Standard document duplicate checking result visual display method and system

Abstract

The invention discloses a standard document duplicate checking result visual display method and system, and relates to the technical field of document information processing, and the method comprises the steps: obtaining a target standard document and a comparison standard document set, and processing the target standard document and the comparison standard document set to generate a structured document object set; performing multi-level semantic association degree fusion based on the structured document object set, constructing a multi-dimensional association knowledge network, and rendering to generate an initial visual view; in the interaction process of the user and the initial visual view, monitoring a user interaction load index and a view complex index of the initial visual view; and evaluating the real-time cognitive load of the user based on the user interaction load index and the view complex index, and dynamically scheduling and optimizing the presentation form of the initial visual view according to the real-time cognitive load to generate an optimized view stream. According to the method, through adaptive view optimization and interactive strategy deduction, the analysis efficiency and the decision support capability in a complex information environment are effectively improved.

Description

Technical Field

[0001] This invention relates to the field of document information processing technology, and in particular to a method and system for visualizing the results of standard document plagiarism checks. Background Technology

[0002] With the application of artificial intelligence technology, standard document deduplication has evolved from keyword matching to semantic analysis based on deep learning. Current mainstream solutions typically parse documents and use large language models to generate semantic vectors. By calculating vector similarity, document associations are quantified, and the results are finally presented in the form of a list or static knowledge graph. This method can reveal the semantic connections between documents and provides richer information than traditional retrieval.

[0003] When faced with complex interconnected networks composed of massive amounts of documents, existing visualization technologies have limitations. They statically present the status of plagiarism detection results and lack dynamic support for the user's analysis process. Complex networks can easily lead to visual information overload, forcing users to spend a lot of energy manually adjusting the view, thus diverting the in-depth analysis of core legal and technical issues. Existing solutions are read-only, and users cannot directly make virtual modifications to the technical solutions in the visualization environment, nor can they observe the dynamic impact of different modification strategies on the interconnected structure in real time and in comparison. This makes it difficult for the analysis process to effectively support key strategy deduction and decision-making. Summary of the Invention

[0004] In view of the aforementioned existing problems, the present invention is proposed.

[0005] Therefore, this invention provides a method for visualizing standard document plagiarism detection results, which solves the problems of excessive user cognitive load caused by static presentation and read-only interaction mode in the prior art, as well as the inability to directly perform dynamic strategy deduction and decision support in a visual environment.

[0006] To solve the above-mentioned technical problems, the present invention provides the following technical solution: In a first aspect, the present invention provides a method for visualizing the results of standard document plagiarism detection, which includes obtaining a target standard document and a set of comparison standard documents, and processing them to generate a set of structured document objects; Multi-level semantic association is fused based on a set of structured document objects to construct a multi-dimensional associated knowledge network and render an initial visualization view. During user interaction with the initial visual view, monitor user interaction load metrics and the view complexity metrics of the initial visual view; Based on user interaction load indicators and view complexity indicators, the real-time cognitive load of users is evaluated. The presentation form of the initial visualization view is dynamically scheduled and optimized according to the real-time cognitive load, and an optimized view flow is generated. The sand table simulation branches are created by receiving virtual modification instructions from users on the target standard document. The structural evolution of the multidimensional related knowledge network caused by each sand table simulation branch is derived and compared. Extract trade-off patterns and potential paths from structural evolution, and provide strategic path visualization guidance and advantageous strategy marking in the optimization view flow; Record the scheduling logs of the optimized view flow and the guidance information of the sandbox simulation branches, and generate a comprehensive report by combining the multi-dimensional related knowledge network status corresponding to the user's sandbox simulation branches.

[0007] As a preferred embodiment of the standard document plagiarism detection result visualization method of the present invention, the method includes the following steps: obtaining the target standard document and the comparison standard document set, and processing them to generate a set of structured document objects: Receive the target standard document, retrieve the set of comparison standard documents from the comparison document database, parse the document format of the target standard document and the set of comparison standard documents, and extract the clean text; The clean text is analyzed using rules and natural language processing models to divide it into technical field blocks, background technology blocks, and technical protection scope blocks. Using a pre-trained language model, semantic vectors are generated for the overall text, technical field section blocks, background technology section blocks, and technical protection scope section blocks of each document in the target standard document and the comparison standard document set. The plain text, technical field section blocks, background technology section blocks, technical protection scope section blocks, and semantic vectors of each part of each document in the target standard document and comparison standard document sets are encapsulated into a structured document object collection.

[0008] As a preferred embodiment of the standard document plagiarism detection result visualization method described in this invention, the method includes the following steps: multi-level semantic association fusion based on a structured document object set to construct a multi-dimensional association knowledge network and render an initial visualization view: Document-level relevance is obtained by comparing the overall text semantic vector of the target standard document in the set of structured document objects with the overall text semantic vector of each comparison standard document; The correlation degree at the technical protection scope level is obtained by comparing the semantic vectors of the technical protection scope chapter blocks of the target standard document in the set of structured document objects with the semantic vectors of the technical protection scope chapter blocks of each comparative standard document. The background-level relevance is obtained by comparing the semantic vectors of the background technology section blocks of the target standard document with the semantic vectors of the background technology section blocks of each comparison standard document in the set of structured document objects. By applying preset weighted fusion rules, the document-level relevance, technical protection scope-level relevance, and background-level relevance of the target standard document and each comparison standard document are fused to generate a multi-level semantic relevance fusion score. Using all documents in a structured document object collection as nodes, and multi-level semantic association scores and inter-document reference relationships as edges, a multi-dimensional related knowledge network is constructed. The nodes and edges in the multidimensional knowledge network are encoded according to the preset visual mapping rules, and then rendered on an interactive canvas using a graphics layout algorithm to generate an initial visual view.

[0009] As a preferred embodiment of the standard document plagiarism detection result visualization method described in this invention, the following steps are included: Monitoring the user interaction load index and the view complexity index of the initial visualization view during user interaction with the initial visualization view includes the following steps: During the user's interaction with the initial visual view, every click, drag, and zoom event is captured to form interactive event events. Based on interactive events, the click frequency, average speed of drag operations, and frequency of zoom operations per unit time are statistically analyzed as user interaction load indicators. During the user's interaction with the initial visual view, the total number of nodes rendered in the current visible area of ​​the initial visual view is counted and used as a complex indicator of the total number of nodes. During the user's interaction with the initial visual view, the total density of edges rendered by the initial visual view within the current visible area is calculated and used as a complex indicator of total edge density. During the user's interaction with the initial visual view, the number of different visual encoding attributes used by the initial visual view within the current visible area is counted as an indicator of visual encoding complexity. The user interaction load index, the total number of nodes complexity index, the total edge density complexity index, and the visual coding type complexity index are collectively referred to as the user interaction load index and the view complexity index of the initial visualization view.

[0010] As a preferred embodiment of the standard document plagiarism detection result visualization method described in this invention, the method involves: assessing the user's real-time cognitive load based on user interaction load and view complexity indicators, including the following steps: The click frequency, average speed of drag operations, and frequency of zoom operations in the user interaction load indicators are compared with the preset cognitive load benchmark parameters to obtain the first load assessment value. The total number of nodes, total edge density, and visual coding type complexity indicators in the view complexity index are compared with their corresponding visual complexity thresholds, and then weighted and summed to obtain the second load assessment value. Based on predefined cognitive load synthesis rules, the first load assessment value and the second load assessment value are synthesized to output the user's real-time cognitive load value.

[0011] As a preferred embodiment of the standard document plagiarism detection result visualization method described in this invention, the optimized view stream is generated based on the dynamic scheduling of real-time cognitive load and the presentation format of the initial visualization view, including the following steps: The real-time cognitive load value is compared with the first load threshold and the second load threshold to determine the current cognitive state of the user. When the real-time cognitive load value exceeds the first load threshold, the view simplification rule is triggered, and the initial visualization view is subjected to operations such as fading secondary related edges, collapsing non-focus area nodes, and hiding some metadata labels. When the real-time cognitive load value is lower than the second load threshold and the interactive events are concentrated in a specific area, the view focus rule is triggered to perform an operation to expand the details of the specific area in the initial visualization view. The adjusted initial visualization, generated by view simplification rules or view focus rules, is serialized into an optimized view flow.

[0012] As a preferred embodiment of the standard document plagiarism detection result visualization method described in this invention, the method includes the following steps: receiving a user's virtual modification instruction for the target standard document to create a sandbox simulation branch, and deriving and comparing the structural evolution of the multidimensional related knowledge network caused by each sandbox simulation branch. In the optimized view flow, virtual modification instructions for the content of the target standard document are received, and the corresponding semantic vector of the target standard document is updated. Using the updated target standard document semantic vector, the multi-level semantic association degree fusion score of all document nodes in the multi-dimensional association knowledge network is re-derived, and the updated multi-dimensional association knowledge network corresponding to each sand table simulation branch is reconstructed. By comparing the updated multidimensional association knowledge network corresponding to each sand table simulation branch with the original multidimensional association knowledge network in parallel, the strengthening, weakening and disappearance of the association edges between nodes are identified as the structural evolution of the multidimensional association knowledge network.

[0013] As a preferred embodiment of the standard document plagiarism detection result visualization method described in this invention, the method involves: extracting trade-off patterns and potential paths from structural evolution, and visually guiding strategy paths and marking advantageous strategies in the optimized view flow, including the following steps: Based on the structural evolution of the multidimensional related knowledge network, by comparing the combination of changes in related edges caused by different sand table simulation branches, we can summarize the related adversarial mode of increasing and decreasing related strength and the related synergistic mode of weakening multi-sided related strength together, which together constitute a trade-off mode. Based on the trade-off model, the logically continuous structural evolution is linked together to plan the evolution sequence from the current network state to the low correlation density region as a potential path; Based on the evolution direction and key nodes of the potential path, a dynamically extending semi-transparent guide trajectory is generated over the current view in the optimized view flow to determine the strategy path visualization guidance. By applying preset multi-objective evaluation rules, the performance of the sand table simulation branches in terms of modification costs and risk reduction effects is quantitatively scored. Branches with scores higher than the preset advantage threshold are identified as advantageous strategies and marked with specific visual symbols that distinguish them from ordinary branches in the optimization view flow, thus determining the advantageous strategy markers.

[0014] As a preferred embodiment of the standard document plagiarism detection result visualization method described in this invention, the following steps are included: recording the scheduling log of the optimized view flow and the guidance information of the sandbox simulation branches, and combining the multi-dimensional related knowledge network status corresponding to the user's sandbox simulation branches to generate a comprehensive report: Capture and store the scheduling logs of the optimized view flow, which include the trigger time, reason and specific operation records of view simplification rules and view focus rules; Capture and store guidance information for sand table simulation branches, including trajectory data of strategy path visualization guidance and scoring criteria for advantageous strategy marking; When a user selects a sand table simulation branch as the final strategy, the updated state of the multidimensional related knowledge network corresponding to the sand table simulation branch is locked. Based on the scheduling logs, guidance information, and the updated multidimensional related knowledge network status corresponding to the final strategy, a comprehensive report is output by formatting, integrating, and displaying text and graphics according to a preset report template.

[0015] Secondly, the present invention provides a standard document plagiarism detection result visualization system, including a data acquisition module, which acquires the target standard document and the comparison standard document set, and processes them to generate a structured document object set; The rendering module performs multi-level semantic association fusion based on a set of structured document objects, constructs a multi-dimensional related knowledge network, and renders and generates an initial visual view. The monitoring module monitors user interaction load metrics and the view complexity metrics of the initial visualization view during user interaction with the initial visualization view. The derivation module assesses the user's real-time cognitive load based on user interaction load and view complexity indicators. It dynamically schedules and optimizes the presentation of the initial visualization view based on the real-time cognitive load, generates an optimized view stream, receives the user's virtual modification instructions for the target standard document to create sand table simulation branches, and derives and compares the structural evolution of the multidimensional related knowledge network caused by each sand table simulation branch. The strategy generation module extracts trade-off patterns and potential paths from structural evolution, and provides visual guidance on strategy paths and marks advantageous strategies in the optimization view flow. The comprehensive report generation module records the scheduling logs of the optimized view flow and the guidance information of the sandbox simulation branches. It combines the multi-dimensional related knowledge network status corresponding to the user's sandbox simulation branches to generate a comprehensive report.

[0016] The beneficial effects of this invention are as follows: A document object set is constructed through structured processing and semantic vectorization. Based on this, a multi-dimensional related knowledge network is built through multi-level semantic correlation fusion to generate an initial visual view. Cognitive load is assessed by real-time monitoring of user interaction load and view complexity indicators, and the optimized view is dynamically scheduled and optimized to generate an optimized view flow. A sandbox simulation mechanism is introduced into the optimized view flow, allowing users to make virtual modifications and deduce the structural evolution of the related network in real time. This allows for the extraction of trade-off patterns and potential paths for visual guidance and strategy labeling. By recording the complete interaction and simulation process and combining it with the network state corresponding to the final strategy, a comprehensive analysis report is generated. This realizes the evolution of plagiarism detection results from static display to dynamic interaction. Through adaptive view optimization and interactive strategy simulation, the analysis efficiency and decision support capabilities in complex information environments are effectively improved. Attached Figure Description

[0017] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0018] Figure 1 A flowchart for visualizing the results of standard document plagiarism checks.

[0019] Figure 2 This is a schematic diagram of a system for visually displaying the results of standard document plagiarism checks. Detailed Implementation

[0020] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0021] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.

[0022] Secondly, the term "one embodiment" or "example" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the invention. The appearance of an embodiment in different places in this specification does not necessarily refer to the same embodiment, nor is it a single or selective embodiment that mutually excludes other embodiments.

[0023] Reference Figures 1-2 This is one embodiment of the present invention, which provides a method for visualizing standard document plagiarism detection results, including the following steps: S1. Obtain the target standard document and the set of comparison standard documents, and process them to generate a set of structured document objects.

[0024] S1.1 Receive the target standard document, retrieve the set of comparison standard documents from the comparison document database, parse the document formats of the target standard document and the set of comparison standard documents, and extract the clean text.

[0025] Furthermore, the process of receiving the target standard document and retrieving the set of comparative standard documents from the comparative document database ensures the specificity of the analysis object and the clarity of the comparison scope. The comparative document database can be a structured knowledge base containing standard documents of literature, technical standards, and academic papers. Parsing the document formats of the target standard document and the set of comparative standard documents and extracting clean text is a prerequisite for all subsequent advanced semantic processing. The diversity of document formats, such as PDF, DOCX, or HTML, requires the parsing process to be compatible with different structural tags and encoding methods, accurately extracting the pure text content while excluding layout, images, and other non-textual interference information. The extracted clean text provides a unified, machine-readable input for subsequent in-depth analysis and understanding. The accurate extraction of clean text ensures the source quality of semantic analysis and avoids semantic deviations caused by format noise. Specifically, for example, a document instruction manual may contain complex chemical formula diagrams. The parsing process needs to accurately identify and skip these non-text areas, extracting only descriptive text paragraphs. This relies on mature document parsing technology, which completes document reception, retrieval, parsing, and text extraction to output a clean set of text.

[0026] S1.2. Use rules and natural language processing models to analyze the text structure of the clean text and segment it into technical field chapter blocks, background technology chapter blocks, and technical protection scope chapter blocks.

[0027] Furthermore, given the unique and highly structured stylistic features of standard documents, precise semantic unit segmentation is crucial. Traditional text processing methods often treat documents as a whole or perform simple paragraph divisions, which cannot meet the needs of differentiated and refined comparisons of different chapters in standard document plagiarism detection. Rule-based pattern matching and semantic understanding based on natural language processing models are essential. Rules can be used for rapid location and initial screening based on common chapter title keywords, numbering systems, and fixed sentence structures in standard documents. For example, using technical field patterns to locate the beginning of technical field chapter blocks. However, rules alone may fail due to non-standard document formats; a natural language processing model is needed to understand the contextual semantics and assist in determining chapter boundaries, especially for paragraphs without explicit titles but semantically belonging to a specific chapter.

[0028] Specifically, this combination of rules and models ensures both the efficiency and accuracy of parsing standardized documents. The documents are divided into technical field sections, background technology sections, and scope of protection sections. The technical field sections define the scope of the document, the background technology sections describe existing technical issues, and the scope of protection sections define the legal protection scope. This semantic-based segmentation allows subsequent semantic vector generation and correlation calculations to be performed at the corresponding levels, achieving multi-level, refined semantic comparison from macro-level technical background to micro-level scope of protection. Through parsing the structure of the clean text, a set of text fragments precisely segmented into technical field sections, background technology sections, and scope of protection sections is output.

[0029] S1.3 Using a pre-trained language model, semantic vectors are generated for the overall text, technical field section blocks, background technology section blocks, and technical protection scope section blocks of each document in the target standard document and the comparison standard document set.

[0030] Furthermore, it achieves a distributed representation of text semantics from discrete symbols to continuous high-dimensional vectors, and this representation is hierarchical. Pre-trained language models, such as large-scale models based on the Transformer architecture, have internalized rich linguistic and world knowledge through self-supervised learning on ultra-large-scale corpora. They can encode the semantics of words, phrases, and even paragraphs into a dense vector of fixed dimensions, generate a global semantic vector for the entire document to capture the central theme of the document, and generate local semantic vectors for semantic units with specific functions, such as technical field sections, background technology sections, and technical protection scope sections. It preserves the semantic specificity of different components within the document.

[0031] S1.4. Encapsulate the clean text, technical field section blocks, background technology section blocks, technical protection scope section blocks, and semantic vectors of each part of each document in the target standard document and comparison standard document sets into a structured document object collection.

[0032] Furthermore, through encapsulation, a logical digital twin is created for each document. Each object in the structured document object collection completely encapsulates all the key information of a document, from its original content and structured chunks to its deep semantic representation. There is no need to go back to the original document file or repeat parsing and vectorization; you can directly manipulate this structured object collection. Encapsulation greatly improves the efficiency of the process and the consistency of the data, so that the original text, logical structure, and semantic information of the document are organically linked together to form a semantically enhanced document representation.

[0033] Specifically, for example, when it's necessary to trace a highly relevant calculation result, the specific chapter or block text that generates the association can be directly located from the object, achieving a seamless link from data to interpretable insights. The construction of the structured document object collection is a crucial link in the entire methodology, solidifying the results of previous data processing into a stable and reliable analytical foundation, enabling subsequent complex interactive analysis and visualization. Through encapsulation operations, a structured document object collection containing complete semantic and structural information is generated, which can be directly invoked by subsequent processes.

[0034] S2. Based on a set of structured document objects, perform multi-level semantic association fusion to construct a multi-dimensional associated knowledge network and render an initial visual view.

[0035] S2.1. Based on comparing the overall text semantic vector of the target standard document in the structured document object set with the overall text semantic vector of each comparison standard document, obtain the document-level relevance.

[0036] Furthermore, it provides a benchmark metric for macro-level thematic similarity between documents. The overall text semantic vector captures the global semantic theme of the document. By obtaining the similarity between two such vectors, such as cosine similarity, a comprehensive score reflecting the similarity between the two documents in terms of overall technical theme and solutions can be obtained. Document-level relevance helps to quickly identify a set of related documents that belong to the same technical field or solve similar technical problems as the target standard document.

[0037] S2.2. Based on the semantic vectors of the technical protection scope chapter blocks of the target standard document in the comparative structured document object set and the semantic vectors of the technical protection scope chapter blocks of each comparative standard document, obtain the correlation degree at the technical protection scope level.

[0038] Furthermore, the section on the scope of technology protection defines the specific scope of protection sought by the technical solution and serves as the basis for infringement determination. Extracting and comparing the semantic vectors of the blocks in the section on the scope of technology protection is essentially a specific assessment of the degree of overlap and conflict between the two documents in terms of legal claims and the boundaries of technological exclusivity. This comparison is more accurate and legally significant than a comparison of the entire document.

[0039] S2.3. Based on the semantic vectors of the background technology section blocks of the target standard document and the semantic vectors of the background technology section blocks of each comparison standard document in the set of structured document objects, obtain the background-level correlation.

[0040] Furthermore, by establishing deeper connections between documents from the perspective of technical problems and the development of technology, the background technology section typically explains the technical problems that the document aims to solve, the shortcomings of existing technologies, and the starting point of this document. By comparing the semantic vectors of the background technology section blocks, it is possible to effectively identify existing technical documents that are committed to solving the same or similar technical problems as the target standard document. This is of great significance for judging the actual contribution of the document, understanding the path of technological evolution, and conducting existing technology searches.

[0041] S2.4 Apply the preset weighted fusion rules to fuse the document-level relevance, technical protection scope-level relevance, and background-level relevance of the target standard document with each comparison standard document, and generate a multi-level semantic relevance fusion score.

[0042] Furthermore, recognizing that different levels of relevance have different importance and indicative significance, and cannot be simply averaged, the pre-set weighted fusion rules reflect the integration of knowledge from the standard document analysis domain. Typically, the relevance at the technical protection scope level is given the highest weight because it directly relates to the scope of legal protection, as it most directly indicates the likelihood of infringement risk. The document-level relevance provides relevance in the macro-technical field and has the next highest weight. The background-level relevance helps to understand the technical context and has a relatively low but not negligible weight. Through a weighted formula, such as linear weighted summation, the relevance of these three different dimensions and weights is integrated into a single multi-level semantic relevance fusion score.

[0043] Specifically, the score comprehensively reflects the multi-level similarity between documents, from macro-themes and core protection scope to technical background. For example, even if the document-level correlation between two documents is average, if their technical protection scope correlation is extremely high, the fusion score will be high, accurately indicating high risk. Conversely, if only the background technology is similar but the technical protection scope is completely different, the fusion score will not be too high, avoiding false alarms. This fusion method overcomes the one-sidedness of single-level comparison, making the correlation metric used to build the network more robust and practically valuable. By applying weighted fusion rules, a comprehensive multi-level semantic correlation fusion score is generated for each pair of target and comparison documents.

[0044] S2.5. Using all documents in the structured document object collection as nodes, and multi-level semantic association scores and inter-document reference relationships as edges, construct a multi-dimensional related knowledge network.

[0045] Furthermore, the discrete document set is transformed into an interconnected graph-structured knowledge system rich in semantics and citation relationships. Nodes represent each independent document entity, while edges carry two key relationships: one is a semantic similarity relationship based on a multi-level semantic relevance fusion score, where the edge weight is determined by this score—a higher score results in a stronger or thicker edge, indicating a closer semantic connection; the other is an explicit technical relationship based on inter-document citation relationships, such as citation relationships in literature. These edges are typically represented in different types or styles. Constructing such a multi-dimensional relational knowledge network has significant advantages, revealing a network of complex inter-document relationships across the entire technical field. In this network, the target document is placed at the center and connected to multiple comparative documents through semantic and citation edges, and various connections may also exist between the comparative documents.

[0046] S2.6. Encode the nodes and edges in the multidimensional related knowledge network according to the preset visual mapping rules, and render them on the interactive canvas using a graphics layout algorithm to generate an initial visual view.

[0047] Furthermore, visual mapping rules define how data attributes are mapped to visual attributes. For example, the color of a node can map to the legal status of a document, the size of a node can map to the number of times a document is cited or its importance, the thickness and color intensity of edges can map to the level of semantic relevance fusion scores at multiple levels, and the line type of an edge can distinguish between semantic relevance edges and reference relationship edges. The graph layout algorithm is responsible for calculating the position of nodes on the canvas, striving to present the network structure clearly and aesthetically. For example, the force-directed layout algorithm is used to group closely connected nodes together and keep sparsely connected nodes away from each other. Rendering on an interactive canvas means that the generated view allows users to pan, zoom, and click to view details.

[0048] Specifically, the complex multidimensional knowledge network is presented in an easy-to-understand and exploratory way. Visual mapping rules enable users to quickly capture key information, such as identifying high-risk, highly related nodes and important core reference nodes at a glance. Interactive features allow users to delve deeper into the parts of interest. The initial visualization view serves as the starting point and main interface of the entire analysis process, providing users with a powerful visual tool for understanding the global relationship between documents. By applying visual mapping rules and graphical layout algorithms, an initial visualization view showcasing the multidimensional knowledge network is generated.

[0049] S3. During the user's interaction with the initial visualization view, monitor the user interaction load index and the view complexity index of the initial visualization view.

[0050] S3.1 During the user's interaction with the initial visual view, capture every click, drag, and zoom operation event to form interactive operation events.

[0051] Furthermore, all of a user's active exploration of the visual interface is transformed into a quantifiable sequence of low-level events. Click events typically indicate a user's attention to and selection of specific graphical elements such as nodes or edges; drag events reflect a user's intention to pan the view within the canvas space to find areas of interest; zoom events reflect a user's need to adjust the view granularity to observe macro layouts or micro details. Capturing these basic interaction events constitutes a primary source of data for understanding users' real-time behavioral patterns and intentions.

[0052] Specifically, these event records include event type, timestamp of occurrence, target graphic element or spatial coordinate metadata, and aggregate all discrete interactive behaviors into an interactive operation event stream ordered by time. By capturing and recording the user's click, drag and zoom operations in real time, an interactive operation event stream reflecting the dynamic interaction process between the user and the view is generated.

[0053] S3.2 Based on interactive operation events, the click frequency, average speed of drag operations, and frequency of zoom operations per unit time are statistically analyzed as user interaction load indicators.

[0054] Furthermore, behavioral features that indirectly reflect a user's cognitive effort and operational fluency can be extracted from the event stream. Click frequency refers to the number of times a user triggers a click event within a fixed time window. A high click frequency may indicate that the user is intensively exploring different document nodes to find clues, or frequently switching points of interest in a complex network. This may suggest that the user is facing a high information filtering cost or has an unclear goal. The average speed of dragging operations is obtained by obtaining the ratio of the displacement to the time spent of all dragging operation events over a period of time. A sustained low average dragging speed may mean that the user is carefully and systematically examining the view content, in a state of deep search; while a sudden high average speed may indicate that the user is quickly skipping the current area to find a new point of interest, or is losing patience due to information overload.

[0055] Specifically, the frequency of zoom operations is counted as the number of times a user performs zoom operations per unit of time. High-frequency zooming may indicate that the user is repeatedly switching between macro overview and micro details, attempting to establish a connection between the overall structure and local information. This context switching itself requires high cognitive resources. Click frequency, average speed of drag operations, and zoom operation frequency are used together as user interaction load indicators to quantify the intensity of user interaction and exploration patterns from different perspectives. This provides an objective behavioral signal for assessing the cognitive pressure that users experience when processing complex visual information. By performing statistical analysis on the interaction operation event flow within a time window, user interaction load indicators are output.

[0056] S3.3 During the interaction between the user and the initial visual view, the total number of nodes rendered by the initial visual view within the current visible area is counted as a complex indicator of the total number of nodes.

[0057] Furthermore, the visual complexity of the view is quantified, and the number of nodes is one of the most direct factors affecting complexity. The current visible area refers to the rectangular range that the user can currently see on the interactive canvas. This takes into account the changes in the view's field of view made by the user through zooming and dragging operations. The total number of nodes actually rendered within the area is counted, rather than the total number of nodes in the entire network, because the user's cognitive processing ability at any given moment is mainly focused on the content within their field of vision.

[0058] Specifically, for example, when a user zooms in to focus on a sub-cluster of the network, the total number of nodes complexity metric decreases significantly; while when the user zooms out to observe the whole picture, the metric increases sharply, indicating that the user is trying to process a larger number of information entities within a single screen. The total number of nodes complexity metric is directly related to the load on visual working memory. Too many nodes presented at the same time force users to expend more effort in visual search, differentiation, and memorization, increasing the cognitive burden. The metric provides a key dimension for subsequent cognitive load assessment regarding the information carrying capacity of the view. By counting the number of rendered nodes in the visible area in real time, the total number of nodes complexity metric, which reflects the density of information entities in the view, is obtained.

[0059] S3.4 During the user's interaction with the initial visual view, the total density of edges rendered by the initial visual view within the current visible area is calculated and used as a complex indicator of total edge density.

[0060] Furthermore, the study focuses on the visual complexity arising from the connections between graphical elements. The total edge density can be defined as the ratio of the total length of edges within a visible area to the area of ​​the visible area, or as a measure of the ratio of the total number of edges to the total number of nodes within a simple visible area. High-density edges can visually form intricate network structures. These intersecting and overlapping lines can severely interfere with a user's ability to trace specific related paths, increasing visual clutter and parsing difficulty.

[0061] Specifically, for example, in a highly interconnected technical knowledge network, core nodes may be surrounded by a large number of related edges, creating a visual "fuzzy ball effect" that makes it difficult for users to quickly identify which connections are key. The total edge density complexity metric complements the total node complexity metric because even with a small number of nodes, a dense network of connections can still result in high visual complexity and parsing difficulty. This metric quantifies the visual burden of a view at the relational connection level. Based on the degree of clustering of edge elements within the currently visible area, the total edge density complexity metric, which measures the visual complexity of a view's relational network, is derived.

[0062] S3.5 During the interaction between the user and the initial visual view, count the number of different visual encoding attributes used by the initial visual view within the current visible area, as an indicator of visual encoding complexity.

[0063] Furthermore, the study identified the impact of visual symbol diversity on cognitive load. Visual encoding attributes include, but are not limited to, color, shape, size, texture, line type, and arrow style. These attributes are used to encode different data dimensions of nodes or edges, such as document type, legal status, association strength, and relationship type. The study counted the number of different visual encoding attributes actually used in the current visible area, reflecting how many categories of visual symbols the user needs to understand and distinguish simultaneously. Too many types of visual encoding will force the user to constantly switch their visual legends to interpret the meaning of different symbols, increasing cognitive conversion costs and memory burden.

[0064] Specifically, for example, if a node simultaneously uses color to represent legal status, shape to represent document type, and size to represent citation count, and edges use color to represent association type and thickness to represent association strength, then the user needs to maintain multiple visual mapping rules in working memory at the same time. The visual coding type complexity index aims to capture this cognitive overhead caused by visual channel overload. By enumerating the different visual coding attribute categories active in the visible area, a visual coding type complexity index that characterizes the complexity of the view symbol system is obtained.

[0065] S3.6. The user interaction load index, the total number of nodes complexity index, the total edge density complexity index, and the visual coding type complexity index are collectively referred to as the user interaction load index and the view complexity index of the initial visualization view.

[0066] Furthermore, by integrating multi-dimensional indicators that characterize user behavior and the objective state of the view into a comprehensive observation framework, we recognize that cognitive load is the result of the combined effect of user subjective effort and the objective difficulty of the view. The user interaction load indicator infers the degree of effort invested and the possible state of confusion from the user's operation behavior. The total number of nodes complexity indicator, the total edge density complexity indicator, and the visual encoding type complexity indicator objectively describe the inherent complexity of the currently presented content from three aspects: the information content, structural complexity, and symbol diversity of the view itself.

[0067] Specifically, these four categories of indicators are collectively referred to as a set of overall indicators. This means that the subsequent cognitive load assessment logic will simultaneously consider how users operate and what they are looking at, thereby achieving a more comprehensive and accurate estimate of the overall cognitive pressure experienced by users in the current specific visualization context. Combining the ideas of user behavior signals and view feature signals, the various indicators are conceptually classified and integrated, completing the definition and preparation of user interaction load indicators and the view complexity indicators of the initial visualization view, providing a complete input dataset for the next step of cognitive load assessment.

[0068] S4. Assess users' real-time cognitive load based on user interaction load metrics and view complexity metrics.

[0069] S4.1. Compare the click frequency, average speed of drag operations, and frequency of zoom operations in the user interaction load indicators with the preset cognitive load benchmark parameters to obtain the first load assessment value.

[0070] Furthermore, user behavior pattern data is transformed into a quantitative estimate representing the cognitive load level reflected by interactive behavior. Pre-defined cognitive load benchmark parameters are a set of reference values ​​or ranges determined based on observations and analyses of typical user interaction behavior under ideal or average conditions. For example, a normal click frequency range, a smooth drag operation average speed range, and a gradual zoom operation frequency level are preset. The comparison process typically involves calculating the differences or deviations between the currently observed click frequency, average drag operation speed, and zoom operation frequency and these benchmark parameters. A high click frequency may indicate that the user is actively and tentatively searching for information, deviating from the normal range, thus contributing a positive load factor.

[0071] Specifically, an average drag speed below the smooth range may indicate that the user is scrutinizing details, while a speed significantly above the smooth range may suggest that the user is quickly skipping information. Both of these deviations from the norm may be associated with higher cognitive load. An abnormally high frequency of zoom operations reflects the user's cognitive effort in frequently switching between macro and micro perspectives. Through this comparison, the degree of deviation of each behavioral indicator is mapped to a sub-score. These sub-scores are then combined into a first load assessment value according to certain rules, such as summation or weighted averaging. By quantitatively analyzing the user's natural interaction behavior, their cognitive state can be indirectly inferred, allowing load assessment to be seamlessly and non-intrusively integrated into the interaction process. After comparing the user's interaction load indicators with preset cognitive load benchmark parameters, the first load assessment value is output, reflecting the cognitive load level indicated by the user's interaction behavior pattern.

[0072] S4.2. Compare the total number of nodes, total edge density, and visual coding type complexity indicators in the view complexity indicators with their corresponding visual complexity thresholds, and then sum them by weight to obtain the second load evaluation value.

[0073] Furthermore, the objective complexity attribute of the visualization itself is transformed into a quantitative indicator of the estimated user cognitive processing burden. The visual complexity threshold is an empirical threshold value set for each view's complexity indicator, representing a certain level at which the indicator may begin to burden the user's visual perception and information processing. The threshold for the total number of nodes complexity indicator corresponds to the theoretical or empirical upper limit of the number of information entities that can be effectively processed simultaneously in working memory; the threshold for the total edge density complexity indicator is related to the limit of the visual system's ability to resolve complex network structures without confusion; and the threshold for the visual encoding type complexity indicator reflects the limitation on the number of visual symbol categories that the user can simultaneously understand and distinguish. The comparison process involves comparing the current total number of nodes complexity indicator, total edge density complexity indicator, and visual encoding type complexity indicator with their respective visual complexity thresholds. The degree to which these indicators approach the threshold reflects the difference in the importance of different complexity indicators in contributing to the overall cognitive load.

[0074] Specifically, for example, the absolute number of nodes may have a greater impact on the burden of basic visual search than the density of edges, so the total number of nodes complexity index may be given a higher weight; while the increase in visual encoding types may lead to switching costs in cognitive legends, and its weight setting may take into account the difficulty of symbol decoding. By weighted summation, the visual complexity impact of the three aspects is integrated into a comprehensive second load assessment value, which objectively assesses the inherent difficulty of information presentation from the source of view presentation, and provides an independent, view feature-based estimate for assessing how much cognitive resources users need to invest to understand the current view. By comparing each view complexity index with the corresponding threshold and weighted synthesis, the second load assessment value is output, which characterizes the cognitive load estimated by the objective visual complexity of the current initial visualization view.

[0075] S4.3. Based on the predefined cognitive load synthesis rules, synthesize the first load assessment value and the second load assessment value, and output the user's real-time cognitive load value.

[0076] Furthermore, predefined cognitive load synthesis rules define how to combine the first and second load assessment values. A simple synthesis rule could be to take the maximum of the two, reflecting the barrel principle, where the overall load is determined by the dimension with the higher load. The synthesis rule might use a weighted average, where the weights can be dynamically adjusted. For example, in the early stages of interaction, view complexity might dominate; while as users enter the deep exploration phase, their interaction patterns might better reflect their true cognitive effort. Another synthesis rule might be based on multiplication or more complex nonlinear functions to simulate the interaction effect between user behavior and view complexity. For example, under high view complexity, even moderate interaction anomalies can lead to a high overall perceptual load.

[0077] Specifically, a user's real-time cognitive load is the result of a dynamic interaction between subjective cognitive ability and objective task difficulty. The first load assessment value captures the behavioral signals exhibited by the user when dealing with the current task, while the second load assessment value characterizes the objective difficulty of the task itself as presented through the view. Relying solely on behavioral signals may be misled by the user's operating habits; relying solely on view features ignores individual differences and the user's current state. By organically combining the two through predefined cognitive load synthesis rules, a more robust and accurate real-time cognitive load estimate can be generated as the core input for subsequent dynamic view scheduling decisions. This allows view optimization to be based on a comprehensive understanding of the user's current cognitive state. The two assessment values ​​are fused according to the predefined cognitive load synthesis rules to output the real-time cognitive load value.

[0078] S5. Based on the real-time cognitive load dynamic scheduling and optimization of the initial visualization view presentation format, generate an optimized view stream.

[0079] S5.1. Compare the real-time cognitive load value with the first load threshold and the second load threshold to determine the current cognitive state of the user.

[0080] Furthermore, the first and second load thresholds are two pre-defined key thresholds. The first load threshold is typically set at a higher level to identify when a user may be experiencing cognitive overload or difficulty processing information. When the real-time cognitive load value exceeds the first load threshold, the user is determined to be in a high-load state. The second load threshold is typically set at a lower level to identify when a user may be in a low-load state, with relatively focused attention or effortless information processing. When the real-time cognitive load value is below the second load threshold, the user is determined to be in a low-load state. Load values ​​between the two thresholds may correspond to a normal or moderate load state.

[0081] Specifically, different intervention mechanisms can be triggered by setting clear boundaries. For example, the first load threshold may be determined based on load observations of a large number of users when they begin to show interaction hesitation or an increase in error rate under a highly complex view. By comparing the real-time cognitive load value with these two thresholds, the user's real-time cognitive state can be clearly classified into a high-load state that requires view simplification intervention, a low-load concentrated state that is suitable for providing support for in-depth exploration, or a medium-load state that does not require active intervention for the time being. By comparing the real-time cognitive load value with the preset threshold, the judgment of whether the user's current cognition is in a high-load state or a low-load concentrated state is completed.

[0082] S5.2 When the real-time cognitive load value exceeds the first load threshold, the view simplification rule is triggered, and the initial visualization view is subjected to operations such as fading secondary related edges, collapsing non-focus area nodes, and hiding some metadata labels.

[0083] Furthermore, proactive, pre-defined information de-loading strategies are employed for high cognitive load states to reduce the visual complexity and information density of the view, thereby helping users restore information processing efficiency. View simplification rules are a set of predefined graphical simplification operations. Diminishing secondary related edges means reducing the visual prominence of edges whose relatedness is below a certain threshold, such as changing their color to light gray or to dashed lines, making them visually recede into the background, thus allowing users to focus more on strong related relationships. Collapsing non-focus area nodes refers to aggregating a group of closely connected nodes outside the current user's focus area into a super node representing a subgroup, which can be expanded when clicked. This significantly reduces the number of basic nodes displayed on the screen at the same time. Hiding some metadata labels refers to temporarily removing non-core text labels on nodes or edges, which are then triggered to appear again by hovering when needed by the user. The common goal is to reduce the number and competition of visually distracting elements and reduce the user's perceptual and cognitive burden.

[0084] Specifically, for example, when a user zooms in to the global view to understand the entire network layout, the large number of nodes and edges may cause the real-time cognitive load to exceed the first load threshold. Automatically triggering these simplification operations can provide a clearer and easier-to-understand overview view without losing overall structural information. Instead of requiring users to manually perform tedious filtering and settings, visual noise reduction services are automatically provided when excessive user load is detected, through the execution of specific graphical operations within the view simplification rules. S5.3 When the real-time cognitive load value is lower than the second load threshold and the interactive operation events are concentrated in a specific area, the view focus rule is triggered to perform the operation of expanding details on the specific area in the initial visualization view.

[0085] Furthermore, by identifying moments when users are low-load and focused, and proactively providing support functions for in-depth analysis, the depth and efficiency of exploration can be improved. View focus rules are another set of preset graphics enhancement operations. Detailed operations include: restoring collapsed supernodes in specific areas where users continuously interact (such as mouse hover or click) to display all their child nodes; increasing the detail of label information for regional nodes and edges; or rendering regions at a larger scale. For example, providing animated highlighting of related paths might mean that when a user selects a node, it automatically highlights several of the most important related paths originating from that node with dynamic flowing light effects or color gradients, thus intuitively revealing its connection network.

[0086] Specifically, the triggering conditions are complex. First, the real-time cognitive load must be below the second load threshold, indicating relatively abundant cognitive resources for the user. Second, interactive events must be concentrated in a specific area, which usually means the user has sustained interest in the content of that area or is exploring it in depth. For example, if a user slowly drags and clicks frequently near a certain technology sub-cluster while the overall cognitive load is low, it suggests the user may be carefully studying the relationships within that cluster. This triggers the view focus rule, automatically expanding the details of the cluster or highlighting its core connections. This adapts to the user's deep analytical intent, providing just the right amount of information support without requiring the user to manually search for and invoke these functions.

[0087] S5.4 The adjusted initial visualization view generated by the view simplification rule or view focus rule is serialized into an optimized view flow.

[0088] Furthermore, each time a view simplification rule or view focus rule is applied, the resulting adjusted view state is recorded in chronological order to form a historical or real-time view state stream. This optimized view stream is not a static final image, but a record of how the view dynamically changes according to the user's cognitive state and interaction intent. The purpose of organizing the adjusted view into an optimized view stream is to maintain the continuity and contextual traceability of the user experience. When the user's cognitive state or interaction focus changes again, new optimization rules are applied from the current view state, making the evolution of the view smooth and natural.

[0089] Specifically, optimizing the view flow provides a complete data foundation for subsequent recording and analysis processes and report generation. By replaying the user's analysis path and the corresponding changes in the view, the results of the aforementioned adaptive adjustments are solidified into a continuous and responsive visualization output process. This transforms the initial visualization view from a static starting point into a dynamic sequence of views that intelligently responds to user states and continuously optimizes its presentation. By serializing and outputting the adjusted views, a coherent optimized view flow that reflects the dynamic optimization process of the views is generated.

[0090] S6. Receive the user's virtual modification instructions for the target standard document, create sand table simulation branches, deduce and compare the structural evolution of the multidimensional related knowledge network caused by each sand table simulation branch.

[0091] S6.1 In the optimized view flow, receive virtual modification instructions for the content of the target standard document and update the corresponding semantic vector of the target standard document.

[0092] Furthermore, it allows users to directly perform hypothetical editing of the text content of technical solutions in a visual analysis environment, and instantly converts such editing into updates to the underlying semantic representation, thereby laying the foundation for subsequent impact deduction. Virtual modification instructions are the intentions of users to change the text of specific parts of the target standard document through interface interaction, such as deleting a technical feature in the scope of technical protection, replacing a description in the background technology, or adding new technical effect descriptions to the document content. The instructions do not actually modify the original document file, but create a temporary document version in memory that represents the modification intention.

[0093] Specifically, upon receiving the instruction, the corresponding semantic vector of the target standard document needs to be updated. This typically involves regenerating the semantic vector of the modified section / block within the target standard document using a pre-trained language model. If the modification affects the overall semantics, the overall text semantic vector may also need to be updated. The semantic vectors of other unaffected section / blocks remain unchanged. By receiving virtual modification instructions and updating the semantic vectors accordingly, a modified semantic representation of the target standard document is created for each sandbox simulation branch.

[0094] S6.2 Using the updated target standard document semantic vector, re-derive the multi-level semantic association degree fusion score of all document nodes in the multi-dimensional association knowledge network, and reconstruct the updated multi-dimensional association knowledge network corresponding to each sand table simulation branch.

[0095] Furthermore, the study simulated the chain reaction of semantic changes in a single document on the overall technological landscape, enabling rapid extrapolation of the global impact of local modifications. A multi-level semantic relevance fusion method was implemented, but the semantic vector of the target standard document in the input was replaced with the updated semantic vector of a specific sandbox simulation branch. For each sandbox simulation branch, its unique updated overall text semantic vector of the target standard document, the semantic vector of the technical protection scope section block, and the semantic vector of the background technology section block were compared and fused again with the section block semantic vector corresponding to each comparison standard document node in the multi-dimensional relevance knowledge network. This generated a new set of multi-level semantic relevance fusion scores reflecting the modified semantic relevance state for each target-comparison document pair. This replaced or updated the weight attributes of the corresponding edges in the original multi-dimensional relevance knowledge network. Based on these updated edge weights, and combined with the invariant nodes and other relationships in the network, a multi-dimensional relevance knowledge network reflecting the new relevance state under the sandbox simulation branch was reconstructed.

[0096] S6.3. In parallel comparison of the updated multidimensional association knowledge network corresponding to each sand table simulation branch with the original multidimensional association knowledge network, the strengthening, weakening and disappearance of the association edges between nodes are identified as the structural evolution of the multidimensional association knowledge network.

[0097] Furthermore, from the dynamic changes in network topology and connection strength, key comparative information that can intuitively and quantitatively reflect the differences in the effects of different modification strategies is extracted. Parallel comparison means that the updated network of all active sandbox simulation branches is analyzed against the original network at the same time. The identification process focuses on the changes in the associated edges between nodes in the network: the strengthening of associated edges means that the weight (association strength) of the same edge in the new network increases significantly; the weakening of associated edges means that the weight decreases; the disappearance of associated edges means that the weight decreases below a certain display threshold or the association score is below a certain threshold, causing the edge to be visually or logically regarded as non-existent. Through comparison, a change list or change graph is generated for each sandbox simulation branch, clearly indicating which comparison documents have become stronger, weaker, or no longer relevant to the target document.

[0098] Specifically, the abstract changes in correlation scores are transformed into concrete and observable changes in network connections. The structural evolution provides a direct data foundation for subsequent pattern and path extraction and strategy evaluation. Through parallel comparison and difference identification, the structural evolution of the multidimensional relational knowledge network triggered by each sand table simulation branch, with the changes in relational edges as the core feature, is output.

[0099] S7. Extract trade-off patterns and potential paths from structural evolution, and provide visual guidance on strategy paths and mark advantageous strategies in the optimization view flow.

[0100] S7.1 Based on the structural evolution of the multidimensional related knowledge network, by comparing the combination of related edge changes caused by different sand table simulation branches, we can summarize the related adversarial mode of increasing and decreasing related strength and the related synergistic mode of weakening multi-sided related strength together to form a trade-off mode.

[0101] Furthermore, by comparing the structural evolution of different sand table simulation branches, it can be found that certain modification strategies will lead to an enhanced correlation between the target document and the comparison document A, while the correlation with the comparison document B will be significantly weakened. This phenomenon of one gaining and the other losing is summarized as the correlation adversarial mode, which reveals a typical trade-off relationship. That is, when modifying to avoid a certain existing technology risk, it may inadvertently make the technical solution closer to another existing technology.

[0102] Specifically, for example, by modifying a basic technical means commonly used in multiple existing technical documents, the similarity with this series of documents can be reduced simultaneously. The significance of the inductive trade-off model lies in sublimating a large amount of edge change data and gaining insight into the inherent laws of modification strategies. This helps users understand the nonlinear and networked impacts brought about by different strategies, rather than just the increase or decrease of a single correlation. Through the regularity analysis of the combination of correlation edge changes in structural evolution, two core trade-off models, correlation adversarial mode and correlation collaborative mode, are identified and defined.

[0103] S7.2. Based on the trade-off model, connect the logically continuous structural evolutions and plan the evolution sequence from the current network state to the low correlation density region as a potential path.

[0104] Furthermore, the static, point-like strategy comparison is extended to dynamic, sequential strategy evolution path planning. This simulates the strategy exploration process of step-by-step modifications to achieve the best results. The planning of potential paths is not a random combination. It is based on the understanding of correlation adversarial mode and correlation synergy mode. For example, a potential path may first suggest a modification that triggers a strong correlation adversarial mode, that is, first significantly weaken the correlation with the core risk document through a high-risk, high-reward change, and then link another modification. This modification aims to alleviate the side effects of the correlation enhancement in the adversarial mode triggered by the first modification. The ultimate goal is to guide the network state to a region with generally low correlation density, that is, a state in which the correlation between the target document and most high-risk comparison documents is effectively weakened.

[0105] Specifically, the planning process assesses the changing trend of network association density after each evolution step to ensure that the entire sequence evolves in the direction of reducing overall risk. The connection logic may be based on heuristic rules, such as prioritizing modifications that generate association and cooperation patterns, or focusing on enhanced association edges in association and adversarial patterns as the next key targets. The identification of potential paths guides users from choosing among numerous parallel strategies to thinking about how to strategically combine modifications in stages to achieve the best obstacle avoidance effect. By logically connecting multiple structural evolutions according to the goal of reducing overall association density, guiding potential paths are planned.

[0106] S7.3 Based on the evolution direction and key nodes of the potential path, generate a dynamically extended semi-transparent guide trajectory that overlays the current view in the optimized view flow to determine the strategy path visualization guidance.

[0107] Furthermore, abstract and logical potential paths are transformed into intuitive and dynamic visual cues, providing users with operational guidance directly within the visualization context. The strategy path visualization guidance is not a simple text description, but rather maps potential paths to graphical elements based on the network structure displayed in the current optimized view flow. The direction of evolution is represented by the extension direction of the trajectory and arrow indicators. Key nodes, i.e. document nodes in the path that require special attention or modification, are emphasized by highlighting, enlarging, or adding special markers when the trajectory passes through them.

[0108] Specifically, dynamic extension means that the guidance trajectory is not displayed all at once, but can unfold gradually like an animation, simulating how the network will change if the first step of modification is made; then how it will change if the second step is made. The semi-transparent visual effect ensures that the guidance trajectory does not completely cover the underlying network view. Users can see the guidance information and the original network state at the same time. Visual guidance seamlessly integrates strategy suggestions into the user's interactive exploration environment, making complex multi-step strategies easy to understand and follow. For example, a guidance trajectory may start from a node representing the current target document, point to a suggested technical feature location for modification, and then extend the trajectory to a region in the network. After the first step of modification, the correlation will weaken, and then the trajectory may turn back to point to another suggested modification point to deal with the new correlation that may have been generated in the first step. By generating and overlaying dynamic semi-transparent guidance trajectories, visual guidance of the strategy path is realized in the optimization view flow.

[0109] S7.4 Apply the preset multi-objective evaluation rules to quantify the performance of the sand table simulation branches in terms of modification costs and risk reduction effects. Branches with scores higher than the preset advantage threshold are identified as advantageous strategies. They are marked with specific visual symbols that distinguish them from ordinary branches in the optimization view flow to determine the advantageous strategy markers.

[0110] Furthermore, a comprehensive and quantifiable evaluation standard is provided for multiple parallel sandbox simulation branches, and the evaluation results are presented with significant visual feedback to help users quickly focus on the optimal option. The preset multi-objective evaluation rules define the dimensions for comprehensively evaluating a strategy, including at least the modification cost and risk reduction effect. The modification cost can be estimated based on the complexity of the virtual modification instructions, the length of the text involved, or the potential impact of the modification on the protection scope of the original technical solution. The risk reduction effect can be measured by calculating the decrease in the sum of the correlation degree with high-risk comparison documents in the multi-dimensional related knowledge network after modification, or the number of missing related edges. The quantitative scoring process combines the indicators of these two dimensions into a comprehensive score according to certain rules (such as weighted summation). The preset advantage threshold is a score threshold used to distinguish between strategies with excellent performance and those with average performance.

[0111] Specifically, the sandbox simulation branches determined to be advantageous strategies are marked with specific visual symbols in the optimization view flow, such as adding a star, a crown icon, or using a special colored border next to their corresponding views or controls. This marking allows users to quickly identify the recommended solutions that achieve the best balance between cost and effect among a variety of strategies.

[0112] S8. Record the scheduling logs of the optimized view flow and the guidance information of the sand table simulation branches, and generate a comprehensive report by combining the multi-dimensional related knowledge network status corresponding to the user's sand table simulation branches.

[0113] S8.1 Capture and store the scheduling log of the optimized view flow. The scheduling log includes the trigger time, reason and specific operation record of the view simplification rule and the view focus rule.

[0114] Furthermore, the system automatically generates a log entry containing key metadata each time a view simplification rule or view focus rule is triggered and executed. The trigger time records the moment the rule takes effect, and the trigger reason records the specific conditions that led to the rule being triggered. For example, it records whether the real-time cognitive load value exceeded the first load threshold, or whether the real-time cognitive load value was below the second load threshold and the interactive events were concentrated in a specific coordinate area of ​​the canvas. The specific operation record describes in detail the specific content of the rule execution, such as which related edges were faded, which area of ​​nodes were collapsed, which labels were hidden, or the details of which specific area was expanded and which related path was highlighted. This constitutes a complete optimized view flow scheduling log, recording why the user sees this view and how the view gradually becomes the current state. This makes the entire analysis process change from static result backtracking to dynamic process reproduction.

[0115] Specifically, for example, the report could explain that in the early stages of the analysis, due to the overload of global network view information, a view simplification was automatically triggered at a certain moment, collapsing the peripheral technology clusters. This explains how the focused view that the user ultimately sees is formed. By capturing and storing scheduling logs containing trigger details, the entire process of optimizing view flow generation and evolution is recorded.

[0116] S8.2. Capture and store the guidance information of the sand table simulation branches. The guidance information includes the trajectory data of the strategy path visualization guidance and the scoring basis of the advantageous strategy mark.

[0117] Furthermore, the decision support information generated during the intelligent analysis process is structured and archived, enabling the machine-generated suggestions and insights to be fully reproduced and understood. The trajectory data of the strategy path visualization guide records the graphical definition information of the semi-transparent guide trajectories displayed to the user in the optimization view flow, such as the starting coordinates of the trajectory, the sequence of key nodes passed through, the ending coordinates, and the order of dynamic extension. The scoring basis of the advantageous strategy mark records the detailed process and intermediate results of applying the preset multi-objective evaluation rules to quantify and score each sandbox simulation branch. For example, it records the specific scores of each branch in the modification cost dimension and risk reduction effect dimension, as well as the calculation logic of the weighted synthesis of the final comprehensive score. The purpose of capturing and storing this guidance information is to preserve the reasoning chain of intelligent analysis.

[0118] Specifically, this allows the final report to not only state that a strategy is recommended, but also explain why it is recommended. That is, what path planning was used and what scoring criteria led to the conclusion that the strategy is an advantageous strategy. For example, the report can include a diagram of the guiding path and explain that this path is recommended because it is modified in two steps: first, it uses the association and collaboration model to generally reduce risk, and then it deals with the remaining association adversarial, ultimately causing the target document to fall into a low association density area. This enhances the interpretability of the analysis process and the persuasiveness of the report's conclusions. By capturing and storing the detailed basis for the strategy path and advantage markers, all the intelligent guidance information generated during the sandbox simulation is preserved.

[0119] S8.3 When a user selects a sand table simulation branch as the final strategy, the updated multidimensional related knowledge network state corresponding to the sand table simulation branch is locked.

[0120] Furthermore, by anchoring the instantaneous results of user decisions to a complete and deterministic analytical snapshot, a unique and accurate data benchmark is provided for generating the final report. The locking operation means fixing and saving the state of the updated multidimensional related knowledge network corresponding to the specific sandbox simulation branch selected by the user, which has been virtually modified and recalculated. This includes the final weights of all nodes and edges in the network, as well as the network's topology. It represents the new position of the target document in the technical knowledge network and its new relationship with other documents after adopting the virtual modification scheme selected by the user. This locked state is the core data source for the subsequent report generation and reflects the final deduplication results after adopting the suggested strategy.

[0121] S8.4 Based on the scheduling log, guidance information, and the updated multidimensional related knowledge network status corresponding to the final strategy, a comprehensive report is output by formatting, integrating, and displaying text and graphics according to the preset report template.

[0122] Furthermore, the system organically integrates scattered process records, intelligent analysis insights, and final data status to automatically generate a deep analysis document that not only includes results but also elucidates the process and rationale. Preset report templates define the report's structure, chapters, and placeholders for embedding various types of information. Formatted integration refers to extracting, transforming, and filling data from captured scheduling logs, guidance information, and locked multidimensional related knowledge network states into the corresponding locations according to the template requirements. For example, key view optimization events recorded in the scheduling logs are organized chronologically into an "Analysis Process Adaptive Adjustment Record" chapter; strategy path trajectory data from guidance information is rendered into images with accompanying descriptions to form a "Strategy Deduction and Intelligent Suggestion" chapter; and the final multidimensional related knowledge network state is rendered into a final association graph, with key risk comparison data extracted to form a "Final Debate Conclusion and Risk Analysis" chapter.

[0123] Specifically, the mixed text and graphics layout ensures that the report not only includes lengthy text descriptions but also key visualizations, such as snapshots of the optimization process, strategy path diagrams, final network state diagrams, and correlation comparison tables. The resulting comprehensive report is a complete analytical product that presents what it is (final state), explains how it came about (optimization process), and why it is (guiding basis). This achieves transparency in the analysis process, traceability of conclusions, and rationalization of decision-making. By integrating and formatting all information according to the template, a comprehensive report containing the entire intelligent analysis process and the basis for the final decision is generated.

[0124] This embodiment also provides a standard document plagiarism detection result visualization system, including: a data acquisition module, which acquires the target standard document and the comparison standard document set, and processes them to generate a structured document object set; The rendering module performs multi-level semantic association fusion based on a set of structured document objects, constructs a multi-dimensional related knowledge network, and renders and generates an initial visual view. The monitoring module monitors user interaction load metrics and the view complexity metrics of the initial visualization view during user interaction with the initial visualization view. The derivation module assesses the user's real-time cognitive load based on user interaction load and view complexity indicators. It dynamically schedules and optimizes the presentation of the initial visualization view based on the real-time cognitive load, generates an optimized view stream, receives the user's virtual modification instructions for the target standard document to create sand table simulation branches, and derives and compares the structural evolution of the multidimensional related knowledge network caused by each sand table simulation branch. The strategy generation module extracts trade-off patterns and potential paths from structural evolution, and provides visual guidance on strategy paths and marks advantageous strategies in the optimization view flow. The comprehensive report generation module records the scheduling logs of the optimized view flow and the guidance information of the sandbox simulation branches. It combines the multi-dimensional related knowledge network status corresponding to the user's sandbox simulation branches to generate a comprehensive report.

[0125] This embodiment also provides a computer device applicable to the standard document plagiarism detection result visualization method, including: a memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions to realize the standard document plagiarism detection result visualization method proposed in the above embodiment.

[0126] The computer device can be a terminal, comprising a processor, memory, communication interface, display screen, and input devices connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The communication interface is used for wired or wireless communication with external terminals; wireless communication can be achieved through Wi-Fi, carrier networks, NFC (Near Field Communication), or other technologies. The display screen can be an LCD screen or an e-ink screen. The input devices can be a touch layer covering the display screen, buttons, a trackball, or a touchpad on the computer device's casing, or an external keyboard, touchpad, or mouse.

[0127] This embodiment also provides a storage medium storing a computer program. When executed by a processor, the program implements the method for visualizing the results of standard document plagiarism detection as proposed in the above embodiments. The storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as Static Random Access Memory (SRAM), Electrically Erasable Programmable Read-Only Memory (EEPROM), Erasable Programmable Read Only Memory (EPROM), Programmable Red-Only Memory (PROM), Read-Only Memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk.

[0128] In summary, this invention constructs a document object set through structured processing and semantic vectorization. Based on this, it builds a multi-dimensional related knowledge network by fusing multi-level semantic correlation to generate an initial visual view. It assesses cognitive load by monitoring user interaction load and view complexity indicators in real time, and dynamically schedules and optimizes the view to generate an optimized view flow. A sandbox simulation mechanism is introduced into the optimized view flow, allowing users to virtually modify the data and deduce the structural evolution of the related network in real time. This allows for the extraction of trade-off patterns and potential paths for visual guidance and strategy labeling. By recording the complete interaction and deduction process and combining it with the network state corresponding to the final strategy, a comprehensive analysis report is generated. This achieves the evolution of plagiarism detection results from static display to dynamic interaction. Through adaptive view optimization and interactive strategy deduction, it effectively improves analysis efficiency and decision support capabilities in complex information environments.

[0129] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the technical protection scope of the present invention.

Claims

1. A method for visually displaying standard document plagiarism detection results, characterized in that: This includes acquiring the target standard document and the set of comparison standard documents, and processing them to generate a collection of structured document objects; Multi-level semantic association is fused based on a set of structured document objects to construct a multi-dimensional associated knowledge network and render an initial visualization view. During user interaction with the initial visual view, monitor user interaction load metrics and the view complexity metrics of the initial visual view; Based on user interaction load indicators and view complexity indicators, the real-time cognitive load of users is evaluated. The presentation form of the initial visualization view is dynamically scheduled and optimized according to the real-time cognitive load, and an optimized view flow is generated. The sand table simulation branches are created by receiving virtual modification instructions from users on the target standard document. The structural evolution of the multidimensional related knowledge network caused by each sand table simulation branch is derived and compared. Extract trade-off patterns and potential paths from structural evolution, and provide strategic path visualization guidance and advantageous strategy marking in the optimization view flow; Record the scheduling logs of the optimized view flow and the guidance information of the sandbox simulation branches, and generate a comprehensive report by combining the multi-dimensional related knowledge network status corresponding to the user's sandbox simulation branches.

2. The method for visualizing the results of standard document plagiarism detection as described in technical protection scope 1, characterized in that: Obtain the target standard document and the set of comparison standard documents, and process them to generate a collection of structured document objects, including the following steps: Receive the target standard document, retrieve the set of comparison standard documents from the comparison document database, parse the document format of the target standard document and the set of comparison standard documents, and extract the clean text; The clean text is analyzed using rules and natural language processing models to divide it into technical field blocks, background technology blocks, and technical protection scope blocks. Using a pre-trained language model, semantic vectors are generated for the overall text, technical field section blocks, background technology section blocks, and technical protection scope section blocks of each document in the target standard document and the comparison standard document set. The plain text, technical field section blocks, background technology section blocks, technical protection scope section blocks, and semantic vectors of each part of each document in the target standard document and comparison standard document sets are encapsulated into a structured document object collection.

3. The method for visualizing the results of standard document plagiarism detection as described in technical protection scope 2, characterized in that: Based on a collection of structured document objects, multi-level semantic association is fused to construct a multi-dimensional associative knowledge network, and an initial visual view is rendered, including the following steps: Document-level relevance is obtained by comparing the overall text semantic vector of the target standard document in the set of structured document objects with the overall text semantic vector of each comparison standard document; The correlation degree at the technical protection scope level is obtained by comparing the semantic vectors of the technical protection scope chapter blocks of the target standard document in the set of structured document objects with the semantic vectors of the technical protection scope chapter blocks of each comparative standard document. The background-level relevance is obtained by comparing the semantic vectors of the background technology section blocks of the target standard document with the semantic vectors of the background technology section blocks of each comparison standard document in the set of structured document objects. By applying preset weighted fusion rules, the document-level relevance, technical protection scope-level relevance, and background-level relevance of the target standard document and each comparison standard document are fused to generate a multi-level semantic relevance fusion score. Using all documents in a structured document object collection as nodes, and multi-level semantic association scores and inter-document reference relationships as edges, a multi-dimensional related knowledge network is constructed. The nodes and edges in the multidimensional knowledge network are encoded according to the preset visual mapping rules, and then rendered on an interactive canvas using a graphics layout algorithm to generate an initial visual view.

4. The method for visualizing the results of standard document plagiarism detection as described in technical protection scope 3, characterized in that: During user interaction with the initial visual view, monitor user interaction load metrics and view complexity metrics of the initial visual view, including the following steps: During the user's interaction with the initial visual view, every click, drag, and zoom event is captured to form interactive event events. Based on interactive events, the click frequency, average speed of drag operations, and frequency of zoom operations per unit time are statistically analyzed as user interaction load indicators. During the user's interaction with the initial visual view, the total number of nodes rendered in the current visible area of ​​the initial visual view is counted and used as a complex indicator of the total number of nodes. During the user's interaction with the initial visual view, the total density of edges rendered by the initial visual view within the current visible area is calculated and used as a complex indicator of total edge density. During the user's interaction with the initial visual view, the number of different visual encoding attributes used by the initial visual view within the current visible area is counted as an indicator of visual encoding complexity. The user interaction load index, the total number of nodes complexity index, the total edge density complexity index, and the visual coding type complexity index are collectively referred to as the user interaction load index and the view complexity index of the initial visualization view.

5. The method for visualizing the results of standard document plagiarism detection as described in technical protection scope 4, characterized in that: Based on user interaction load metrics and view complexity metrics, assess the user's real-time cognitive load, including the following steps: The click frequency, average speed of drag operations, and frequency of zoom operations in the user interaction load indicators are compared with the preset cognitive load benchmark parameters to obtain the first load assessment value. The total number of nodes, total edge density, and visual coding type complexity indicators in the view complexity index are compared with their corresponding visual complexity thresholds, and then weighted and summed to obtain the second load assessment value. Based on predefined cognitive load synthesis rules, the first load assessment value and the second load assessment value are synthesized to output the user's real-time cognitive load value.

6. The method for visualizing the results of standard document plagiarism detection as described in technical protection scope 5, characterized in that: Based on the presentation format of the initial visualization view of real-time cognitive load dynamic scheduling and optimization, an optimized view stream is generated, including the following steps: The real-time cognitive load value is compared with the first load threshold and the second load threshold to determine the current cognitive state of the user. When the real-time cognitive load value exceeds the first load threshold, the view simplification rule is triggered, and the initial visualization view is subjected to operations such as fading secondary related edges, collapsing non-focus area nodes, and hiding some metadata labels. When the real-time cognitive load value is lower than the second load threshold and the interactive events are concentrated in a specific area, the view focus rule is triggered to perform an operation to expand the details of the specific area in the initial visualization view. The adjusted initial visualization, generated by view simplification rules or view focus rules, is serialized into an optimized view flow.

7. The method for visualizing the results of standard document plagiarism detection as described in technical protection scope 6, characterized in that: The process involves receiving virtual modification instructions from users on target standard documents, creating sandbox simulation branches, deriving and comparing the structural evolution of the multidimensional interconnected knowledge network triggered by each sandbox simulation branch, including the following steps: In the optimized view flow, virtual modification instructions for the content of the target standard document are received, and the corresponding semantic vector of the target standard document is updated. Using the updated target standard document semantic vector, the multi-level semantic association degree fusion score of all document nodes in the multi-dimensional association knowledge network is re-derived, and the updated multi-dimensional association knowledge network corresponding to each sand table simulation branch is reconstructed. By comparing the updated multidimensional association knowledge network corresponding to each sand table simulation branch with the original multidimensional association knowledge network in parallel, the strengthening, weakening and disappearance of the association edges between nodes are identified as the structural evolution of the multidimensional association knowledge network.

8. The method for visualizing the results of standard document plagiarism detection as described in technical protection scope 7, characterized in that: Extracting trade-off patterns and potential paths from structural evolution, and visually guiding strategic paths and marking advantageous strategies within the optimization view flow, includes the following steps: Based on the structural evolution of the multidimensional related knowledge network, by comparing the combination of changes in related edges caused by different sand table simulation branches, we can summarize the related adversarial mode of increasing and decreasing related strength and the related synergistic mode of weakening multi-sided related strength together, which together constitute a trade-off mode. Based on the trade-off model, the logically continuous structural evolution is linked together to plan the evolution sequence from the current network state to the low correlation density region as a potential path; Based on the evolution direction and key nodes of the potential path, a dynamically extending semi-transparent guide trajectory is generated over the current view in the optimized view flow to determine the strategy path visualization guidance. By applying preset multi-objective evaluation rules, the performance of the sand table simulation branches in terms of modification costs and risk reduction effects is quantitatively scored. Branches with scores higher than the preset advantage threshold are identified as advantageous strategies and marked with specific visual symbols that distinguish them from ordinary branches in the optimization view flow, thus determining the advantageous strategy markers.

9. The method for visualizing the results of standard document plagiarism detection as described in technical protection scope 8, characterized in that: Record the scheduling logs of the optimized view flow and the guidance information of the sandbox simulation branches. Combine this with the multi-dimensional related knowledge network status corresponding to the user's sandbox simulation branches to generate a comprehensive report, including the following steps: Capture and store the scheduling logs of the optimized view flow, which include the trigger time, reason and specific operation records of view simplification rules and view focus rules; Capture and store guidance information for sand table simulation branches, including trajectory data of strategy path visualization guidance and scoring criteria for advantageous strategy marking; When a user selects a sand table simulation branch as the final strategy, the updated state of the multidimensional related knowledge network corresponding to the sand table simulation branch is locked. Based on the scheduling logs, guidance information, and the updated multidimensional related knowledge network status corresponding to the final strategy, a comprehensive report is output by formatting, integrating, and displaying text and graphics according to a preset report template.

10. A standard document plagiarism detection result visualization system, based on the standard document plagiarism detection result visualization method described in any one of technical protection scopes 1 to 9, characterized in that: This includes a data acquisition module, which acquires the target standard document and the comparison standard document set, and processes them to generate a collection of structured document objects; The rendering module performs multi-level semantic association fusion based on a set of structured document objects, constructs a multi-dimensional related knowledge network, and renders and generates an initial visual view. The monitoring module monitors user interaction load metrics and the view complexity metrics of the initial visualization view during user interaction with the initial visualization view. The derivation module assesses the user's real-time cognitive load based on user interaction load and view complexity indicators. It dynamically schedules and optimizes the presentation of the initial visualization view based on the real-time cognitive load, generates an optimized view stream, receives the user's virtual modification instructions for the target standard document to create sand table simulation branches, and derives and compares the structural evolution of the multidimensional related knowledge network caused by each sand table simulation branch. The strategy generation module extracts trade-off patterns and potential paths from structural evolution, and provides visual guidance on strategy paths and marks advantageous strategies in the optimization view flow. The comprehensive report generation module records the scheduling logs of the optimized view flow and the guidance information of the sandbox simulation branches. It combines the multi-dimensional related knowledge network status corresponding to the user's sandbox simulation branches to generate a comprehensive report.

Images