry point and controller for the entire analysis process.

[0033] 2. Data Acquisition Module: Connected to the task management module, it automatically and selectively collects raw data from multiple heterogeneous data sources such as the Internet, internal databases, and file systems, based on task configuration.

[0034] 3. Data Processing and AI Engine Module: Connected to the data acquisition module, this is the system's "computing brain." It is responsible for cleaning and deduplicating the raw data, and calling NLP models to perform entity recognition, relationship extraction, and event extraction, generating structured preliminary intelligence data.

[0035] 4. Knowledge Graph Storage and Query Module: Connected to the data processing and AI engine module, this module stores and manages two types of knowledge graphs: one is the "preliminary knowledge graph" generated by the AI ​​engine and awaiting review; the other is the "master knowledge graph," which serves as an institutional asset after confirmation and correction by analysts. This module provides efficient graph query and update interfaces.

[0036] 5. Interactive Visualization Module: Connected to both the data processing and AI engine module and the knowledge graph storage and query module, this module serves as the system's "human-computer interface." It is responsible for presenting knowledge graph data graphically and intuitively, receiving interactive operations from analysts such as clicking, dragging, filtering, and path queries, and updating the view in real time.

[0037] 6. Human-Machine Collaborative Review Module: Connected to the interactive visualization module and the knowledge graph storage and query module, this is the "collaborative core" of the invention. It provides an interactive interface for reviewing, correcting, and supplementing knowledge, and transforms the analyst's operation commands into update operations on the main knowledge graph, thereby achieving knowledge accumulation.

[0038] 7. Report Generation Assistance Module: Connected to the interactive visualization module and the knowledge graph storage and query module, this module automatically integrates information and generates a preliminary report framework containing data references and graph fragments based on the key content selected by the analyst in the graph, for the analyst's final use.

[0039] The detailed design of each module is as follows: 1. Detailed design of the task management module; 1.1. Module Overview; The task management module is the starting point and control center of the "human-machine collaborative intelligent intelligence analysis system" described in this invention. It provides analysts with a structured and visual interface to create, define, initiate, monitor, and manage the entire lifecycle of intelligence analysis tasks. Its core value lies in transforming vague analytical needs into structured task instructions that are understandable, executable, and traceable by machines, ensuring that all subsequent stages, including data collection, processing, analysis, and report generation, are closely aligned with clear objectives, thereby achieving precise "task-driven" analysis.

[0040] 1.2. Functional Design; 1.2.1 Task creation and configuration; The module provides a wizard-style configuration interface that guides analysts through defining a new task.

[0041] 1.2.1.1 Basic Information; Task Name: A user-defined, easily recognizable name, such as "Analysis of X Company's Technology Acquisition Intent_2025Q3".

[0042] Task Description: An optional text box for describing the task background, special requirements, or analytical perspective in more detail.

[0043] Responsible Person / Collaborator: Specifies the main person in charge of the task and the team members who can collaborate to view / edit it, supporting access control (read-only / edit).

[0044] Priority: Sets the priority of a task (e.g., high, medium, low), which is used for task scheduling and resource allocation.

[0045] 1.2.1.2 Keyword Configuration; Core Entities: The core objects of the analysis task, usually proper nouns. The system should support entity type annotation.

[0046] Event / Topic Keywords: Describes the type of event or topic area of ​​interest. Synonym expansion is supported.

[0047] Exclusion words: Used to filter out noise and ensure data relevance.

[0048] Logical combination: Supports simple logical combinations.

[0049] 1.2.1.3 Data source range; Provide a list of selectable data sources and support fine-grained configuration for each data source. Data sources should be managed in a plug-in manner for easy future expansion.

[0050] 1.2.1.4 Analysis Objective; This is the key innovation of the invention: it tells the system the analyst's ultimate intent, guiding subsequent AI processing and report generation. It provides preset target templates and allows for customization.

[0051] Preset target template: Entity profile: "Please provide a comprehensive analysis of the [core entity's] basic information, core business, key personnel, partners, and competitors." Relationship network analysis: "Please uncover all known and potential association paths between [core entity A] and [core entity B], and analyze the nature of their relationships." Event Origin and Evolution: "Please trace the cause, development process, key milestones, and current status of [an event, such as a 'data breach incident']." Trend and Intent Assessment: "Please analyze the latest developments, technological layout, and future strategic intentions of [the core entity] in [a certain field, such as 'artificial intelligence']." Risk warning: "Please monitor negative information, legal proceedings, operational risks, etc. related to [core entity], and assess their potential impact." Custom Objectives: A text box is provided to allow analysts to describe more complex analytical objectives not covered by the template in natural language. The system backend can use NLP technology to perform preliminary parsing of the custom objectives and extract key instructions.

[0052] 1.2.2 Task execution and monitoring; One-click start: After configuration, the analyst clicks the "Start Task" button. The module serializes the structured task configuration information and distributes it to the data acquisition module 202 via message queue or API call, officially starting the analysis process.

[0053] Real-time status dashboard: Provides a visual task monitoring interface that displays the status of all tasks in the form of cards or lists.

[0054] 1.2.3 Task Management and Operation; Task List: Displays all tasks in a table format, supporting sorting and filtering by name, responsible person, status, creation time, etc.

[0055] Task Operation: View: Enter the task details page to view all configuration information and execution results.

[0056] Edit: For tasks that are pending or have failed, allow modification of their configuration and restart.

[0057] Pause / Resume: For tasks in progress, analysts are allowed to manually pause them to free up system resources or resume execution later.

[0058] Terminate: Forcefully stop a running task.

[0059] Copy: Based on a completed task, quickly create a new task with a similar configuration, requiring only minor modifications to parameters (such as the time range), greatly improving the efficiency of repetitive analysis.

[0060] Archive / Delete: Archive (hides but retains data) or completely delete completed or obsolete tasks.

[0061] 1.2.4 Task History and Reuse; Task Details Page: This is the page you enter after clicking on a task; it's a hub for information.

[0062] Configuration Summary: Clearly displays all the initial configurations for this task (keywords, data sources, targets, etc.).

[0063] Execution History: Displays every key milestone and status change of a task from creation to completion in a timeline format.

[0064] Results entry: Provides links to directly jump to other modules: View raw data -> output linked to the data acquisition module.

[0065] Explore the knowledge graph -> Link to the interactive visualization module and load the preliminary knowledge graph generated by the task.

[0066] View the final report -> Link to the final report generated by the report generation module.

[0067] Knowledge Accumulation and Retrospection: This is an advanced feature. In the knowledge graph, every piece of knowledge "confirmed" by an analyst should be traceable to its source—which task, which original data, and which analysis first discovered and verified it. The task details page can display "This task contributed XX new pieces of knowledge to the organization's knowledge base."

[0068] 1.2.5 Permissions and Collaboration; Administrator: Has all permissions, including user management, system configuration, viewing and operating all tasks.

[0069] Analysts can create, edit, and execute tasks they are responsible for, and view tasks that they are authorized to collaborate on.

[0070] Observers can only view authorized tasks and their results; they cannot make any modifications or operations.

[0071] Task-level authorization: When creating or editing a task, the task owner can designate other users or user groups as "collaborators" or "observers" and assign corresponding permissions.

[0072] 1.3 User Interface Design; Layout: Uses a classic left-right or top-bottom layout.

[0073] Left / Top: This area serves as navigation and the task list, displaying task cards (including name, status, responsible person, and progress).

[0074] Right side / main area: This is the content display area, showing task creation wizards, task monitoring dashboards, or task details pages based on user actions.

[0075] Task creation wizard: It adopts a multi-step (Step-by-Step) form design, with each step corresponding to a configuration item (basic information -> keywords -> data source -> analysis target). Each step has clear guidance and examples, reducing the cognitive burden on users.

[0076] Monitoring Dashboard: Uses a card-based layout, with one card for each task. Different colors indicate the status on the cards, a progress bar displays the progress, and quick action buttons (such as pause and view) are provided.

[0077] Interactive feedback: All operations (such as clicking, saving, and launching) should have immediate visual feedback (such as loading animations, success toasts, and error warning pop-ups) so that users are clearly aware of the results of their operations.

[0078] 2. Detailed design of the data acquisition module; 2.1. Module Overview; The data acquisition module is the data entry point for the "intelligent intelligence analysis system based on human-machine collaboration" described in this invention. It receives structured instructions from the task management module and is responsible for targeted and precise data capture and aggregation from multiple heterogeneous data sources such as the Internet, internal databases, and file systems. Its core value lies in transforming the "analysis goals" defined by analysts into an executable data acquisition plan and transmitting the raw data to the downstream data processing and AI engine modules in a standardized format, providing high-quality and highly relevant "raw materials" for subsequent intelligent analysis.

[0079] 2.2. Functional Design; 2.2.1 Data Acquisition Task Parsing Submodule; This module is the "brain" of the data acquisition module, responsible for receiving task instructions from the task management module and translating them into specific, executable acquisition plans.

[0080] Input: A structured task object from the task management module, containing fields such as task_id, keywords, data_sources, analysis_goal, and time_range.

[0081] Processing logic: Task verification: Verify the completeness and validity of task instructions.

[0082] Keyword strategy generation: Transforms a simple list of keywords input by the user into complex query expressions for different data sources.

[0083] Data collection plan generation: Based on the data source type and keyword strategy, one or more specific data collection execution units are generated. Each execution unit specifies the data collection source, data collection method, query parameters, crawling depth, page range, etc.

[0084] Output: One or more standardized data collection execution plans for use by the data collection execution submodule.

[0085] 2.2.2 Data Acquisition and Execution Submodule; This module acts as the "hands and feet" of the data acquisition module, responsible for actually performing data scraping. It is a pluggable and scalable execution framework that integrates multiple data collectors.

[0086] Core components: General-purpose web crawler: Built on frameworks such as Scrapy and Selenium, used to crawl websites with no API or incomplete API functionality. Supports dynamic page rendering (JavaScript), simulated login, session persistence, and cookie management.

[0087] API Data Collector: Specifically designed for calling various public or private APIs (such as news APIs, social media APIs, and patent database APIs). It handles API authentication (API Key, OAuth), request frequency limiting, pagination, and data format parsing (typically JSON / XML).

[0088] Database connectors: Used to query data on demand from internal databases such as MySQL, PostgreSQL, MongoDB, and Elasticsearch. It executes predefined SQL or NoSQL queries.

[0089] File system monitor: Used to monitor specified local or network folders (such as shared drives, FTP servers). When new files (such as PDF, Word, Excel, TXT) are added, the content is automatically read and included in the collection process.

[0090] Processing logic: Obtain the acquisition plan from the acquisition task parsing submodule.

[0091] Schedule the appropriate collectors based on the planned data source type.

[0092] The collector performs crawling according to the plan and passes the raw, unprocessed HTML, JSON, file binary streams, and other data, along with their metadata (source URL, collection time, task ID, etc.), to the data preprocessing and caching submodule.

[0093] 2.2.3 Data Preprocessing and Caching Submodule; This module serves as the "quality control and transfer station" for the data acquisition module. It is responsible for the initial cleaning, format standardization, and temporary storage of the collected raw data, providing clean and standardized data for downstream modules.

[0094] Data cleaning: i. Content extraction: For HTML pages, algorithms such as Readability and Trafilatura are used to extract the core content and remove noise such as navigation bars, advertisements, and footers.

[0095] ii. Text encoding conversion: Convert all text to UTF-8 encoding to solve the garbled text problem.

[0096] iii. Deduplication: Based on content fingerprints (such as SimHash) or URLs, perform preliminary deduplication on the data just collected to avoid duplicate processing.

[0097] Data standardization: Data from different sources and in different formats (HTML, JSON, PDF text, etc.) is uniformly encapsulated into a standard internal data object.

[0098] 2.2.4 Data Acquisition Strategy and Scheduling Submodule; This module is the "commander" of the data acquisition module, responsible for managing the macro-schedule, resource allocation and strategy control of acquisition tasks, ensuring that the entire acquisition system operates efficiently, stably and compliantly.

[0099] Task queue management: Maintain a priority task queue. Dynamically adjust the execution order of data collection tasks based on factors such as task urgency and user permissions.

[0100] Distributed scheduling: If the system adopts a distributed deployment (multiple collection nodes), this module is responsible for distributing the collection plan to different working nodes and monitoring the node status to achieve load balancing.

[0101] Rate control: Implement granular rate limiting policies to prevent being blocked by target websites due to excessively fast requests. Policies include: Global rate limit: Sets the upper limit of the total request frequency for the entire system.

[0102] Rate limiting for domains: Set independent request intervals and concurrency for each domain (e.g., news.sina.com.cn).

[0103] Adaptive Delay: Automatically adjusts the request frequency based on the HTTP status code returned by the server (such as 429 Too Many Requests).

[0104] Proxy IP Pool Management: Integrates proxy IP services to maintain a usable proxy IP pool. Automatically rotates IPs during data collection to circumvent anti-scraping blocking. Automatically detects and removes invalid proxy IPs.

[0105] Scheduled data collection: Supports setting up periodic data collection tasks.

[0106] 2.2.5 Monitoring and Logging Submodule; This module serves as the "black box" and "dashboard" of the data acquisition module, responsible for comprehensively recording all activities and statuses during the acquisition process, and providing a visual monitoring interface for operation and maintenance personnel and analysts to view.

[0107] Detailed logging: Records detailed logs for each data collection task, including: Start / End Time: Records the lifecycle of the task.

[0108] Data collection statistics: number of pages successfully crawled, number of failures, amount of new data, and amount of data after deduplication.

[0109] Error Log: Records in detail all kinds of errors encountered during the data collection process (such as 404 Not Found, 503 Service Unavailable, connection timeout, parsing failure, etc.), and includes error stack information to facilitate troubleshooting.

[0110] Performance monitoring: Real-time monitoring of various performance indicators of the data acquisition system, such as: System load: CPU, memory, and network I / O utilization.

[0111] Queue length: The number of tasks waiting to be processed.

[0112] Data capture rate: Number of pages processed per second / minute.

[0113] Visual dashboard: Provides a web interface that visually displays information such as the progress, success rate, data source distribution, and error type statistics of data collection tasks in the form of charts (line charts, bar charts, pie charts). Users can filter and view data by task ID, time range, and other dimensions.

[0114] 2.3. Workflow; A complete data collection task process is as follows: (1) Start: The analyst creates and starts a task in the task management module.

[0115] (2) Analysis: The task instruction is sent to the acquisition task analysis submodule, which generates one or more detailed acquisition plans.

[0116] (3) Scheduling: The acquisition strategy and scheduling submodule receives the acquisition plan, puts it into the task queue, and distributes it to an idle acquisition execution submodule according to priority and resource status.

[0117] (4) Execution: The collection execution submodule calls the corresponding collector (web crawler / API collector, etc.) to retrieve raw data from the target data source according to the plan.

[0118] (5) Preprocessing: The captured raw data is sent to the data preprocessing and caching submodule in real time. This module performs content extraction, deduplication, and format standardization, and stores it in a temporary cache (such as Kafka).

[0119] (6) Delivery: The data processing and AI engine module acts as a consumer, subscribing to and obtaining standardized data from the cache to begin the subsequent intelligent analysis process.

[0120] (7) Monitoring: Throughout the process, the monitoring and logging submodule continuously records all activities, statuses, and performance data and updates them to the visualization dashboard. If an error occurs, an error log will be recorded, which may trigger an alarm.

[0121] 3. Detailed design of the data processing and AI engine module; 3.1. Module Overview; The data processing and AI engine module is the core computing hub of the human-machine collaborative intelligent intelligence analysis system described in this invention. It receives raw, heterogeneous, and unstructured data from the data acquisition module, and through a series of automated, pipeline-style processing steps, cleans, integrates, and analyzes it, ultimately transforming it into structured intelligence elements (entities, relationships, events) suitable for knowledge graph construction and visualization exploration. Its core value lies in utilizing cutting-edge natural language processing and machine learning technologies to "translate" massive amounts of textual information into machine-understandable and human-verifiable knowledge, laying a solid foundation for subsequent human-machine interaction and knowledge accumulation.

[0122] 3.2. Submodule design within the module; 3.2.1 Data Preprocessing Submodule; This is the first stop on the production line, responsible for "rough processing" of the raw data to make it meet the requirements of subsequent AI model processing.

[0123] Format unification and parsing: Unify the parsing of input data in different formats (HTML web pages, PDF documents, Word files, JSON API returned data, etc.) into plain text or structured text (such as retaining meta-information such as headings, paragraphs, and lists).

[0124] Text cleaning: Remove irrelevant content such as webpage navigation bars, advertisements, copyright notices, script code, etc. Remove special characters and garbled text, handle non-standard encoding, and clean up meaningless symbols. Standardize text, such as full-width / half-width conversion, simplified / traditional character conversion, and case unification.

[0125] Document deduplication: Based on text fingerprints (such as SimHash) or content similarity algorithms, identify and remove documents with highly repeated content to avoid redundant calculations and information redundancy.

[0126] Language detection and sentence segmentation: Automatically detect the language in the text and use appropriate tools to segment long texts into independent sentences. Sentences are the basic unit for subsequent NLP processing.

[0127] A standardized collection of cleaned and deduplicated sentences, where each sentence retains the metadata of its source document (such as source URL, publication time, etc.).

[0128] 3.2.2 Core NLP Processing Submodule; This is the core of the pipeline, responsible for "reading" and "extracting" key intelligence elements from text. This submodule is itself an internal pipeline.

[0129] Basic NLP analysis: Each sentence is segmented and tagged with parts of speech. Dependency parsing is performed to understand the grammatical structure between words in the sentence (such as subject-verb-object, attributive, adverbial, and complement), providing a grammatical basis for relation extraction.

[0130] Named Entity Recognition (NER): Identifies entities with specific meanings in text and categorizes them into predefined categories. Depending on the intelligence analysis scenario, a rich set of types can be defined, such as: people, organizations (companies, governments, universities), geographical locations, products, technologies, events, dates, amounts, etc. It employs a NER model based on deep learning models (such as BERT-BiLSTM-CRF), which has been fine-tuned on domain-specific intelligence corpora to identify domain-specific entities (such as "mergers and acquisitions" and "AI chips").

[0131] Relation extraction: Based on entity identification, this step determines whether predefined semantic relationships exist between entities and identifies the relationship type. This also needs to be defined according to the scenario, such as: employed by ([person]-[organization]), invested in ([organization]-[organization]), acquired ([organization]-[organization]), partner, located in ([organization]-[geographical location]), founder ([person]-[organization]), etc. Relation extraction models using remote supervision or Few-shot learning are employed. These models automatically generate training data using existing knowledge bases (such as master knowledge graphs), effectively addressing the problem of scarce labeled data. The model output is typically a triple: (entity 1, relation, entity 2).

[0132] Event Extraction: Identifying specific types of events described in text and extracting event trigger words and participants (arguments). Examples include merger and acquisition events, financing events, product launch events, and senior management personnel changes. A deep learning-based joint extraction model can simultaneously identify event trigger words and event arguments, outputting a structured event framework, such as: {Event Type: Merger and Acquisition Event, Acquirer: Company X, Acquired Company: Company Z, Time: October 2023, Amount: $1 billion}.

[0133] 3.2.3 Knowledge Integration and Standardization Submodule; The results extracted by AI models are often "coarse" and "fragmented". This submodule is responsible for "refining" this preliminary knowledge and resolving ambiguity and inconsistency issues.

[0134] Entity linking: Resolves the issues of synonyms and heteronyms. It links entity references extracted from the text (such as "Apple" or "Apple Inc.") to unique, normalized entity IDs in the knowledge base (such as Q42, representing Apple Inc.).

[0135] Candidate generation: Based on the entity name, retrieve multiple candidate entities from the main knowledge graph or external knowledge bases (such as Wikidata).

[0136] Disambiguation ranking: Calculate the similarity between the contextual information of the entity reference (such as surrounding words and document topic) and the descriptive information of the candidate entity, and select the most matching candidate entity.

[0137] New entity discovery: If no matching candidate entity can be found, it is considered a newly discovered entity, a new unique ID is generated for it, and it is prepared to be added to the knowledge graph.

[0138] Attribute completion: For linked entities, try to extract more attribute information from the text, such as the person's job title, the company's establishment date, the product's release date, etc., to enrich the entity's description.

[0139] Relationship and event normalization: The extracted relationships and events are mapped to a predefined standardized ontology. For example, different expressions such as "acquired", "merged", and "purchased" are standardized into an acquisition relationship.

[0140] Output: High-quality structured knowledge that has been integrated and normalized, with all entities having unique normalized IDs and relationships and events conforming to the system ontology definition.

[0141] 3.2.4 Confidence Assessment and Chain of Evidence Sub-module; This is key to achieving "explainability" and "traceability," providing a basis for decision-making in the analyst review process.

[0142] Confidence score calculation: For each extracted relation and event, a confidence score is calculated.

[0143] Evidence chain construction: Each structured knowledge (triple or event) is linked to its original source of evidence. Evidence includes not only the source sentence but also more granular location information, such as document ID -> paragraph number -> sentence number. During visualization, when an analyst clicks on a relationship on the graph, the system should highlight or directly jump to the original textual evidence supporting that relationship.

[0144] Output: The final set of knowledge elements, where each piece of knowledge is accompanied by a confidence score and one or more chains of evidence pointing to the original text.

[0145] 4. Detailed design of the knowledge graph storage and query module; 4.1. Module Overview; The knowledge graph storage and query module serves as the "memory center" and "knowledge asset repository" of the human-machine collaborative intelligent intelligence analysis system described in this invention. It is responsible for the efficient and reliable storage, management, and querying of two types of core knowledge graphs: the "preliminary knowledge graph," automatically generated by the AI ​​engine and awaiting review, and the "master knowledge graph," which serves as the organization's core asset after confirmation, correction, and supplementation by analysts. This module not only provides powerful graph data storage and complex path query capabilities but also deeply integrates into the human-machine collaborative review process, supporting versioned knowledge management, source tracing, and continuous evolution. It is the carrier for realizing the concepts of "knowledge accumulation" and "breaking down knowledge silos."

[0146] 4.2. Module Architecture; 4.2.1. Storage Engine Layer; Responsible for the physical storage and underlying indexing of graph data.

[0147] Native graph databases store data in a "node-relationship" manner, offering deep optimizations for graph traversal and path lookup, resulting in performance far superior to relational databases or non-native graph databases. At the database level, physical isolation between the "preliminary knowledge graph" and the "main knowledge graph" is achieved through different namespaces, graph tags, or independent database instances. For example, in Neo4j, the `__Preliminary__` tag can be added to nodes and relationships in the preliminary graph, while the main graph does not. Indexes are created for the unique identifiers (such as `id`) and commonly used attributes (such as `name`) of all entities to accelerate node lookups. Indexes are also created for relation types to accelerate relation traversal.

[0148] 4.2.2. Core Functional Layer; The core of this module is to encapsulate operations on the storage engine and provide APIs at the business logic level.

[0149] Core components and functional design: Graph writing interface: addEntity(entity): Receives a structured entity from the data processing and AI engine module and writes it into the initial knowledge graph.

[0150] addRelation(relation): Receives structured relations and writes them into the initial knowledge graph.

[0151] updateEntity(entity): Updates entity properties.

[0152] `promoteToMain(knowledgeId, analystId, timestamp)`: The core collaboration interface. It receives instructions from the human-machine collaborative review module, completely copies the knowledge (entity / relationship) with the specified ID from the preliminary knowledge graph, writes it into the main knowledge graph, and marks it as "promoted" or deletes it directly from its original location.

[0153] addKnowledgeToMain(knowledge): Receives new knowledge created manually by analysts and writes it directly into the main knowledge graph.

[0154] Map query interface: getEntityById(id): Retrieves entity details based on ID.

[0155] findEntityByName(name, type): Performs a fuzzy search for an entity based on its name and type.

[0156] getRelationsOfEntity(entityId, relationType, direction): Query the relationships (incoming / outgoing edges) of a given entity of a specified type.

[0157] `findPath(startId, endId, maxDepth)`: The core analysis interface. Queries all paths between two entities (e.g., the shortest path, all paths with no more than K hops). This is the foundation for implementing interactive path query functionality.

[0158] getSubGraph(centerId, hops, filters): Queries the subgraph centered on a given entity and within a specified number of hops, used for local rendering in the visualization module.

[0159] Knowledge Origin and Version Management Components: Traceability information storage: In addition to business attributes, each entity and relationship node must have traceability attributes attached.

[0160] Versioning (optional, advanced feature): Version control can be introduced for critical knowledge. When an analyst modifies an entity (such as changing its name), instead of overwriting it directly, a new version is created, and the old version is retained. This enables historical backtracking and impact analysis of knowledge. This can be achieved by adding a version number attribute inside nodes or by using specialized graph version control tools.

[0161] Schema management components: Dynamic schema support: The system should allow the dynamic addition of new entity types (such as CryptoCurrency), relation types (such as is_sanctioned), or attributes. Schema information itself can also be stored and managed as part of the graph, enabling the knowledge model to be self-describing and evolving.

[0162] Constraint Management: Define some basic constraints, such as "the 'establishment time' of a company entity must be a date type" and "an individual can have at most one 'CEO' relationship" to ensure data quality.

[0163] 4.2.3 API Service Layer; The core functional layer interfaces are encapsulated into standard, easy-to-call services for use by other modules of the system (such as interactive visualization modules and human-computer collaborative auditing modules).

[0164] RESTful APIs are simple, versatile, and suitable for most web application scenarios. They use JSON as the data exchange format.

[0165] GraphQL: Better suited for complex front-end query scenarios. The front-end can precisely specify the data fields to be returned, avoiding data redundancy or multiple requests issues found in REST APIs, making it ideal for highly interconnected data like graphs.

[0166] gRPC: If the communication between modules within the system has extremely high performance requirements, gRPC is a better choice. It is based on HTTP / 2 and Protocol Buffers and has excellent performance.

[0167] 5. Detailed design of the interactive visualization module; 5.1 Module Overview; The interactive visualization module serves as the "human-computer interaction window" and "cognitive enhancement interface" of the "intelligent intelligence analysis method and system based on human-computer collaboration" described in this invention. It transforms the abstract and complex entity and relationship data stored in the knowledge graph into an intuitive and interactive graphical language, presenting it to analysts. The core mission of this module is not merely to "display" data, but to "empower" analysts by reducing their cognitive load in discovering hidden patterns and understanding complex relationships through rich interactive methods, and providing a seamless and intuitive entry point for subsequent knowledge review and consolidation operations. It is a crucial link in realizing the transformation from "viewing data" to "gaining knowledge insight."

[0168] 5.2 Module Architecture; This module adopts a classic front-end and back-end separation architecture to ensure performance, scalability, and maintainability.

[0169] Frontend (Client): Employs a modern web frontend framework for building a component-based user interface. The graphics rendering engine uses the professional graphics library D3.js.

[0170] Core components: The graph rendering engine is responsible for drawing graphics on a Canvas or SVG based on graph data (nodes and edges). It handles the shape, color, and size of nodes, the type (straight line, curve), thickness, and arrows of edges, as well as the calculation and application of overall layout algorithms (such as force-directed layout, hierarchical layout, and circular layout).

[0171] Interaction Controller: Listens for and processes user mouse, keyboard, and touch events, such as clicks, hovers, drags, selection, zooming, and panning, and translates these events into operation commands for the graph view and data.

[0172] UI Control Panel: Provides a series of graphical user interface controls, such as search boxes, filters (sliders, checkboxes, drop-down menus), layout toggle buttons, toolbars, etc., and is the main entry point for analysts to issue complex instructions.

[0173] Information Display Panel: A sidebar or pop-up window used to display detailed information about a graph element as the user interacts with it. Clicking on a node displays all the entity's attributes, source evidence, related events, etc.

[0174] API Communication Layer: Responsible for asynchronous data communication with the backend, sending query requests to the backend (such as obtaining map data, executing path queries), receiving data returned by the backend, and driving the update of the frontend view.

[0175] Backend (server-side): Employs a high-performance web framework, Python's FastAPI.

[0176] Core functions: API Interface Service: Provides a set of RESTful APIs or GraphQL APIs for front-end calls.

[0177] a.GET / api / v1 / graph / initial?taskId={taskId}: Retrieves the initial graph data to be loaded based on the task ID (usually containing only core entities and relationships to avoid loading too much data at once on the front end).

[0178] b.POST / api / v1 / graph / node / detail: Retrieves detailed information (including all attributes and origin) for the specified node.

[0179] c.POST / api / v1 / graph / path / find: Receives two entity IDs, calls the knowledge graph storage and query module, performs a path query, and returns the path data.

[0180] d.POST / api / v1 / graph / subgraph / expand: Receives an entity ID and a relation type, and returns the N-degree neighbor subgraph of that entity, which is used to implement "drill-down" exploration of the graph.

[0181] Data aggregation and formatting: After obtaining raw data from the knowledge graph storage and query module, necessary aggregation, calculation and formatting are performed to convert it into a specific JSON format required by the front-end rendering engine, reducing the computational burden on the front end.

[0182] 5.3. Interaction Design; Explore and navigate the interaction: Zoom and pan: Supports mouse wheel zooming and drag-to-pan, allowing analysts to freely switch between macro views and micro details.

[0183] Hover Highlight: When the mouse hovers over a node or edge, the element and its directly related neighbors and connections are highlighted, while other irrelevant elements are faded out, quickly focusing on the local network.

[0184] Click to focus: Click on a node to place it in the center of the canvas and rearrange its first-degree relationship network around it to achieve "drill-down" exploration.

[0185] Selecting multiple nodes by dragging the mouse, allowing for batch operations on the selected nodes (such as viewing details or adding to a report).

[0186] Information query and filtering interaction: Global Search: Enter the entity name in the search box at the top, and the system will automatically locate and highlight the node, placing it in the center of the view.

[0187] Multi-dimensional filters: Provides a powerful filtering panel in the sidebar.

[0188] Filter by type: Select to display only nodes of the "Company" and "People" types using checkboxes.

[0189] Filter by Relationship: Select from the drop-down menu to display only relationships of the "Acquisition" and "Investment" types.

[0190] Filter by time: Use the time range slider to display only events or relationships that occur within a specific time period.

[0191] Filter by confidence level: Use the slider to display only relationships with a confidence level higher than a certain threshold.

[0192] Path Lookup: A dedicated "Path Lookup" entry is provided. Analysts can enter or select two entities by dragging and dropping, and click the "Find Path" button. The system will automatically calculate and highlight all possible related paths (such as the shortest path, all paths with no more than 5 hops), and list the path details in the information panel.

[0193] Human-machine collaborative review interaction (linked with the "Human-machine collaborative review module"): Right-click menu: Right-click on any node or edge to bring up a context menu.

[0194] For the initial knowledge generated by AI: the menu items include [View Source], [Confirm], [Correct], and [Delete].

[0195] Clicking "Confirm" will change the color of the node / edge to the "Verified" style (e.g., become solid, thicken the border) and trigger the backend API to write the knowledge into the main knowledge graph.

[0196] Click on "Correct": A small form will pop up, allowing analysts to modify relationship types, entity names, etc. After submission, the graph and backend knowledge base will be updated.

[0197] Clicking "Delete" removes the element from the current view (or marks it as "ignored") and notifies the backend.

[0198] For manually adding new knowledge: menu items include

Add Relationship

Add Entity

[0199] Click on "Add Relationship": First, select a node, then hold down Ctrl / Cmd and click on another node. Then, select the relationship type from the menu to create a new edge between the two nodes.

[0200] Operational Feedback: All review operations should provide immediate visual feedback. After clicking "Confirm," a green "√" animation should appear on the node / edge, along with a brief message "Confirmed and saved."

[0201] Report generation interaction (linked with the "Report Assistance Generation Module"): Content selection: Analysts can use Ctrl / Cmd + click or select by box to choose multiple nodes and edges that they consider key in the graph.

[0202] Access point: After selecting content, the "Add to Report" button in the toolbar or right-click menu will be activated.

[0203] Report Preview and Generation: Click "Add to Report," and the selected elements will be highlighted. Analysts can continue selecting and finally click the "Generate Report" button. The system will collect all selected elements and their associated information, call the report generation module, and generate the report framework.

[0204] 5.4 User Interface Layout Design; The UI layout includes the following areas: Top navigation bar: Includes system logo, current task name, global search box, user information, etc.

[0205] Left-side control panel: Integrates all interactive controls, such as layout switchers, entity / relationship filters, time range selectors, legends, etc. This panel can be collapsed to maximize the graph view.

[0206] Central Main Canvas: Occupies the largest area of ​​the interface and is used for rendering and interacting with the knowledge graph. The canvas background should be a dark or light grid to enhance the sense of space.

[0207] The right-hand information panel is hidden by default and slides out when the user clicks on a node, edge, or path. It displays detailed information about the selected object, including its properties, a list of source links, and a summary of related events. This panel can also be collapsed.

[0208] Bottom status bar: Displays information such as the number of nodes / edges in the current graph, the current zoom level, and system performance status.

[0209] 6. Detailed design of the human-machine collaborative review module; 6.1 Module Overview; The human-machine collaborative review module is the core of collaboration and the entry point for knowledge accumulation in the "intelligent intelligence analysis method and system based on human-machine collaboration" described in this invention. It is tightly integrated into the interactive visualization module, providing analysts with a set of direct, efficient, and traceable interactive controls for reviewing, correcting, supplementing, and annotating the preliminary knowledge (entities, relationships, events, attributes) automatically generated by the AI ​​engine. The module's design philosophy is "what you see can be reviewed, what you review can be accumulated," seamlessly transforming analysts' professional judgments into updates to the institutional-level knowledge base. This is a crucial step in achieving the leap from "personal cognition" to "organizational assets."

[0210] 6.2 Module Architecture; The human-machine collaborative review module can be logically divided into three sub-modules, which work together to complete the entire process from receiving instructions to updating the knowledge base.

[0211] 6.2.1 Review and Interaction Submodule; Directly facing the user, these are the "hands and feet" of the module, responsible for receiving review instructions from analysts.

[0212] Context menu: This is the primary interaction method. When an analyst right-clicks on any element (node, edge, label) in the knowledge graph interface, a context-sensitive menu will pop up.

[0213] Menu items for entities (nodes)

Confirm Entity

[0214]

Modify Entity

[0215]

Merge Entities

[0216]

Add Attribute

[0217] [Add Comment]: A pop-up text box allows analysts to add a free-format comment or judgment to the entity, such as "This person is a key figure in the industry and needs close attention."

[0218] [Mark as irrelevant]: Remove the entity from the current task view and record the judgment.

[0219] Menu items for relations (edges) [Confirm Relationship]: Mark the relationship as "verified".

[0220]

Modify Relationship

[0221] [Delete Relationship]: If the relationship is deemed to have been mistakenly extracted by the AI, it will be deleted.

[0222] [Add Comment]: Add a comment to the relationship, such as "This collaboration news is questionable, only one media outlet reported it, and further verification is needed."

[0223] Menu items for events (special nodes or subgraphs): similar to entities and relationships, providing operations such as [Confirm], [Correct], [Delete], and [Add Comment].

[0224] Quick Operation Bar: In the toolbar or sidebar of the graph interface, quick buttons for some frequently used review functions are provided.

[0225] Batch Confirmation: Analysts can select multiple elements within a region and then click the "Batch Confirmation" button to mark them as verified all at once.

[0226] Pending Review List: After clicking, a list will pop up in the sidebar, listing all knowledge items in the current graph with a confidence level below a certain threshold or that have never been reviewed. Analysts can process them one by one or in batches.

[0227] Intelligent Recommendation Hints: When the system is running in the background, if it detects certain knowledge that requires special attention, it will display a visual marker on the corresponding element. Analysts can click on this icon to view the reason for the recommendation and review it.

[0228] 6.2.2 Audit Logic Submodule; The "brain" of the module is responsible for processing instructions from the interactive sub-modules, performing logical judgments and business processing.

[0229] Operation command parsing: Receives commands from the interaction submodule (such as "Confirm Relationship ID: rel_12345") and parses their meaning and parameters.

[0230] Conflict Detection and Resolution: Merge Conflicts: When performing the "Merge Entities" operation, check for conflicts in the attributes and relationships of the entities to be merged (e.g., one entity's "Birthplace" is Beijing, and the other's is Shanghai). If a conflict exists, a pop-up window should prompt the analyst to decide which value to retain or create a new value.

[0231] Conflict Correction: When an analyst corrects knowledge that has already been confirmed by another analyst, the system should detect that this is an "overwrite" operation and may require a reason for the modification or trigger a higher-level approval process (in scenarios requiring strict access control).

[0232] Knowledge confidence update: The confidence level of knowledge items is dynamically updated based on the analyst's actions.

[0233] Confirmation action: Increase confidence to the highest level (e.g., 100%) and lock the status (or mark it as "verified").

[0234] Correction action: Mark the original knowledge item as "obsolete" and create a new knowledge item whose confidence level inherits the analyst's authority.

[0235] Deletion operation: Reduce the confidence level to 0 and mark it as "deleted".

[0236] Generate a knowledge change set: Package one or a series of related review operations into a "knowledge change set". This change set contains all the knowledge information to be updated and is a standardized data package passed to the knowledge accumulation submodule.

[0237] 6.2.3 Knowledge Accumulation Submodule; The "executor" of the module is responsible for persisting the "knowledge change set" processed by the review logic submodule to the knowledge graph storage and query module.

[0238] API integration with the knowledge graph storage and query module: call its provided update interfaces (such as adding nodes, updating edges, deleting attributes, etc.).

[0239] Atomicity operations: ensure that all operations in a "knowledge change set" either all succeed or all fail and roll back, avoiding the knowledge base being in an inconsistent intermediate state.

[0240] Writing to the audit log: This is the core of knowledge accumulation. Every successful update operation must be recorded in the audit log as an immutable entry. Log entries should include: Operation ID: A unique identifier.

[0241] Timestamp: accurate to milliseconds.

[0242] Operator: The analyst ID that performs the operation.

[0243] Task ID: The analysis task context in which the operation occurred.

[0244] Operation types: CREATE, UPDATE, DELETE, MERGE, CONFIRM.

[0245] Operation object: The ID of the entity / relationship / event being operated on.

[0246] Content before the change: A snapshot of the data before the operation (JSON format).

[0247] Changes: A snapshot of the data after the operation (JSON format).

[0248] Reason for change / Notes: The reasons or notes provided by the analyst.

[0249] Source evidence chain: A list of original data source IDs supporting this operation.

[0250] Trigger knowledge graph version management (optional advanced feature): For critical knowledge changes, you can trigger knowledge graph version management to create a new knowledge snapshot, allowing the system to revert to the knowledge state at any historical moment.

[0251] 6.4 Data Flow and Interface Design; Input: User interaction events from the interactive visualization module (such as right-clicks, drag-and-drop selections). Knowledge data to be reviewed from the knowledge graph storage and query module (entities, relationships, events, their confidence levels, sources, etc.).

[0252] Output: Sends UI update commands (such as changing node / edge colors, hiding elements, and popping up prompts) to the interactive visualization module. Sends knowledge update requests (API calls) to the knowledge graph storage and query module. Writes new audit records to the audit log database.

[0253] 6.3 Examples of User Interface and Interaction Design; The analyst saw a connection on the graph between "Company X" and "Company A," labeled "Acquisition." But he knew it was old news; the most recent acquisition target was "Company Z."

[0254] Problem identified: The analyst hovered his mouse over the line connecting "Company X" and "Company A" and saw the relationship details: "Acquisition," source: a 2022 news article. He realized this information was outdated and could interfere with his current analysis.

[0255] To perform the audit: He right-clicked the connection and selected "Delete Relationship" from the pop-up context menu.

[0256] Confirm Operation (Optional): The system will display a simple confirmation dialog box: "Are you sure you want to delete the relationship 'Company X - [Acquisition] -> Company A'?", and provide a text box "Please enter a reason (optional):". The analyst enters: "This acquisition information is outdated and irrelevant to the current analysis task.", and then clicks "Confirm".

[0257] Instant feedback: The lines on the graph disappear immediately. Simultaneously, the line's color may first turn gray and then fade out, providing the user with a clear visual feedback.

[0258] Background data accumulation: i. The audit logic submodule generates a "knowledge change set" that includes the "delete relationship" operation.

[0259] ii. The knowledge accumulation submodule calls the API of the knowledge graph storage module to perform deletion.

[0260] iii. At the same time, write a new record in the audit log, detailing all information about this deletion operation, including the reason entered by the analyst.

[0261] • Additional Information: The analyst then right-clicks the "Company X" node, selects "Add Relationship," and in the pop-up form, enters "Company Z" as the target entity, selects "Negotiation for Acquisition" as the relationship type, and can add "Time Range: Second Half of 2023" to the attributes. After clicking "Create," a new green line (indicating that it has been created by the analyst) appears on the graph.

[0262] 7. Detailed design of the report generation module; 7.1 Module Overview; The report generation module is the "value outlet" and "results converter" of the human-machine collaborative intelligent intelligence analysis system described in this invention. It receives analyst selections (such as key entities, relationships, paths, and annotations) from the interactive visualization module and deeply integrates structured knowledge from the knowledge graph storage and query module to automatically generate a preliminary report framework that is structurally complete, content-rich, well-supported, and traceable. The core objective of this module is not to replace analysts in writing reports, but rather to act as a "super assistant," automatically organizing the essential information selected by analysts from the knowledge graph into the report's skeleton and attaching all necessary evidence chains. This frees analysts from tedious tasks such as data organization, formatting, and citation verification, allowing them to focus on in-depth analysis, strategic recommendations, and language polishing, greatly improving the efficiency and quality of report generation.

[0263] 7.2 Module Architecture; 7.2.1 Report Content Aggregation Submodule; "Responsible for collecting and organizing all the original materials needed to generate the report."

[0264] Input: A user selection set from the interactive visualization module: structured data containing information such as the selected entity ID, relationship ID, path ID, and annotation text. A knowledge details query request from the knowledge graph storage and query module.

[0265] Processing flow: Receive and parse the user's selections to determine which knowledge elements need to be written to the report. Based on the IDs in the selection set, call the API of the knowledge graph storage and query module to retrieve detailed information for each element in batches. Entity: Retrieves all its attributes (such as full company name, establishment date, and main business) and all related relationships.

[0266] Relationship: Obtain its type, attributes (such as time of occurrence, amount), and most importantly—the source evidence chain (i.e., the URLs, titles, summaries, etc. of all original data sources that support the relationship).

[0267] Path: Obtain the sequence of all entities and relationships that constitute the path, as well as the chain of evidence for each link.

[0268] Note: Retrieves text comments added by analysts during the review process.

[0269] All the retrieved information is organized into an internally structured data object (such as JSON format), which serves as the data source for subsequent template rendering.

[0270] Output: A structured report material data package.

[0271] 7.2.2 Report Template Engine Submodule; This function is responsible for populating the source data package into a predefined report template to generate the report's text content. This is the core of achieving a structured and logical report.

[0272] Input: A report material data package from the Content Aggregation submodule. A selected report template from the template library.

[0273] Processing flow: (1) Template definition: The report template is a file defined using a specific markup language (such as Markdown, HTML, or a dedicated template syntax). (2) Data rendering: The template engine reads the template file, traverses the report material data package, replaces placeholders (such as `{{task.name}}`) with actual data, processes loops (such as `{{#each}}`) and conditional judgments (such as `{{#if}}`), and finally generates a structured report text (such as Markdown or HTML format).

[0274] Output: A structured draft of a preliminary report containing text and tags.

[0275] 7.2.3 Report Editing and Export Submodule; Responsible for providing users with a WYSIWYG editing environment and exporting the final report in a formatted manner.

[0276] Input: A preliminary report draft from the template engine submodule.

[0277] Processing flow: Rich text editor integration: Integrate a powerful rich text editor (such as TinyMCE, Quill.js, or a self-developed editor) into the system's front-end interface.

[0278] Content loading: The generated report draft (such as Markdown) is rendered into a rich text editor. At this point, the report's framework, key findings, evidence links, graph screenshots, etc., are all in place.

[0279] Analyst Interaction: Analysts perform their final work in this editor. i. Editing and polishing: Modify any automatically generated text, adjust paragraph order, and optimize language expression.

[0280] ii. Content Supplementation: In the pre-set supplementary areas such as "Conclusions and Recommendations", write your own in-depth analysis, strategic thinking, risk warnings, and other core value content.

[0281] iii. Formatting: Use the editor's toolbar to adjust the font, font size, color, and add tables, lists, etc., to make the report layout more aesthetically pleasing.

[0282] iv. Graph Screenshot Embedding: The system automatically embeds high-resolution screenshots generated from the paths or subgraphs selected by the analyst into the specified locations in the report. Analysts can adjust their size and position.

[0283] Export in multiple formats: After editing, the analyst clicks the "Export" button. This submodule offers several export options: i. Export to PDF: Call the browser's printing API or a backend conversion service (such as wkhtmltopdf, Pandoc) to convert the editor's content into a beautifully formatted PDF document, suitable for formal archiving and printing.

[0284] ii. Export to Word (.docx): Use libraries (such as docx.js) or call backend services to convert the content into Word format, making it convenient for users to perform further offline editing.

[0285] iii. Export as HTML: Saves the content as a separate HTML file, preserving all hyperlinks, suitable for sharing on web pages.

[0286] Output: Final, distributable intelligence analysis report files (PDF, Word, HTML, etc.).

[0287] Example: Example 1: Analyzing a technology company's intent to acquire technology; Suppose an analyst from a financial institution needs to analyze Company X's technology acquisition intentions over the past six months in order to assess its investment risks and future development direction.

[0288] Step 1: Task creation (S101); The analyst logs into the system and creates a new task in "Task Management Module 201".

[0289] Task Title: Analysis of Company X's Technology Acquisition Intent; Keywords: "Company X", "mergers and acquisitions", "acquisition", "investment", "technology cooperation", "artificial intelligence", "semiconductors"; Data sources include: mainstream financial news websites, technology media, official announcements from Company X, patent databases, and relevant industry research report databases.

[0290] Analysis objective: To identify potential M&A targets and key technology deployments of Company X, and to assess its strategic moves.

[0291] Step 2: Automatic processing (S102); According to its configuration, the "Data Acquisition Module 202" retrieved thousands of documents and patent information containing keywords from the past six months.

[0292] The "Data Processing and AI Engine Module 203" begins operation: It cleans all text, removing advertisements and irrelevant content. Entity recognition is performed, identifying entities such as "Company X," "University Y," "Z-chip startup," and "Mr. Wang (CTO of Company X)." Relationship extraction is performed, extracting relationships such as "Company X and Z-chip startup are in acquisition talks," "Company X obtained an image recognition patent license from University Y," and "Mr. Wang attended an AI summit." Event extraction is performed, extracting events such as "Company X announced the establishment of an AI Research Institute." All structured data is stored in the preliminary knowledge graph of the "Knowledge Graph Storage and Query Module 204."

[0293] Step 3: Interactive exploration (S103); The analyst opened "Interactive Visualization Module 205," and a knowledge graph centered on "Company X" appeared on the screen.

[0294] He noticed that "Company X" was linked to multiple "acquisition" relationships, with the line pointing to "Z Chip Startup" being particularly thick, indicating that this relationship had been mentioned in several reports. Clicking on "Z Chip Startup," he discovered that the company focused on "edge computing chips," and its founder, "Dr. Li," was a professor at "University Y." He performed a path search, entering "Company X" and "University Y," and the system immediately highlighted a path: "Company X" -> "License Acquired" -> "Image Recognition Patent" <- "Patent Inventor" <- "Dr. Li" <- "Founder" <- "Z Chip Startup." This path revealed a potentially close connection between Company X, University Y, and Company Z through talent and technology patents—a deep association that would be difficult to discover quickly by simply reading news articles.

[0295] Step 4: Knowledge Confirmation and Consolidation (S104); During the investigation, analysts discovered: The system incorrectly identified an entity named "Company A" as an acquisition target of Company X, which was later verified to be a company with the same name. He deleted the erroneous relationship by clicking "Delete" in the "Human-Machine Collaborative Audit Module 206". He found a report mentioning a potential collaboration between Company X and "Biotechnology Company B", but the system did not extract it. He manually added a "potential collaboration" relationship between "Company X" and "Biotechnology Company B". He confirmed all the related information about "Z Chip Startup" and clicked "Confirm". The system updated all his operations (deletion, addition, confirmation) to the "Main Knowledge Graph" in real time. This knowledge will become an institutional asset, and any future analytical tasks about Company X or related fields will directly reuse this verified knowledge.

[0296] Step 5: Assisted report generation (S105); The analysts circled "Company X", "Z chip startup", "University Y", "Dr. Li" in the graph, as well as the key relationships and paths between them.

[0297] He clicked the "Generate Report" button, and the "Report Generation Assistance Module 207" automatically generated a report framework, which included: Abstract: Company X is actively expanding into the fields of edge computing and artificial intelligence, and its potential acquisition target, the chip startup Z, has a deep technical and talent connection with Company X.

[0298] Key Finding 1: Company X is in acquisition talks with chip startup Z. (Links to 5 news sources are attached.) Key Finding 2: Company X established connections with University Y through patent licensing, and Dr. Li, the founder of Company Z, is from University Y, constituting a potential channel for talent and technology transfer. (Attached are screenshots of patent links and pathway maps); Conclusion and Recommendations: We recommend closely monitoring Company X's subsequent actions in the semiconductor field, as its strategic intentions are clear and may trigger changes in the industry landscape.

[0299] Based on this framework, the analyst added their own professional judgment and risk warnings, ultimately producing a high-quality, highly credible intelligence analysis report.

[0300] Through the above embodiments, the present invention perfectly combines the professional wisdom of analysts with the powerful computing capabilities of machines, not only efficiently completing the analysis task, but more importantly, precipitating the key insights generated during the analysis process into permanent knowledge assets of the organization, realizing the concept of "intelligent co-pilot" in human-machine collaboration.

Claims

1. A human-machine collaborative intelligent intelligence analysis method, characterized in that, Includes the following steps: Step 1: Task-driven data acquisition; An analysis task is created through the system interface; the task has a structured configuration; the system plans the data collection strategy based on the task configuration and automatically retrieves raw data from the specified multi-source heterogeneous data sources. Step 2: Automated assembly line processing; The system automatically processes the collected raw data in the background, removing advertisements and invalid information, and merging duplicate content; Utilize NLP models for entity recognition, relation extraction, and event extraction to generate structured preliminary intelligence data, which exists in the form of triples or event frames. Step 3: Interactive knowledge graph exploration; The structured data generated in step 2 is visualized on an interactive interface in the form of a knowledge graph. Step 4: Knowledge confirmation and accumulation in human-machine collaboration; When exploring the map, any errors or omissions that may exist in the results automatically extracted by the system can be reviewed, corrected, or supplemented directly in the map interface. Step 5: Assist in generating the report; After completing the exploration and review to form the core judgment, the most critical entities, relationships and paths are selected in the knowledge graph; the system automatically generates a preliminary analysis report framework based on the selected content and the initial analysis task objectives.

2. An analysis system employing the intelligent intelligence analysis method as described in claim 1, characterized in that, include: Task Management Module: Used to receive, create and manage analysis tasks, define task keywords, data sources and analysis objectives, and serve as the entry point and controller for the entire analysis process; Data acquisition module: Connects to the task management module and automatically and selectively collects raw data from multiple heterogeneous data sources, including the Internet, internal databases, and file systems, according to the task configuration. Data Processing and AI Engine Module: Connected to the data acquisition module, it is responsible for cleaning and deduplicating the raw data, and calling NLP models to perform entity recognition, relationship extraction and event extraction to generate structured preliminary intelligence data; Knowledge Graph Storage and Query Module: Connected to the Data Processing and AI Engine Module, this module stores and manages two types of knowledge graphs: one is the "preliminary knowledge graph" generated by the AI ​​engine and awaiting review; the other is the "master knowledge graph" that has been confirmed and corrected and serves as an institutional asset. This module provides efficient graph query and update interfaces. Interactive visualization module: Connected to the data processing and AI engine module and the knowledge graph storage and query module respectively, it is the system's "human-computer interaction interface"; it is responsible for presenting knowledge graph data in a graphical way, and receiving interactive operations such as clicking, dragging, filtering, and path querying, and updating the view in real time. Human-machine collaborative review module: Connected with the interactive visualization module and the knowledge graph storage and query module, it provides an interactive interface for reviewing, correcting and supplementing knowledge, and transforms the analyst's operation instructions into update operations on the main knowledge graph, thereby realizing the accumulation of knowledge; Report generation module: Connected to the interactive visualization module and the knowledge graph storage and query module, it automatically integrates information and generates a preliminary report framework containing data references and graph fragments based on the key content selected by the analyst in the graph, for the analyst's final use.

3. The analysis system according to claim 2, characterized in that, The task management module is the starting point and control center of the intelligent intelligence analysis system. It provides a structured and visual interface to create, define, start, monitor and manage the entire lifecycle of intelligence analysis tasks, transforming vague analysis requirements into structured task instructions that are understandable, executable and traceable by machines, ensuring that all subsequent data collection, processing, analysis and report generation are closely aligned with clear objectives.

4. The analysis system according to claim 3, characterized in that, The data acquisition module is the data entry point of the intelligent intelligence analysis system. It receives structured instructions from the task management module and is responsible for data capture and aggregation from multiple heterogeneous data sources, including the Internet, internal databases, and file systems. It transforms the analysis objectives into an executable data acquisition plan and transmits the raw data to the downstream data processing and AI engine modules in a standardized format, providing raw materials for subsequent intelligent analysis.

5. The analysis system according to claim 4, characterized in that, The data processing and AI engine module is the core computing hub of the intelligent intelligence analysis system. It receives raw, heterogeneous, and unstructured data from the data acquisition module, and cleans, integrates, and analyzes it through a series of automated, pipeline-style processing steps, ultimately transforming it into structured intelligence elements that can be used for knowledge graph construction and visualization exploration. The data processing and AI engine module uses natural language processing and machine learning technologies to translate massive amounts of text information into machine-understandable and human-verifiable knowledge, laying a solid foundation for subsequent human-computer interaction and knowledge accumulation.

6. The analysis system according to claim 5, characterized in that, The knowledge graph storage and query module is the "memory center" and "knowledge asset repository" of the intelligent intelligence analysis system. It is responsible for efficiently and reliably storing, managing and querying two types of core knowledge graphs: the "preliminary knowledge graph" automatically generated by the AI ​​engine and awaiting review, and the "master knowledge graph" formed after being confirmed, corrected and supplemented by analysts, which serves as the core asset of the organization.

7. The analysis system according to claim 6, characterized in that, The interactive visualization module is the "human-computer interaction window" and "cognitive enhancement interface" of the "intelligent intelligence analysis system". It transforms the abstract and complex entity and relationship data stored in the knowledge graph into an intuitive and interactive graphical language and presents it to the analyst.

8. The analysis system according to claim 7, characterized in that, The human-machine collaborative review module is the collaborative core and knowledge accumulation entry point of the intelligent intelligence analysis system. It is integrated into the interactive visualization module and provides a set of direct, efficient, and traceable interactive controls for reviewing, correcting, supplementing, and annotating the preliminary knowledge automatically generated by the AI ​​engine.

9. The analysis system according to claim 8, characterized in that, The report generation module is the "value outlet and result converter" of the intelligent intelligence analysis system. It receives analyst selections from the interactive visualization module and combines structured knowledge from the knowledge graph storage and query module to automatically generate a preliminary report framework that is structurally complete, detailed in content, well-supported by evidence, and traceable.

10. The analysis system according to claim 2, characterized in that, The raw data includes text and images; the entity recognition results include names, place names, organization names, and technical terms; the intelligence elements include entities, relationships, and events; and the preliminary knowledge automatically generated by the AI ​​engine includes entities, relationships, events, and attributes.

Images