A content intelligent referencing method for document editing

Abstract

This invention provides a method for intelligent content referencing in document editing, relating to the technical field of intelligent document referencing technology. The method includes: displaying a source document selection interface in response to a user's insertion operation in the main document; displaying at least one page of the source document in response to the user's selection of the source document; determining the selected target page in response to the user's selection of the page; parsing the document structure of the target page, identifying and visually displaying at least one content block within the page; recommending a corresponding insertion method based on the type of the content block; and inserting the content block into the main document according to the selected insertion method in response to the user's selection of the content block and confirmation of the insertion method. This invention solves the problem of low content referencing efficiency, thereby improving the efficiency and accuracy of content referencing.

Description

Technical Field

[0001] This invention relates to the field of automatic document editing technology, and more specifically, to a method for intelligent content citation in document editing. Background Technology

[0002] In modern office and knowledge work, document writers often need to reference content from multiple existing documents to integrate information, support arguments, or reuse existing results.

[0003] Existing document editing tools generally suffer from problems such as cumbersome operation processes, coarse granularity of citation, poor format compatibility, and low level of intelligence when providing cross-document content citation methods, which significantly restrict work efficiency and document quality. Summary of the Invention

[0004] This invention provides a method for intelligent content citation in document editing, which at least solves the problem of low content citation efficiency in related technologies.

[0005] According to an embodiment of the present invention, a method for intelligent content citation in document editing is provided, comprising: In response to the user's insert operation in the main document, the source document selection interface is displayed; In response to the user's selection of the source document, at least one page of the source document is displayed in a paginated preview mode; In response to the user's selection operation on the page, the selected target page is determined; The target page is parsed to analyze its document structure, and at least one content block within the page is identified and visualized. The content block includes at least one of paragraphs, tables, charts, images, and presentation content blocks. Based on the type of the content block, a corresponding insertion method is recommended. The insertion method includes at least one of copying content, inserting a reference block, inserting an image snapshot, and inserting a link. In response to the user's selection of the content block and confirmation of the insertion method, the content block is inserted into the main document according to the selected insertion method.

[0006] In one exemplary embodiment, identifying content blocks within the page includes: Based on at least one piece of information from the document, such as paragraph marks, table marks, chart titles, and list structure, content block boundaries are defined. Each presentation document's slides are parsed as independent content block units, and at least one content block from text boxes, images, charts, and tables within each slide is identified.

[0007] In one exemplary embodiment, the visualization includes: The selectable area of ​​each content block is marked using at least one of the following methods: highlighted border, semi-transparent overlay, or mouse hover preview.

[0008] In one exemplary embodiment, the source document includes at least one of PDF document, Word document, Excel document, image, and presentation document. Different types of source documents are uniformly parsed into paginated preview view and content blocks; wherein, each slide of the presentation document is displayed and selected as an independent content block unit.

[0009] In one exemplary embodiment, recommending the corresponding insertion method based on the type of content block includes: For text paragraphs, it is recommended to copy the content or quote the block; For tables, it is recommended to copy the content; for charts or images, it is recommended to insert an image snapshot.

[0010] In one exemplary embodiment, the automatic recommendation of the corresponding insertion method based on the type of content block further includes: The system records the user's final choice of insertion method for each content block type, constructs a probability matrix, and dynamically updates the recommendation priority through an exponentially weighted moving average.

[0011] In one exemplary embodiment, the insertion method includes at least one of copying content, inserting a reference block, inserting an image snapshot, and inserting a link; wherein, copying content preserves the editability of the content block, inserting a reference block preserves the referencing relationship, and inserting an image snapshot preserves the original layout style.

[0012] In an exemplary embodiment, when a user selects multiple content blocks in the same slide of a presentation document at the same time, the system obtains the bounding box coordinates of each content block, calculates the relative position, and creates a floating canvas in the target document to render each content block onto the canvas according to its original relative position, and outputs it as a vector graphic or a first combined object.

[0013] According to yet another embodiment of the present invention, a computer-readable storage medium is also provided, wherein a computer program is stored therein, wherein the computer program is configured to perform the steps in any of the above method embodiments when executed.

[0014] According to yet another embodiment of the present invention, an electronic device is also provided, including a memory and a processor, wherein the memory stores a computer program and the processor is configured to run the computer program to perform the steps in any of the above method embodiments.

[0015] This application automatically deconstructs the page into units with independent semantics, such as paragraphs, tables, and charts, enabling users to directly select and reference the core information they need with unprecedented precision. It achieves an adjustment from "page-level" to "semantic block-level," and in particular solves the problem of difficulty in finely referencing content in presentations (PPT), as well as the problem of low content referencing efficiency, thus improving the efficiency of content referencing. Attached Figure Description

[0016] Figure 1 This is a structural block diagram of a document editing intelligent content referencing system according to an embodiment of the present invention; Figure 2 This is a flowchart of a method for intelligent content referencing in document editing according to an embodiment of the present invention. Detailed Implementation

[0017] The technical solutions of the embodiments of this application will be described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments.

[0018] In the following description, the terms "first," "second," etc., are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Therefore, a feature defined with "first," "second," etc., may explicitly or implicitly include one or more of that feature. In the description of this application, unless otherwise stated, "a plurality of" means two or more.

[0019] Furthermore, in this application, directional terms such as "upper," "lower," "left," and "right" may be defined relative to the orientation of the components shown in the accompanying drawings. It should be understood that these directional terms can be relative concepts, used for relative description and clarification, and may change accordingly depending on the orientation of the components in the accompanying drawings.

[0020] In this application, unless otherwise expressly specified and limited, the term "connection" should be interpreted broadly. For example, "connection" can be a fixed connection, a detachable connection, or an integral part; it can be a direct connection or an indirect connection through an intermediate medium. Furthermore, the term "coupled" can refer to an electrical connection that enables signal transmission.

[0021] As used herein, “about,” “approximately,” or “approximately” includes the stated value and the average value within an acceptable range of deviation from the given value, wherein the acceptable range of deviation is determined by a person skilled in the art taking into account the measurement under discussion and the error associated with the measurement of the given quantity (i.e., the limitations of the measurement system).

[0022] Example 1 This embodiment provides an intelligent content referencing system for document editing. The system comprises a five-layer architecture: a source document parsing layer, a pagination rendering layer, a content block recognition layer, a user interaction layer, and an insertion execution layer. It can perform unified parsing and visual preview of source documents in various formats. The system is suitable for automatically identifying and allowing users to precisely select semantic content blocks such as paragraphs, tables, and charts within a page. Based on content type, user behavior, and context, it intelligently recommends insertion methods to solve the technical problems of cumbersome operation processes and coarse referencing granularity in existing technologies for cross-document content referencing, thereby effectively improving the efficiency and accuracy of content referencing.

[0023] Reference Figure 1 The system includes: a source document selection module 10, a pagination preview module 20, a content block recognition module 30, an intelligent recommendation module 40, and a content insertion module 50.

[0024] Reference Figure 2 The intelligent content referencing method executed by this system specifically includes the following steps: Suppose you are editing a Word document "Project Griffin - Phase II DiagnosticReport.docx" whose embedded metadata contains the project ID PROJ-GRF-02.

[0025] Source document 1 (PDF) is a 150-page document titled "Griffin Engine G-750 Technical Manual.pdf", with a key performance parameter table on page 88. Source document 2 (PPTX) is a 20-page document titled "Project Griffin - Kick-off Meeting.pptx", with a project milestone chart and concluding text on slide 12. The following steps should then be performed: S100: In response to the user's insert operation in the main document, display the source document selection interface. After a user triggers an OnClick event, the source document selection module 10 is activated. The source document selection module 10 renders the user interface in the editor sidebar, which is divided into three areas: "Local Files", "Project Space" and "Smart Recommendation".

[0026] In the "Intelligent Recommendation" section, a context-aware recommendation algorithm is executed. The input to this algorithm is a feature vector generated from the parsing of the main document's context. , The generation process includes: (1) Read the metadata field project_id and get the value PROJ-GRF-02; (2) The document type is determined to be "diagnostic report" by template matching or filename heuristic rules; (3) The document text is extracted by a TF-IDF (Term Frequency-Inverse Document Frequency) processor to generate a weighted keyword list containing terms such as “Griffin Engine”, “G-750”, “performance”, and “fault”.

[0027] Based on this feature vector The system sends an HTTP POST request to the knowledge base's search engine API endpoint, and the request body contains... The server-side search engine searches for each indexed document in the database. Iteratively calculate a comprehensive relevance score. :

[0028] Among them, weight , , The values ​​are preset to [0.4, 0.3, 0.3], and their sum is 1, where the weight parameters are... The initial baseline value can be determined by the following method: the system pre-collects a log dataset containing cross-document referencing behavior of millions of users offline, and optimizes it using a grid search algorithm with the objective function of maximizing the mean accuracy (mAP) of retrieval, thereby obtaining the initial baseline value; in order to address the differences in feature emphasis when referencing different document types, an adaptive weight adjustment unit can be used for adaptive adjustment. This unit is preloaded with a two-dimensional cross-format weight correction matrix. .when The main document type and candidate documents represented in the middle When the types of documents change, the weights are dynamically reallocated. For example, when the main document is a Word document and the candidate documents are... When using Excel, the system determines that the user prefers to find specific data rather than reports of similar projects. In response to this determination, the weights are dynamically adjusted. (Reduce metadata weight) (Increase the weight of precise word matching) And so on.

[0029] The metadata similarity is a binary function whose input is the item ID of D_i and the... If the two strings match exactly, output 1; otherwise, output 0.

[0030] For word similarity, it can be calculated The cosine similarity between the keyword vector and the pre-stored keyword vector of D_i is used to derive the result.

[0031] For semantic similarity, it can be achieved by... The title and abstract are input into a pre-trained BERT model, generating a 768-dimensional semantic embedding vector, and then... The pre-computed semantic embedding vectors are subjected to a dot product operation, and the result is normalized to the [0,1] interval.

[0032] For example, for source document 1 "Technical Manual.pdf", its metadata does not contain a project ID. The value is 0; however, its content is highly relevant to the keywords. It is 0.95. The value is 0.80. For source document 2 "Kick-off Meeting.pptx", its project ID matches. The score is 1; content relevance is the second highest. It is 0.60. The value is 0.75. The final score is calculated as follows:

[0033]

[0034] The system will sort the returned document list in descending order of score and render the clickable links of the top 5 documents on the UI, so users can directly select "Technical Manual.pdf" from the recommended list, and so on.

[0035] S200: In response to the user's selection of the source document, display at least one page of the source document in a paginated preview mode. After the user selects "Technical Manual.pdf", the pagination preview module 20 is activated. The PDF parsing engine inside this module starts in an isolated sandbox environment in the client browser to load the document data stream. This engine is built on WebAssembly.

[0036] The engine executes two independent tasks in parallel: Rendering Task: This task is responsible for rasterizing each page of the PDF content into a bitmap. To optimize initial screen load time, a lazy loading and tiered rendering strategy can be used. Specifically: The engine first quickly traverses all 150 pages, generating a low-resolution PNG thumbnail with a width of 200 pixels for each page. These thumbnails are immediately pushed to the front-end UI and displayed in a scrollable grid view. Only when the user's scrollbar stops and a thumbnail of a page enters the viewport are high-priority tasks scheduled to generate a high-resolution view of that page with a width of 1024 pixels.

[0037] Structured parsing task: This task is responsible for building the Document Object Model (DOM) for each page. Specifically, the parser iteratively processes the page content stream to extract the Unicode code point of each text character, precisely... Geometric information of objects such as coordinates, font attributes (name, point value, weight, color), vector paths, and raster images.

[0038] For example, for page 88, the output of this task is a structured JSON object `page_88`, which contains all the underlying data for that page, laying the data foundation for subsequent content block recognition. Users can quickly locate and select page 88 by scrolling through the thumbnails and using the magnifying glass preview function when hovering the mouse.

[0039] S300: Perform document structure analysis on the target page, identify and visualize at least one content block within the page. After page 88 is selected, the content block recognition module 30 is triggered, which performs the following steps: S310: Execute content block detection algorithm For PDF pages, the content block recognition module 30 executes an algorithm based on layout and heuristic rules. The input to this algorithm is the DOM data of page_88, and then the following detection is performed: Paragraph block detection: Algorithm iteratively compares adjacent text nodes and font point value and weight The rule base defines font priority conditions: if and (Bold) indicates that the judgment is correct. This is a separate title block; it should be prevented from interacting with the text nodes below it. Merge them into the same paragraph block.

[0040] For example, the page title "Table 5-2" has a point value of 16 and is bold, while the table text below has a point value of 10. Because... Based on the heuristic rules, the system generates a block isolation boundary at this point.

[0041] Table Block Detection: The algorithm first scans the page for a large number of text elements aligned in a grid in two-dimensional space. Simultaneously, it performs a Hough transform to examine the page's vector path data, determining if a set of mutually perpendicular and equally spaced lines exists, which is typically a strong feature of table borders. Based on the text alignment grid and the detected borders, the algorithm regresses the outer bounding box of the table and marks it as a table content block.

[0042] For presentation (PPT) documents, due to their highly flexible layout, the system performs the following detections using a deep learning-based object detection network: Presentation content block detection: The system deploys a detection model based on CSPDarknet53 as the backbone and coupled with a Feature Pyramid Network (FPN) as the neck. In the offline phase, this model is pre-trained using a hybrid reconstructed dataset, which is obtained by crawling publicly available document layout analysis datasets (such as PubLayNet) and sorting them according to... The slides were created by combining 200,000 high-resolution slide renderings, which were obtained from the company's historical PPT library through manual bounding box annotation.

[0043] Classification label setting: The target classification label for the model prediction head is a set of 6 mutually exclusive classes: {Title_Text,Body_Text,Table,Chart,Image,Footer_Header}.

[0044] Inference and Regression: After the RGB image of slide 12 is normalized and input into the network, the multi-task prediction head performs regression on feature maps at different scales. The model then outputs a series of candidate bounding boxes and their confidence scores. The system applies Non-Maximum Suppression (NMS) with an Intersection over Union (IoU) threshold of 0.5 to eliminate duplicate detections of the same object. For example, for slide 12, the model's final output is: {class: 'Chart', bbox: [100, 150, 1080, 500], confidence: 0.98} and {class: 'Body_Text', bbox: [100, 680, 1080, 80], confidence: 0.99}.

[0045] S320: Visual display of optional content blocks Based on the S310's recognition results, the system overlays a rendering layer on the high-definition view on page 88. The bounding box area of ​​each recognized content block is covered by a semi-transparent blue overlay (CSS rgba(0, 0, 255, 0.1)) and a dashed border. When the user's mouse pointer enters the bounding box of any content block, the OnMouseEnter event is triggered to update the block's border style to a solid line and display a tooltip containing a preview of the content. When the user clicks on a table area, the border of that area turns a striking orange, indicating that it is selected.

[0046] S400: Automatically recommends the appropriate insertion method based on the type of content block. Once the table block is selected, the intelligent recommendation module 40 is activated.

[0047] The decision-making logic of this module is a three-layer model: Rule baseline layer: For the "table" type, the default rule recommends "copy content (editable table)".

[0048] User behavior learning layer: The system queries the user's personalized behavior model. For a new user, the probability matrix of their insertion method is... Initialized. For example, for content type... Its corresponding probability vector for These correspond to [copy content, image snapshot, quote block, link], respectively. Therefore, "copy content" is marked as "recommended" by the UI as the most likely option.

[0049] However, to preserve the complex formatting, the user manually selected "Insert Image Snapshot". This selection event was recorded, and the user's choice was encoded into a one-hot vector. The system then applies an exponentially weighted moving average algorithm with a preset learning rate. Update the probability matrix:

[0050]

[0051] Among them, learning rate The range of values ​​is limited to a closed interval. Within this range, the basis for setting this interval is: if The system's learning is too slow, failing to respond promptly to users' preferences for formatting specific projects; if The system may experience sudden probability shifts due to a single misoperation or accidental selection, leading to a loss of recommendation stability. After this update, The probability jumps from 0.15 to 0.32.

[0052] Context-aware layer: The system analyzed the main document's filename "Diagnostic Report.docx" and found no strong contextual keywords such as "Draft" or "Final." Therefore, this layer did not intervene in the recommendation results. The user ultimately confirmed the "Insert image snapshot" option.

[0053] S500: In response to the user's selection of the content block and confirmation of the insertion method, the content block is inserted into the main document according to the selected insertion method. The insertion operation is performed here through the content insertion module 50. Specifically: S510: Perform the specific insertion action For the "Insert Image Snapshot" command, the system uses the bounding box coordinates (bbox: [50, 120, 700, 500] in pixels) of the table content block to perform a precise cropping operation on the high-resolution rendered image on page 88 using an image processing library (such as ImageMagick) on the server side. The resulting 700x500 pixel PNG image, after Base64 encoding, is returned to the client via the API response and used as a snapshot. The element is inserted into the DOM at the current cursor position in the Word document.

[0054] S520: Handling the grouping and insertion of content blocks in a presentation (PPT) For example, after completing the aforementioned operations, the user returns to S100 and selects "Kick-offMeeting.pptx" from the recommendation list. In S200 and S300, the system performs deep learning-based object detection on the 12th slide, identifying the chart and text box. In the visualization interface of S320, the user holds down the Ctrl key and selects both the chart block and the conclusion text box below it.

[0055] The system detected a multi-selection event and recommended "Insert Combined Object (Vector)". After user confirmation, the system retrieved the bounding box coordinates (in pixels) of the two blocks: chart:

[0056] Text box:

[0057] The system constructs an SVG (Scalable Vector Graphics) object in memory, with its canvas size set to just fit the two objects. Pixels. The chart is rendered as a PNG and encapsulated in an SVG. Within the tag, the content of the text box is converted to SVG.<text>The tag retains the original font and size attributes. The generated SVG XML string is inserted into the main document.

[0058] Fault tolerance mechanism: Before insertion, the system queries the target editor's capabilities set via a JS interface. If `editor.capabilities.supportsSVG` returns false, the method will automatically fallback, and the entire capability set will be inserted on the server side instead. The combined area of ​​pixels is rendered into a single high-resolution PNG image for insertion.

[0059] S530: Deep integration with the knowledge base and maintenance of reference relationships Each successful insert operation creates an edge or record in the backend reference relational database (a graph database or a relational database).

[0060] For example, after inserting a table image snapshot, a new record is added to the database: INSERT INTO citation_map (source_doc_uuid, source_element_hash,target_doc_uuid, target_element_id) VALUES ('uuid-pdf-manual', 'md5(content_of_table)', 'uuid-docx-report', 'img-12345'); `source_element_hash` is the value obtained by serializing and MD5 hashing the original table data, used for accurate version comparison. When the "Technical Manual.pdf" file in the knowledge base is updated, a file monitoring service (such as inotify) triggers a hook that recalculates the hash value of the table content on page 88 and compares it with the record in the database. If the hash values ​​do not match, the system pushes a JSON-formatted notification to the client editing "Diagnostic Report.docx" via a WebSocket channel. Upon receiving the notification, the front-end UI renders a prompt asking the user whether to synchronize. If the user confirms, the system will replace the old image with the ID `img-12345` in the document with the new content.

[0061] Example 2 Based on Example 1, this embodiment also includes a semantic-based cross-table data referencing and reorganization mechanism for specific scenarios of financial analysis and report writing.

[0062] For example, a user is editing a "Q3 Quarterly Business Line Profitability Analysis Report.docx" and needs a summary table to show the "revenue", "cost" and "gross profit margin" indicators for each business line (A, B, C).

[0063] The source document (Excel) is an Excel workbook named "Q3 Original Financial Data.xlsx", and its "Profit and Loss Statement" worksheet contains a huge financial statement.

[0064] In response, the following operations are performed in steps S300 and S500: S301: Semantic Recognition of Table Content and Sub-block Segmentation After the user selects the "Profit and Loss Statement" worksheet in S200, the content block recognition module 30 starts a semantic analysis submodule specifically designed for tables.

[0065] Header and Index Recognition: This submodule automatically identifies row headers (usually in column A) and column headers (usually in row 1 or 2) of a table by combining NLP and layout analysis algorithms, and constructs an internal two-dimensional semantic index for each cell. Mapped to a semantic coordinate: ( ).

[0066] Semantic Concept Normalization: The system loads a thesaurus of financial terms to normalize the identified table header labels. This dictionary is constructed as follows: the system automatically retrieves the Extensible Business Reporting Language (XBRL) classification standard document published by the Securities and Exchange Commission (SEC) and extracts standard financial element labels; simultaneously, it uses a Word2Vec word embedding model to perform unsupervised training on a corpus of millions of historical financial statements from enterprises, calculating word similarity to perform K-Means clustering, mapping non-standard words within the same cluster (such as "main business revenue" and "sales revenue") to baseline classification labels (such as "revenue"). This is then reviewed and solidified by domain experts into a structured dictionary. Based on this dictionary, the system accurately maps the various messy table headers identified in the current table to the standard concept space. For example, "operating revenue" and "main business revenue" are mapped to the standard concept "revenue". At the same time, a rule engine is loaded, defining the calculation formulas for derived indicators, such as... .

[0067] Sub-block visualization: In S320, when the system enters "sub-block selection mode", the headers of rows or columns become clickable objects. At this time, the user selects two rows normalized to "Revenue" and "Cost", as well as three columns normalized to "Business Line A", "Business Line B", and "Business Line C".

[0068] S501: Intelligent Table Restructuring and Derivative Calculation Insertion Based on Semantic Selection When a user confirms the newly recommended method of "copying content (intelligent table reorganization)" in S400, the content insertion module 50 executes the following process: Data Extraction and Pivot: Based on user-selected semantic tags, the system accurately extracts corresponding data subsets from the data matrix of the source Excel spreadsheet. Simultaneously, it performs a pivot operation to summarize and sum the data from July, August, and September according to business lines.

[0069] Derived indicator calculation: Based on the rule engine, the system automatically calculates a new row of "gross profit margin" data using the extracted "revenue" and "cost" data.

[0070] Generate and insert the target table: Finally, the system generates and inserts a new table into the target Word document, which contains all the necessary original and derived data.

[0071] Example 3 Based on the above, this embodiment also includes a citation content verification and text consistency check mechanism based on Natural Language Understanding (NLU) for fields such as law, contracts, and patents where the accuracy of citations is extremely important.

[0072] For example, a user is drafting an "Infringement Complaint.docx". The complaint needs to accurately quote the claims of the infringed patent (source document 1) and the technical description in the defendant's product specification (source document 2). Source document 1 (Word) is a "ZL000 Invention Patent Grant Announcement Text.docx". Source document 2 (PDF) is a "Defendant's Product Specification.pdf".

[0073] The following operations are then performed in steps S300 and S500: S302: Semantic verification of referenced content and extraction of entity relationships When a user selects a source document in S200 and a content block (e.g., the full text of claim 1 of a patent) in S320, the system not only recognizes it as a "paragraph block" but also automatically activates a semantic verification submodule, which performs the following operations: Named Entity Recognition (NER): This submodule uses a pre-trained BERT-CRF model for the legal domain to perform named entity recognition on selected content blocks of text. It can effectively recognize technical terms (such as "RF transceiver module"), key parameters (such as "frequency range 2.4GHz to 2.5GHz"), limiting conditions (such as "under microcontroller control"), and legal terms (such as "including" and "at least").

[0074] Relationship Extraction: Building upon NER, the system further extracts relationships between entities. For example, it can construct a knowledge graph triple: ('RF transceiver module', 'has_parameter', 'frequency range 2.4GHz to 2.5GHz').

[0075] Verification Summary Generation: The system will generate a structured "verification summary" from all extracted entities and relationships, and display it next to the visual preview in S320. For example, when the user selects claim 1, a summary box will pop up next to it, clearly listing: "Technical features: radio frequency transceiver module, microcontroller; parameters: 2.4GHz-2.5GHz; subordinate relationship: microcontroller → control → radio frequency transceiver module", which allows the user to quickly identify the core technical points of the content block before citing it.

[0076] S502: Post-insertion consistency scan and risk alert Once the user confirms the insertion in S400 (usually via "copy content"), the content is inserted into the main document. At this point, the system immediately starts a consistency scan and verification process in the background.

[0077] Citation bias detection: The system performs precise character-level and semantic-level comparisons between the text inserted into the main document and the original text extracted from the source document.

[0078] Character-level comparison: Detects for any unintentional additions, deletions, or modifications, such as typos or punctuation errors.

[0079] Semantic-level comparison: The system employs a fine-tuned Sentence-BERT dual-tower model to map two text segments into fixed-length feature vectors and calculate cosine similarity. The threshold setting logic for similarity comparison is not a single checkpoint, but a two-interval stepped judgment logic derived from receiver operating characteristic curve (ROC) optimization. Specifically, a security threshold is set. and warning threshold .like If the system determines that the sentence structure adjustment is legal (such as voice conversion), the operation is allowed; otherwise... The system triggers a "weak warning," generating a yellow alert in the UI sidebar to indicate a potential risk of synonym substitution; if If the system determines that the core technical features have undergone substantial deviation (such as parameter range being tampered with or key steps being omitted), it will trigger a "strong warning", render a red wavy line under the target text and force the user to review it, and so on.

[0080] Global Terminology Consistency Validation: This is the core innovation of this embodiment. The system maintains a dynamically built "terminology-definition" knowledge base within the current main document (the complaint). Whenever a technical term is first referenced or written, the system records its context and usage. When the same term appears again elsewhere in the document, the system verifies whether its usage is consistent with the initial definition.

[0081] For example, in the first part of the complaint, the user cited claim 1, in which the "RF transceiver module" was defined as operating at 2.4 GHz. Subsequently, in the third part of the complaint, the user, in analyzing the defendant's product, may have mistakenly written that "the RF transceiver module in the defendant's product also operates at 5.8 GHz...".

[0082] At this point, the conformance scan process will detect that, in the same complaint, the term "radio frequency transceiver module" is given two contradictory parameters (2.4 GHz and 5.8 GHz).

[0083] Risk warnings and intelligent correction suggestions: When any deviation or inconsistency is detected, the system will not make automatic corrections, but will highlight the corresponding text in the main document with a wavy line and generate a warning in the "Writing Assistant" panel in the sidebar.

[0084] For citation discrepancies, the alert might read: "Alert: This sentence differs from the source patent text." A "Restore Original Text with One Click" button would also be provided.

[0085] For inconsistencies in terminology, the alert might read: "Risk: The term 'RF transceiver module' has an inconsistent definition in this document (defined as 2.4GHz on page 3 and 5.8GHz on page 15). Please check if it is a typo." It may also provide correction suggestions, such as "Should 'RF transceiver module' be corrected to 'Wi-Fi module' here?", and so on.

[0086] Through the above description of the embodiments, those skilled in the art can clearly understand that the methods according to the above embodiments can be implemented by means of software plus necessary general-purpose hardware platforms. Of course, they can also be implemented by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk) and includes several instructions to cause a terminal device (which may be a mobile phone, computer, server, or network device, etc.) to execute the methods described in the various embodiments of the present invention.

[0087] It should be noted that the above modules can be implemented by software or hardware. For the latter, they can be implemented in the following ways, but are not limited to: all the above modules are located in the same processor; or, the above modules are located in different processors in any combination.

[0088] Embodiments of the present invention also provide a computer-readable storage medium storing a computer program, wherein the computer program is configured to perform the steps in any of the above method embodiments when executed.

[0089] In one exemplary embodiment, the aforementioned computer-readable storage medium may include, but is not limited to, various media capable of storing computer programs, such as a USB flash drive, read-only memory (ROM), random access memory (RAM), portable hard disk, magnetic disk, or optical disk.

[0090] Embodiments of the present invention also provide an electronic device including a memory and a processor, the memory storing a computer program and the processor being configured to run the computer program to perform the steps in any of the above method embodiments.

[0091] In one exemplary embodiment, the electronic device may further include a transmission device and an input / output device, wherein the transmission device is connected to the processor and the input / output device is connected to the processor.

[0092] Through the above description of the embodiments, those skilled in the art can clearly understand that, for the sake of convenience and brevity, only the division of the above functional modules is used as an example. In actual applications, the above functions can be assigned to different functional modules as needed, that is, the internal structure of the device can be divided into different functional modules to complete all or part of the functions described above.

[0093] In the several embodiments provided in this application, it should be understood that the disclosed apparatus and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of modules or units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another apparatus, or some features may be ignored or not executed. Furthermore, the mutual coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between apparatuses or units may be electrical, mechanical, or other forms.

[0094] The units described as separate components may or may not be physically separate. A component shown as a unit can be one or more physical units; that is, it can be located in one place or distributed in multiple different locations. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0095] Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0096] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a readable storage medium. Based on this understanding, the technical solutions of the embodiments of this application, in essence, or the parts that contribute to the prior art, or all or part of the technical solutions, can be embodied in the form of a software product. This software product is stored in a storage medium and includes several instructions to cause a device (which may be a microcontroller, chip, etc.) or processor to execute all or part of the steps of the methods of the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0097] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any changes or substitutions within the technical scope disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.< / text>

Claims

1. A method for intelligent content referencing in document editing, characterized in that, include: In response to the user's insert operation in the main document, the source document selection interface is displayed; In response to the user's selection of the source document, at least one page of the source document is displayed in a paginated preview mode; In response to the user's selection operation on the page, the selected target page is determined; The target page is parsed to analyze its document structure, and at least one content block within the page is identified and visualized. The content block includes at least one of paragraphs, tables, charts, images, and presentation content blocks. Based on the type of the content block, a corresponding insertion method is recommended. The insertion method includes at least one of copying content, inserting a reference block, inserting an image snapshot, and inserting a link. In response to the user's selection of the content block and confirmation of the insertion method, the content block is inserted into the main document according to the selected insertion method.

2. The method according to claim 1, characterized in that, Identifying content blocks within the page includes: Based on at least one piece of information from the document, such as paragraph marks, table marks, chart titles, and list structure, content block boundaries are defined. Each presentation document's slides are parsed as independent content block units, and at least one content block from text boxes, images, charts, and tables within each slide is identified.

3. The method according to claim 1, characterized in that, The visualization includes: The selectable area of ​​each content block is marked using at least one of the following methods: highlighted border, semi-transparent overlay, or mouse hover preview.

4. The method according to claim 1, characterized in that, The source documents include at least one of the following: PDF documents, Word documents, Excel documents, images, and presentation documents. Different types of source documents are uniformly parsed into paginated preview views and content blocks; among them, each slide of the presentation document is displayed and selected as an independent content block unit.

5. The method according to claim 1, characterized in that, The method of recommending the corresponding insertion method based on the type of content block includes: For text paragraphs, it is recommended to copy the content or quote the block; For tables, it is recommended to copy the content; for charts or images, it is recommended to insert an image snapshot.

6. The method according to claim 5, characterized in that, The automatic recommendation of the corresponding insertion method based on the type of content block also includes: The system records the user's final choice of insertion method for each content block type, constructs a probability matrix, and dynamically updates the recommendation priority through an exponentially weighted moving average.

7. The method according to claim 1, characterized in that, The insertion methods include at least one of copying content, inserting a reference block, inserting an image snapshot, and inserting a link; wherein, copying content preserves the editability of the content block, inserting a reference block preserves the referencing relationship, and inserting an image snapshot preserves the original layout style.

8. The method according to claim 1, characterized in that, When a user selects multiple content blocks in the same slide of a presentation document, the system obtains the bounding box coordinates of each content block, calculates its relative position, and creates a floating canvas in the target document to render each content block onto the canvas according to its original relative position, and outputs it as a vector graphic or a first composite object.

9. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program, wherein the computer program is configured to perform the method described in any one of claims 1 to 8 when executed.

10. An electronic device comprising a memory and a processor, characterized in that, The memory stores a computer program, and the processor is configured to run the computer program to perform the method as described in any one of claims 1 to 8.

Images