Human-machine collaborative evolutionary event visual analysis method and system based on stage division
By adopting a human-machine collaborative evolution event visualization analysis method based on stage division, the problems of poor overall modeling and difficulty in tracing the source in existing technologies are solved, realizing efficient cross-stage analysis and traceable evidence, and improving the transparency and verifiability of analysis.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- SHANDONG UNIV
- Filing Date
- 2026-06-23
- Publication Date
- 2026-07-21
AI Technical Summary
Existing technologies suffer from poor overall modeling, difficulty in tracing the source, insufficient interaction, and lack of traceability of evidence in the visual analysis of evolutionary events, making it difficult to support cross-stage evolutionary analysis and in-depth human-machine collaboration.
A human-machine collaborative evolution event visualization analysis method based on stage division is adopted. Through techniques such as evidence skeleton extraction, multi-dimensional screening and scoring, and directed acyclic graph scheduling, evidence organization, traceable analysis and generation, and iterative collaborative correction are realized.
It enhances analytical transparency and traceability, supports cross-stage evolutionary analysis, improves analytical efficiency and the verifiability of results, and enables iterative analysis through human-machine collaboration.
Smart Images
Figure CN122432268A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of computer visual analytics technology, and in particular to a method and system for visual analysis of human-computer collaborative evolutionary events based on stage division. Background Technology
[0002] The statements in this section are merely background information related to the present invention and do not necessarily constitute prior art.
[0003] Evolutionary events refer to complex events that develop over a long period and a wide geographical area, commonly found in fields such as public governance and crisis response. These events are typically presented through a large number of scattered reports, with individual reports often only reflecting a partial aspect of the event. When conducting visual analysis of evolutionary events, analysts need to sift through massive amounts of reports to identify key evidence, pinpoint the stages of the event's evolution, analyze the interactions between various stakeholders, and establish a correspondence between analytical conclusions and original evidence.
[0004] With the development of large language model technology, more and more visual analytics systems are incorporating large language models to assist in completing analytical tasks such as information extraction and trend summarization. However, due to the dispersed and phased nature of evolutionary events, existing technologies still suffer from poor overall modeling, difficulty in tracing origins, and insufficient interactivity in visual analytics scenarios oriented towards evolutionary events. Summary of the Invention
[0005] To address the shortcomings of existing technologies, the purpose of this invention is to provide a human-machine collaborative evolutionary event visualization analysis method and system based on stage division, which can realize the organization of evidence for stage perception of evolutionary events, traceable analysis and generation, and iterative collaborative correction within a human-machine collaboration framework.
[0006] To achieve the above objectives, the present invention is implemented through the following technical solution: The first aspect of this invention provides a method for visual analysis of human-machine collaborative evolution events based on stage division, comprising the following steps: The document data for the task to be analyzed is used to construct a document corpus and generate analysis and query commands. The event evolution process is analyzed and divided into stages using a document corpus and analytical query commands. Based on the analysis query instructions and the phase division results, the analysis task is decomposed into sub-tasks undertaken by various dedicated analysis agents. The sub-tasks are organized into an execution plan in the form of a directed acyclic graph, and the execution of the sub-tasks is scheduled in parallel according to the dependencies of the directed acyclic graph. The outputs of each agent are integrated into an analysis report with evidence citations. The analysis report is used by analysts to make analysis statements and provide feedback on the annotation of evidence sources. The analysis report is provided to the analysts for analysis statements and feedback on the annotation of evidence sources. The annotation feedback is used as a constraint to adjust the use of evidence or task plan in subsequent analysis rounds.
[0007] Furthermore, the specific steps for conducting stage analysis and division of the event evolution process using document corpora and analytical query commands are as follows: Structured metadata is extracted from each document in the document corpus to generate an evidence skeleton; Based on the analysis query command, the evidence skeleton is screened and scored in multiple dimensions to obtain the working subset; The event evolution process is divided into stages based on the working subset and analysis query instructions, generating an ordered sequence of event stages.
[0008] Furthermore, the evidence skeleton includes: document identifier, summary, entity set, spatial reference, time attribute, and keywords; the event phase includes: phase identifier, phase name, time range, main geographical area, and dominant entity.
[0009] Furthermore, the specific steps for multi-dimensional screening and scoring of the evidence skeleton based on the analysis query command are as follows: The analysis query command is parsed into structured search constraints that include entities, time ranges, spatial ranges, and keywords; Documents are hard-filtered based on time range, and the remaining documents after filtering are weighted and scored according to three dimensions: entity, space, and keywords. The documents with the highest scores are selected as the working subset.
[0010] Furthermore, the specific steps for dividing the event evolution process into stages based on the working subset and analysis query instructions are as follows: The evidence skeletons of each document in the working subset are organized into a compressed metadata sequence ordered by time. The large language model identifies geographic focus shifts, dominant entity changes, and thematic breakpoints based on compressed metadata sequences as stage boundaries, generating an ordered sequence of several event stages. The stage boundaries are based on logical turning points in the evidence distribution rather than fixed calendar intervals.
[0011] Furthermore, the specific steps for parallel scheduling of subtasks based on the dependencies of a directed acyclic graph are as follows: Loop detection in directed acyclic graphs; Subtasks with an in-degree of zero are added to the ready queue and executed in parallel according to topological order; After a subtask is completed, the structured output is injected into the execution context of the subsequent task and the in-degree of the subsequent task is updated.
[0012] Furthermore, the specific steps for using annotation feedback as a constraint to adjust evidence use or task planning in subsequent analysis rounds are as follows: The constraints generated by the time-based selection, geographic selection, or entity filtering operations performed by analysts on the visualization interface are translated into structured prompt word templates. Once the structured keyword template is confirmed by the analysts, it will be used as a new analytical query to start a new round of analysis.
[0013] A second aspect of the present invention provides a human-machine collaborative evolution event visualization analysis system based on stage division, comprising: The input acquisition module is used to acquire the document data of the task to be analyzed, which constitutes the document corpus, as well as the analysis and query instructions; The evidence retrieval and stage awareness module is used to perform stage analysis and division of the event evolution process using document corpora and analysis query commands; The phase-aware analysis generation module is used to decompose the analysis task into sub-tasks undertaken by various dedicated analysis agents according to the analysis query instructions and phase division results, organize the sub-tasks into an execution plan in the form of a directed acyclic graph, schedule the execution of the sub-tasks in parallel according to the dependencies of the directed acyclic graph, and integrate the output of each agent into an analysis report with evidence citations. The human-machine collaborative iterative optimization module is used to provide analysis reports to analysts for analysis declarations and feedback on the annotation of evidence sources. The annotation feedback is used as a constraint to adjust the use of evidence or task plans in subsequent analysis rounds.
[0014] A third aspect of the present invention provides a computer-readable storage medium storing a computer program adapted for loading by a processor and executing steps in the human-machine collaborative evolution event visualization analysis method based on stage division as described in the first aspect of the present invention.
[0015] A fourth aspect of the present invention provides a computer device comprising: A processor, adapted to execute computer programs; A computer-readable storage medium storing a computer program, which, when executed by the processor, implements the human-machine collaborative evolution event visualization analysis method based on stage division as described in the first aspect of the present invention.
[0016] The above one or more technical solutions have the following beneficial effects: This invention discloses a human-machine collaborative evolutionary event visualization analysis method and system based on stage division. By establishing stage-aware evidence organization and explicit evidence-conclusion correspondence, it improves the transparency and traceability of analysis while supporting iterative evolutionary event analysis in human-machine collaboration.
[0017] This invention divides the event evolution process into several stages with semantic cohesion, and uses the stage as the basic unit for analyzing task decomposition and reporting organization. It overcomes the problem that existing methods treat time as a linear axis and cannot support cross-stage evolution analysis, so that the analysis results can clearly present the dominant entities, geographical focus and thematic characteristics of the event at different stages.
[0018] This invention establishes an explicit correspondence between original documents, intermediate reasoning, and final conclusions by building a traceable evidence model and an analysis report model. This allows each analysis statement to be traced back to its supporting documents, effectively improving the transparency and verifiability of the analysis results and overcoming the shortcomings of existing large language model-assisted analysis methods where evidence is not traceable.
[0019] This invention organizes subtasks using a directed acyclic graph execution plan, supporting parallel execution of independent tasks and sequential triggering of dependent tasks, thus improving overall analysis efficiency while ensuring the integrity of the analysis. At the same time, it employs multiple dedicated analysis agents to work together, producing complementary analysis content from multiple dimensions such as macro-level situation, entity trajectory, and multi-subject relationships.
[0020] This invention provides a three-stage human-machine collaboration mechanism—pre-execution plan review, post-execution evidence annotation, and replanning based on visual interaction—enabling analysts to intervene in the system's analytical intent and evidence usage with fine granularity, achieving true collaborative reasoning rather than a single command-based invocation.
[0021] Advantages of additional aspects of the invention will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of the invention. Attached Figure Description
[0022] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.
[0023] Figure 1 This is a flowchart of the human-machine collaborative evolution event visualization analysis method based on stage division in Embodiment 1 of the present invention; Figure 2 This is a diagram showing the structural relationship between the source tracing evidence model and the analysis report model in Embodiment 1 of the present invention. Detailed Implementation
[0024] It should be noted that the following detailed descriptions are exemplary and intended to provide further illustration of the invention. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.
[0025] It should be noted that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the scope of exemplary embodiments according to the invention. As used herein, unless the context clearly indicates otherwise, the singular form is also intended to include the plural form. Furthermore, it should be understood that when the terms "comprising" and / or "including" are used in this specification, they indicate the presence of features, steps, operations, devices, components, and / or combinations thereof. The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of this application.
[0026] For the visualization and analysis of evolutionary events, analysts need to sift through hundreds or even thousands of reports published across different regions, media, and time spans (usually several years) to trace the phased evolution of the event, identify the roles and relationships of various organizations and industry players at different stages, and provide verifiable analytical evidence for relevant decision-making. In such scenarios, if analysts manually read and cross-reference each report, it is not only extremely time-consuming but also difficult to guarantee the consistency of phase divisions and the completeness of evidence for the conclusions. If they directly use general-purpose large language models to generate analysis reports, the lack of explicit correspondence between the output results and the original reports will fail to meet the stringent verification requirements of professional settings such as research institutions.
[0027] Current evolutionary event analysis schemes based on large language models still have the following shortcomings: First, the content generated by large language models lacks transparency, making it difficult to trace the output results back to specific evidence sources, and analysts cannot effectively verify their reliability; second, existing systems mostly treat time as a linear axis, lacking the ability to model the stage-specific structure of events, making it difficult to support cross-stage evolutionary analysis; third, existing systems mostly adopt a single-round interaction mode, making it difficult for analysts to review and modify analysis plans, resulting in insufficient depth of human-computer collaboration; fourth, the lack of explicit correlation mechanisms between evidence, analysis process, and analysis conclusions leads to poor verifiability and reproducibility of analysis results.
[0028] To overcome the aforementioned shortcomings, this invention discloses a human-machine collaborative evolutionary event visualization analysis method and system based on stage division. This method enables the organization of evidence for stage-based evolutionary events, traceable analysis and generation, and iterative collaborative correction within a human-machine collaborative framework. Specific implementation details are provided in the following embodiments.
[0029] Example 1: Embodiment 1 of the present invention provides a method for visual analysis of human-machine collaborative evolution events based on stage division, such as... Figure 1 As shown, it includes the following steps: S1: Obtain the document data of the task to be analyzed to form a document corpus and analysis query instructions.
[0030] In one specific implementation, a document corpus This is a collection of closed-domain documents gathered by analysts around the theme of a specific evolutionary event. Represents a document. This represents the i-th document. , This refers to the number of documents. Analysis queries are analytical query commands submitted by analysts through an interactive interface in natural language. They describe the analysis objectives, objects of interest, and spatiotemporal constraints, and can be completely replaced.
[0031] S2: Use document corpora and analysis query commands to perform stage analysis and division of the event evolution process.
[0032] S2.1: Extract structured metadata from each document in the document corpus to generate an evidence skeleton.
[0033] In one specific implementation, the evidence skeleton includes: document identifier, summary, entity set, spatial reference, time attribute, and keywords.
[0034] Specifically, this embodiment constructs a source tracing evidence model and an analysis report model, the structural relationship of which is as follows: Figure 2 As shown. The source tracing evidence model consists of the evidence skeleton s(n) and the event stages. The framework is used to structure and organize the original documents and establish evidence-conclusion relationships on them. For each document n in the corpus, a lightweight language model is used to extract the evidence skeleton. Where id is the unique identifier for the document, sum is a concise summary, ent is a set of entities (including regions, organizations, companies, people, ships, etc.), loc is a set of spatial references with optional latitude and longitude, time is a time attribute, and kw is a set of topic keywords. Multiple documents are fed into the model in batches in parallel to improve processing throughput.
[0035] To reduce the issue of inconsistent naming of the same entity in different documents during cross-batch extraction (e.g., "United States", "US", "United States"), a global entity normalization is performed after the skeleton extraction of each batch is completed: the entity mentions extracted from all batches are summarized and fed into a lightweight large model to complete synonym merging in one go, and the entity dictionary consisting of a list of canonical names, types and aliases is output. Subsequently, the entity names in each skeleton are rewritten uniformly based on this dictionary.
[0036] S2.2: Based on the analysis query command, the evidence skeleton is screened and scored in multiple dimensions to obtain the working subset.
[0037] S2.2.1: Parse the analysis query command into structured search constraints that include entities, time ranges, spatial ranges, and keywords.
[0038] S2.2.2: Perform hard filtering on documents based on time range, and then perform weighted matching and scoring on the remaining documents after filtering based on three dimensions: entity, space, and keywords. Select the documents with the highest scores as the working subset.
[0039] In one specific implementation, the analysis query command is parsed to obtain the structured retrieval constraint Q=(ents, locs, kws, ts, te). Here, ts and te represent the start and end times of the time range constrained by the analysis query command, respectively, which can be extracted from the analysis query command by the system during the query parsing phase. Each document is scored as follows: if the document's time is outside the time range [ts, te], it is directly excluded; otherwise, a comprehensive score is calculated using the following formula:
[0040] .
[0041] in, Indicates the overall score. Document The evidence framework; , , These represent the set of entities, the set of spatial references, and the set of keywords extracted from the analysis query command, respectively. , These respectively represent the return of the evidence skeleton. The entity set and the spatial reference set in the text; This indicates that a set of keywords will be returned. Hit count in the summary and keyword fields of the evidence skeleton s(n); The cardinality of a set; , , The weights are assigned to the three dimensions of entity, space, and keyword, respectively, and satisfy the following conditions: Typical value , , This process prioritizes matching entities with the strongest semantic discriminative power. Before entity matching, the entity representations in the query are mapped to canonical names using an entity dictionary, and then sorted by overall score from highest to lowest to obtain the working subset. When the number of filtered results is less than the preset threshold, the system automatically reverts to all documents within the time frame to prevent subsequent stages from failing due to insufficient evidence.
[0042] S2.3: Divide the event evolution process into stages based on the working subset and analysis query instructions to generate an ordered sequence of event stages.
[0043] In one specific implementation, the event phase of this embodiment includes: a phase identifier, a phase name, a time range, a main geographical area, and a dominant entity. Event Phase Represented as = phase_id, name, t_range, loc_focus, key_ent Where phase_id is the phase identifier, name is the phase name, t_range is the phase time range, loc_focus is the main geographic area, and key_ent is the dominant entity within the phase.
[0044] The process of dividing the event evolution process into stages is as follows: the evidence skeleton of each document in the working subset is organized into a compressed metadata sequence ordered by time; the large language model identifies geographical focus drift, dominant entity replacement, and theme breakpoints based on the compressed metadata sequence as stage boundaries, generating an ordered sequence composed of several event stages; wherein, the division of stage boundaries is based on logical turning points in the evidence distribution rather than fixed calendar intervals. In this embodiment, the logical turning point refers to the point in time when the key attributes (geographical focus, dominant entity, theme distribution) of the evidence skeleton change significantly during the event evolution process, specifically obtained by the joint identification of the aforementioned three types of signals: geographical focus drift, dominant entity replacement, and theme breakpoints.
[0045] Specifically, the evidence skeleton of each document in the working subset Nw is compressed into a single-line structure: [Date] [DOC_ID] Title | Abstract | Entity | Location | Keywords. This is then arranged chronologically to form a compressed metadata timeline. This timeline, along with the analysis query command, is fed into an existing large language model with strong reasoning capabilities. Based on three types of signals—geographic focus shift, dominant entity replacement, and topic breakage—it detects stage boundaries and outputs an ordered sequence Φ=( consisting of 2 to 4 stages). 1, …, K ), where K is the total number of stages. Each stage k contains fields such as stage identifier, stage name, start and end dates, spatial focus, and dominant entity; k represents the stage index. This method divides stages based on logical turning points in the evidence distribution, avoiding semantic breaks caused by slicing according to fixed calendar intervals.
[0046] S3: Based on the analysis query instructions and phase division results, the analysis task is decomposed into sub-tasks undertaken by various dedicated analysis agents. The sub-tasks are organized into an execution plan in the form of a directed acyclic graph (DAG), and the sub-tasks are scheduled to be executed in parallel according to the dependencies of the DAG. The outputs of each agent are integrated into an analysis report with evidence citations. The analysis report is used by analysts to make analysis statements and provide feedback on the annotation of evidence sources.
[0047] S3.1: Based on the analysis query instructions and the phase division results, the analysis task is decomposed into sub-tasks undertaken by various dedicated analysis agents.
[0048] In one specific implementation, this embodiment generates a set of subtasks T={τ1, …,τ} based on the analysis query command, the event phase sequence Φ, and available agent roles. M The execution plan P = (T, D) is composed of} and its dependencies D, where M is the total number of subtasks. This embodiment sets up the following three dedicated analysis agents: The Global Monitor Agent performs macro-level situational analysis across different phases, outputting fact statements, topic clusters, and geographic event points with latitude and longitude coordinates.
[0049] Deep Dive Agents perform micro-tracking of specific target entities within one or more stages, outputting entity profiles, evolutionary trajectories, and event nodes with spatiotemporal attributes for visualization.
[0050] Relation Miner Agent extracts relationship edges such as cooperation and trade between multiple entities and provides causal labels and sources of evidence.
[0051] Each subtask declares in its definition the stage identifier it is bound to, the proxy it uses, and its dependencies on other subtasks. After the plan is generated, deterministic normalization is performed: the validity of the proxy name is verified, the intersection of the stage identifier and Φ is taken and if it is empty, it is rolled back to the full stage, duplicate tasks are removed, dependencies pointing to non-existent tasks and self-references are removed, and dependency loops are checked; if a loop exists, execution is aborted and an exception is returned.
[0052] It should be noted that the dependency relationship in this embodiment refers to the input-output relationship between subtasks, that is, the structured output of one subtask serves as the input condition for another subtask; for example, the subtask of the relationship mining agent analyzing the "relationship between entity A and entity B in stage k" can be declared to depend on the deep mining agent's trajectory tracking subtask of entity A in stage k, and the execution of the former can only be triggered after the latter is completed.
[0053] S3.2: Organize the subtasks into an execution plan in the form of a directed acyclic graph, and schedule the subtasks to be executed in parallel according to the dependencies of the directed acyclic graph.
[0054] S3.2.1: First, review the human-machine collaboration plan.
[0055] In one specific implementation, the execution plan is presented to analysts in a visual format before formal execution. Analysts can choose to approve the execution directly or provide modification suggestions (e.g., deleting a subtask, adding a new deep mining task, adjusting the entity of interest, etc.). The system feeds back the modification suggestions to the plan decomposition step to regenerate the execution plan, iterating until the analyst approves it.
[0056] S3.2.2: Construct a directed graph based on the dependencies declared between subtasks.
[0057] S3.2.2: Perform loop detection on directed acyclic graphs.
[0058] Specifically, a topological sort is used to check for cycles in the directed acyclic graph. During traversal, if all nodes with an in-degree of zero can be cleared in sequence, the graph is acyclic; otherwise, a cycle exists. If a cycle exists, an exception is returned.
[0059] S3.2.3: Add subtasks with an in-degree of zero to the ready queue for parallel execution in topological order.
[0060] Specifically, the execution process is as follows: initialize the in-degree of each task; put tasks with an in-degree of zero into the ready queue; and, without exceeding the maximum concurrency, take subtasks from the ready queue and assign them to the document subsets of the corresponding stages for parallel execution.
[0061] In this context, the in-degree refers to the number of dependent edges pointing to the task node. In this embodiment, the in-degree of the task node is the number of dependent tasks declared by the task.
[0062] S3.2.4: After the subtask is completed, the structured output is injected into the execution context of the subsequent task and the in-degree of the subsequent task is updated.
[0063] Specifically, after a subtask is completed, its structured output is injected into the execution context of the subsequent tasks that depend on it, and the in-degree of the subsequent tasks is updated, and the in-degree of the subsequent tasks is reduced accordingly. Once the in-degree is zero, the task is entered into the ready queue. If all dependencies of a task fail, the task is marked as skipped. The above process is repeated until all subtasks are completed.
[0064] S3.3: Integrate the outputs of each agent into an analysis report with evidence citations.
[0065] S4: Provide the analysis report to the analysts for analysis statements and feedback on the annotation of evidence sources. Use the annotation feedback as a constraint to adjust the use of evidence or task plan in subsequent analysis rounds.
[0066] S4.1: Provide the analysis report to the analyst for analysis statements and feedback on the source of evidence.
[0067] In one specific implementation, this embodiment models the analysis report to obtain an analysis report model, such as... Figure 2 As shown, the analysis report adopts a two-tiered structure with phase summaries and thematic chapters. The phase summaries generate situational descriptions stage by stage according to the event phase sequence, summarizing the evolution of events within a time phase. The thematic chapters categorize the analysis content into several thematic paragraphs, each consisting of several analysis statements. Each analysis statement includes the statement text, the phase identifier, the source document identifier supporting the statement, and a marker to distinguish factual statements from inferences.
[0068] Specifically, the integration process involves labeling the outputs of each subtask according to task identifier, agent type, and stage name before sending them into the integration model. The integration model then sorts the outputs by stage summary. and thematic chapters Final analysis report of output from two parallel structures ,in For the report title, A collection of stage summaries, A collection of thematic chapters, This is the overall summary. The phase summary generates a situational description stage by stage according to the event phase sequence; the thematic chapters consist of several thematic paragraphs, each containing several analytical statements. Composition, in which For the statement text, A list of source document identifiers to support this statement, This serves as the identifier for the relevant stage. Boolean tags are used to distinguish factual statements from inference statements. The integrated model only allows references to document identifiers that explicitly appear in the subtask outputs and must not fabricate sources.
[0069] It should be noted that the integrated model in this embodiment is a large language model instance with strong reasoning capabilities. It is responsible for organizing the content according to the structure defined by the analysis report model, and is different from the analysis report model as a data structure specification. It can be implemented using the existing large language model structure, so it will not be described in detail here.
[0070] Each agent's output adopts a two-layer structure with a fact layer and an inference layer: the fact layer consists of several fact statements with document identifiers, which are objective statements directly extracted or summarized by the agent from the source documents. The inference layer consists of several inference statements, which are analytical conclusions derived by the agent based on further reasoning from the fact layer. Typical types include core contradictions, hidden intentions, and trend predictions. Each inference statement also carries the source document identifier it is based on to ensure the traceability of the inference conclusions. The explicit layering of the fact layer and the inference layer makes it easier for analysts to distinguish between objective information and the agent's explanatory conclusions, improving the auditability of the analysis report.
[0071] The annotation feedback serves as a constraint, adjusting the use of evidence or task plans in subsequent analysis rounds. Annotation feedback includes two categories: irrelevant and contradictory. The irrelevant label excludes evidence sources marked as irrelevant from the relevant proxy prompts in the corresponding analysis statement. The contradictory label responds to evidence sources marked as contradictory as explicit counter-evidence. Annotated sources are unaffected in other irrelevant analysis statements.
[0072] Specifically, after the analysis report is generated, analysts can review the analysis statements and the sources of evidence they cite. For each statement's evidence... ,in This indicates a statement. This indicates evidence. Analysts can assign one of two labels: irrelevant, indicating that the source of evidence is unrelated to the topic of the statement; or contradictory, indicating that the content of the source of evidence contradicts the conclusion of the statement. The accumulated set of labels... , ,in, Indicates a label, Indicates an irrelevant tag. The label indicates a contradiction. The set of labels is injected as constraints into the corresponding agent in subsequent analysis rounds: evidence labeled as irrelevant will no longer be cited in the re-reasoning of the corresponding statement; evidence labeled as contradictory will be explicitly prompted to respond to or reconcile the contradiction with the agent. Constraints are injected at the granularity of statement-evidence pairs and do not affect the normal use of the same evidence in other unrelated statements.
[0073] S4.2: Use annotation feedback as a constraint to adjust the use of evidence or task plan in subsequent analysis rounds.
[0074] S4.2.1: Translate the constraints generated by time-based selection, geographic selection, or entity filtering operations performed by analysts on the visualization interface into structured prompt word templates.
[0075] In one specific implementation, when analysts perform interactions such as time range selection, geographic area selection, or entity focusing on the visualization interface, the system translates the interaction status into a structured prompt word template and automatically fills it in the analysis query command input area. After the analyst confirms or edits it, it serves as a new analysis query command to start a new analysis cycle.
[0076] The complete process described above yields an analysis report that explicitly corresponds to the source evidence. Analysts can read this report sequentially by stage, navigate by topic, and verify each piece of evidence. Through multiple iterations, analysts can continuously refine their analysis, ultimately arriving at credible conclusions validated by the evidence.
[0077] S4.2.2: After the structured prompt word template is confirmed by the analyst, it is used as a new analysis query command to start a new analysis round.
[0078] Example 2: Embodiment 2 of the present invention provides a human-machine collaborative evolution event visualization analysis system based on stage division, comprising: The input acquisition module is used to acquire the document data of the task to be analyzed, which constitutes the document corpus, as well as the analysis and query instructions; The evidence retrieval and stage awareness module is used to perform stage analysis and division of the event evolution process using document corpora and analysis query commands; The phase-aware analysis generation module is used to decompose the analysis task into sub-tasks undertaken by various dedicated analysis agents according to the analysis query instructions and phase division results, organize the sub-tasks into an execution plan in the form of a directed acyclic graph, schedule the execution of the sub-tasks in parallel according to the dependencies of the directed acyclic graph, and integrate the output of each agent into an analysis report with evidence citations. The human-machine collaborative iterative optimization module is used to provide analysis reports to analysts for analysis declarations and feedback on the annotation of evidence sources. The annotation feedback is used as a constraint to adjust the use of evidence or task plans in subsequent analysis rounds.
[0079] In one specific implementation, the system is a web-based interactive system with five collaborative views: an input panel, an analysis report panel, a spatiotemporal visualization panel, a stage interaction panel, and an evidence review panel. The report panel presents the source of evidence for each analysis claim via links; the spatiotemporal visualization panel includes a terrain view and a relationship network view, supporting the display of relevant hotspots by geographical location and the display of relationships between entities by force-oriented layout; the stage interaction panel overlays event stages as color-coded stripes on the timeline, supporting range selection and spatial label switching; and the evidence review panel displays the original source documents cited, allowing analysts to verify and annotate the evidence.
[0080] Example 3: Embodiment 3 of the present invention provides a computer-readable storage medium storing a computer program adapted for loading by a processor and executing the steps in the human-machine collaborative evolution event visualization analysis method based on stage division as described in Embodiment 1 of the present invention.
[0081] Example 4: Embodiment 4 of the present invention provides a computer device, the device comprising: A processor, adapted to execute computer programs; A computer-readable storage medium storing a computer program, which, when executed by the processor, implements the steps in the human-machine collaborative evolution event visualization analysis method based on stage division as described in Embodiment 1 of the present invention.
[0082] The steps and methods involved in Examples 2, 3 and 4 above correspond to those in Example 1. For specific implementation details, please refer to the relevant description section of Example 1.
[0083] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed in this application can be implemented in electronic hardware or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.
[0084] In the above embodiments, implementation can be achieved, in whole or in part, through software, hardware, firmware, or any combination thereof. When implemented in software, it can be implemented, in whole or in part, as a computer program product. A computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, all or part of the flow or function according to the embodiments of this application is generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in or transmitted through a computer-readable storage medium. The computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired or wireless means. The computer-readable storage medium can be any available medium that a computer can access or a data processing device such as a server or data center that integrates one or more available media. The available medium can be a magnetic medium, an optical medium, or a semiconductor medium, etc.
[0085] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.
Claims
1. A visual analysis method for human-machine collaborative evolution events based on stage division, characterized in that, Includes the following steps: The document data for the task to be analyzed is used to construct a document corpus and generate analysis and query commands. The event evolution process is analyzed and divided into stages using a document corpus and analytical query commands. Based on the analysis query instructions and the phase division results, the analysis task is decomposed into sub-tasks undertaken by various dedicated analysis agents. The sub-tasks are organized into an execution plan in the form of a directed acyclic graph, and the execution of the sub-tasks is scheduled in parallel according to the dependencies of the directed acyclic graph. The outputs of each agent are integrated into an analysis report with evidence citations. The analysis report is used by analysts to make analysis statements and provide feedback on the annotation of evidence sources. The analysis report is provided to the analysts for analysis statements and feedback on the annotation of evidence sources. The annotation feedback is used as a constraint to adjust the use of evidence or task plan in subsequent analysis rounds.
2. The human-machine collaborative evolution event visualization analysis method based on stage division as described in claim 1, characterized in that, The specific steps for performing stage analysis and division of the event evolution process using document corpora and analysis query commands are as follows: Structured metadata is extracted from each document in the document corpus to generate an evidence skeleton; Based on the analysis query command, the evidence skeleton is screened and scored in multiple dimensions to obtain the working subset; The event evolution process is divided into stages based on the working subset and analysis query instructions, generating an ordered sequence of event stages.
3. The human-machine collaborative evolution event visualization analysis method based on stage division as described in claim 2, characterized in that, The evidence skeleton includes: document identifier, summary, entity set, spatial reference, time attribute, and keywords; the event stage includes: stage identifier, stage name, time range, main geographical area, and dominant entity.
4. The human-machine collaborative evolution event visualization analysis method based on stage division as described in claim 2, characterized in that, The specific steps for multi-dimensional screening and scoring of the evidence skeleton based on the analysis query command are as follows: The analysis query command is parsed into structured search constraints that include entities, time ranges, spatial ranges, and keywords; Documents are hard-filtered based on time range, and the remaining documents after filtering are weighted and scored according to three dimensions: entity, space, and keywords. The documents with the highest scores are selected as the working subset.
5. The human-machine collaborative evolution event visualization analysis method based on stage division as described in claim 2, characterized in that, The specific steps for dividing the event evolution process into stages based on the working subset and analysis query instructions are as follows: The evidence skeletons of each document in the working subset are organized into a compressed metadata sequence ordered by time. The large language model identifies geographic focus shifts, dominant entity changes, and thematic breakpoints based on compressed metadata sequences as stage boundaries, generating an ordered sequence of several event stages. The stage boundaries are based on logical turning points in the evidence distribution rather than fixed calendar intervals.
6. The human-machine collaborative evolution event visualization analysis method based on stage division as described in claim 1, characterized in that, The specific steps for parallel scheduling of subtasks based on the dependencies of a directed acyclic graph are as follows: Loop detection in directed acyclic graphs; Subtasks with an in-degree of zero are added to the ready queue and executed in parallel according to topological order; After a subtask is completed, the structured output is injected into the execution context of the subsequent task and the in-degree of the subsequent task is updated.
7. The human-machine collaborative evolution event visualization analysis method based on stage division as described in claim 1, characterized in that, The specific steps for using annotation feedback as a constraint to adjust evidence use or task planning in subsequent analysis rounds are as follows: The constraints generated by the time-based selection, geographic selection, or entity filtering operations performed by analysts on the visualization interface are translated into structured prompt word templates. Once the structured keyword template is confirmed by the analysts, it will be used as a new analytical query to start a new round of analysis.
8. A human-machine collaborative evolution event visualization analysis system based on stage division, characterized in that, include: The input acquisition module is used to acquire the document data of the task to be analyzed, which constitutes the document corpus, as well as the analysis and query instructions; The evidence retrieval and stage awareness module is used to perform stage analysis and division of the event evolution process using document corpora and analysis query commands; The phase-aware analysis generation module is used to decompose the analysis task into sub-tasks undertaken by various dedicated analysis agents according to the analysis query instructions and phase division results, organize the sub-tasks into an execution plan in the form of a directed acyclic graph, schedule the execution of the sub-tasks in parallel according to the dependencies of the directed acyclic graph, and integrate the output of each agent into an analysis report with evidence citations. The human-machine collaborative iterative optimization module is used to provide analysis reports to analysts for analysis declarations and feedback on the annotation of evidence sources. The annotation feedback is used as a constraint to adjust the use of evidence or task plans in subsequent analysis rounds.
9. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program adapted to be loaded by a processor and executed as described in any one of claims 1-7: Human-Machine Cooperative Evolution Event Visualization Method Based on Stage Division.
10. A computer device, characterized in that, include: A processor, adapted to execute computer programs; A computer-readable storage medium storing a computer program, which, when executed by the processor, implements the human-machine collaborative evolution event visualization analysis method based on stage division as described in any one of claims 1-7.