Teaching manuscript annotation system

Abstract

The invention relates to the technical field of AI education enabling, in particular to an annotation system for teaching manuscripts. The method comprises the steps that a user terminal receives a to-be-annotated teaching manuscript uploaded by a user and uploads the to-be-annotated teaching manuscript to an application server; the application server obtains plain text content and a structured analysis instruction corresponding to the to-be-annotated teaching manuscript according to the received to-be-annotated teaching manuscript and a preset teaching major evaluation index, sends a calling request to an analysis engine, and synchronously transmits the plain text content and the structured analysis instruction to the analysis engine; the analysis engine executes teaching field deep semantic analysis according to the received plain text content and the structured analysis instruction to generate annotation data; and the application server associates and matches the received annotation data with the specific position of the original text of the teaching manuscript to be annotated, and sends the teaching manuscript after annotation integration to the user terminal. The problems of low teaching manuscript review efficiency, insufficient depth, high subjectivity and the like in the prior art can be solved.

Description

Technical Field

[0001] This invention relates to the field of AI-enabled education technology, and in particular to an annotation system for teaching documents. Background Technology

[0002] With the continuous advancement of educational informatization, the quality of core teaching documents such as teaching syllabi and lesson plans directly affects classroom effectiveness and teaching quality. Therefore, how to efficiently improve their professionalism has become a crucial issue in the education field. In recent years, the rapid development of artificial intelligence technology, especially natural language processing technology, has provided new directions for educational text analysis, and various basic text processing tools are gradually being applied to teaching scenarios.

[0003] Currently, the quality optimization of teaching materials mainly relies on two models: one is that teachers use general document processing software such as Word to conduct self-checks for basic text errors such as spelling and grammar; the other is that the teaching and research group leader or senior teachers conduct manual review, and the reviewers provide suggestions for modification through software annotation functions or offline communication, and the teachers then adjust the materials accordingly.

[0004] Existing technologies have significant shortcomings: general document software can only perform surface error checking and cannot reach professional dimensions such as teaching methodology and student learning adaptation; manual review is limited by personal experience and energy, which is not only inefficient and has delayed feedback, but also has problems such as inconsistent evaluation standards and strong subjectivity, making it difficult to meet the needs of large-scale, high-quality teaching manuscript optimization. Summary of the Invention

[0005] This invention provides an annotation system for teaching documents to address the problems of low efficiency, insufficient depth, and strong subjectivity in the review of teaching documents in the prior art.

[0006] In a first aspect, embodiments of the present invention provide an annotation system for teaching documents, comprising: User terminals, application servers, and analytics engines; The user terminal is communicatively connected to the application server and is used to receive teaching documents to be annotated uploaded by the user and upload them to the application server. The application server is also connected to the analysis engine for obtaining the plain text content and structured analysis instructions corresponding to the teaching manuscript to be annotated based on the received teaching manuscript to be annotated and the preset teaching professional evaluation indicators, as well as sending a call request to the analysis engine and synchronously transmitting the plain text content and the structured analysis instructions to the analysis engine. The analysis engine is used to perform deep semantic analysis in the teaching domain based on the received plain text content and the structured analysis instructions, generate annotation data, and return the annotation data to the application server. The application server is also used to associate and match the received annotation data with the specific location of the original text of the teaching document to be annotated, and send the annotated teaching document to the user terminal for the user to view.

[0007] In one possible implementation, the application server includes: a file processing module, a build analysis quality module, and an application programming interface (API) module; The file processing module is used to perform text extraction processing on the received teaching manuscript to be annotated, and obtain the plain text content corresponding to the teaching manuscript to be annotated. The construction and analysis quality module is used to construct structured analysis instructions for the plain text content based on preset teaching professional evaluation indicators; The API interface module is used to establish communication links between the application server, the user terminal, and the analysis engine.

[0008] In one possible implementation, the file processing module is used to scan the teaching manuscript to be annotated, which contains a mixture of text, formulas, charts, or code, to obtain a scanned document. Based on the scanned document, the module uses enhanced optical character recognition (OCR) technology to extract text, while preserving the chapter, paragraph, and list hierarchy of the teaching manuscript to be annotated during the extraction process, to obtain plain text content.

[0009] In one possible implementation, the file processing module includes: a scanning unit, a detection unit, a semantic structure parsing unit, a formula parsing unit, a code parsing unit, and a summarizing unit; The scanning unit is used to scan the teaching manuscript to be annotated, which contains a mixture of text, formulas, charts or codes, to obtain a scanned document; The detection unit is used to detect the document type corresponding to the content at different locations in the scanned document. The document type includes text type, mathematical formula type and programming code type. The text type content is sent to the semantic structure parsing unit for plain text content extraction, the mathematical formula type content is sent to the formula parsing unit for plain text content extraction, and the programming code type content is sent to the code parsing unit for plain text content extraction. The summarization unit is used to receive the parsing results sent by the semantic structure parsing unit, the formula parsing unit, and the code parsing unit, and to integrate the parsing results according to the chapter, paragraph, and list hierarchy of the teaching manuscript to be annotated to obtain plain text content.

[0010] In one possible implementation, the construction and analysis quality module is used to obtain the grade information corresponding to the teaching text to be annotated, obtain the curriculum standards corresponding to the grade information by calling the get_curriculum_standards interface, and generate structured analysis instructions by combining preset teaching professional evaluation indicators, user historical preferences, and real-time context. The preset teaching professional evaluation indicators include: teaching objectives, teaching key points and difficulties, adaptability of teaching activity design, and student learning analysis. The structured analysis instructions require the analysis engine to analyze the plain text content from the dimensions of curriculum standard compliance and teaching logic rationality.

[0011] In one possible implementation, the application server further includes: an annotation management module; The annotation management module is used to bind each annotation in the annotation data to the corresponding segment using the paragraph number and character offset of the teaching manuscript to be annotated as positioning identifiers.

[0012] In one possible implementation, the analysis engine is built on the Dify large-scale language model application platform and is used to decompose the uploaded teaching material content based on the dynamic model routing function of the Dify workflow. Based on the decomposed content, it performs in-depth semantic analysis of the plain text content in the teaching field to generate annotation data. At the same time, through the multi-turn dialogue memory function of the Dify workflow, it controls the coherence of the annotation context during the interaction with the application server.

[0013] In one possible implementation, the analysis engine includes: a Dify workflow and a knowledge retrieval module, a knowledge reasoning module, and a knowledge logic module that are communicatively connected to the Dify workflow. The Dify workflow is used to receive the plain text content and teaching materials, and send the teaching materials to the knowledge retrieval module; The knowledge retrieval module is used to match the plain text content with the textbook, and return the related exercises and / or related knowledge points in the corresponding content of the matched textbook catalog to the Dify workflow; The Dify workflow is also used to send the associated exercises and / or the associated knowledge points to the knowledge reasoning module; The knowledge reasoning module is used to sort out the related exercises and / or related knowledge points based on the plain text content, identify key and difficult points and mark them, and return the marked key and difficult points to the Dify workflow. The Dify workflow is also used to generate a teaching link chain based on the associated exercises and / or the associated knowledge points, the marked key and difficult points, and the matching degree between teaching objectives and learning situation, and the correspondence between teaching activities and key and difficult points, and send it to the knowledge logic module. The knowledge logic module is used to perform logical processing on the teaching process chain based on the plain text content, generate annotation data, and pass it to the Dify workflow; the annotation data is structured data in JSON format, and the structured data contains the original text fragment index, question type, improvement suggestions, and educational theory basis corresponding to each annotation.

[0014] The Dify workflow is also used to send the annotation data to the application server.

[0015] In one possible implementation, the analysis engine is also used to generate hierarchical annotations of different depths based on the importance of the plain text content, and to propose teaching improvement suggestions based on learning theory.

[0016] In one possible implementation, the user terminal is further configured to receive the annotated and integrated teaching text sent by the application server, and display it through a web interface, marking the original text fragments bound with the annotation data; and displaying a "comment" input box below the annotation data display area of ​​the web interface.

[0017] This invention provides a teaching document annotation system. A user terminal communicates with an application server to receive and upload teaching documents to be annotated. The application server also communicates with an analysis engine to obtain the plain text content and structured analysis instructions corresponding to the received teaching document and preset teaching professional evaluation indicators. The system also sends a request to the analysis engine and simultaneously transmits the plain text content and structured analysis instructions to the analysis engine. The analysis engine performs deep semantic analysis of the teaching domain based on the received plain text content and structured analysis instructions, generates annotation data, and returns the annotation data to the application server. The application server further matches the received annotation data with the specific location of the original text in the teaching document and sends the annotated teaching document to the user terminal for viewing, achieving "instant upload and review," greatly improving teachers' work efficiency. This invention, through sending structured analysis instructions generated based on preset teaching professional evaluation indicators to the analysis engine, enables the engine to perform professional and in-depth semantic analysis of the teaching field. Its analytical depth far surpasses traditional software and is more systematic and comprehensive than ordinary manual review. The analysis in this invention, based on a unified model and algorithm, provides objective and consistent evaluation standards, contributing to the standardized management of teaching quality within a region or school. Attached Figure Description

[0018] To more clearly illustrate the technical solutions in the embodiments of the present invention, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0019] Figure 1 This is a schematic diagram of the annotation system for teaching documents provided in an embodiment of the present invention; Figure 2 This is a schematic diagram of an annotation system for teaching documents provided in another embodiment of the present invention; Figure 3 This is a schematic diagram of the user interface provided in an embodiment of the present invention; Figure 4 This is a schematic diagram of the structure of the application server provided in an embodiment of the present invention; Figure 5 This is a schematic diagram of the file processing module provided in an embodiment of the present invention; Figure 6 This is a flowchart illustrating the processing of the annotation system for teaching documents provided in this embodiment of the invention. Figure 7 This is a flowchart illustrating the processing of a teaching document annotation system according to another embodiment of the present invention. Detailed Implementation

[0020] In the following description, specific details such as particular system architectures and techniques are set forth for illustrative purposes and not for limitation, in order to provide a thorough understanding of the embodiments of the invention. However, those skilled in the art will understand that the invention can be implemented in other embodiments without these specific details. In other instances, detailed descriptions of well-known systems, apparatuses, circuits, and methods are omitted so as not to obscure the description of the invention with unnecessary detail.

[0021] To make the objectives, technical solutions, and advantages of the present invention clearer, specific embodiments will be described below in conjunction with the accompanying drawings.

[0022] The current technology for approving teaching materials uses a model of "using general-purpose document processing software (such as Microsoft Word) combined with manual review." Specific description: (1) System composition: a computer with general office software (such as Microsoft Word) installed, and one or more teachers to review the documents.

[0023] (2) Method and steps: 1. Manuscript writing: Teacher A writes the teaching syllabus or lesson plan in software such as Microsoft Word.

[0024] 2. Initial Self-Check: Teacher A used Microsoft Word's built-in "Spell & Grammar Check" feature to perform basic text error checking on the document. This feature matches the text against a built-in dictionary and grammar rule base to identify potential errors.

[0025] 3. Human review: Teacher A sends the document to an experienced colleague (Teacher B).

[0026] 4. Annotations and Feedback: Teacher B opens the document and uses the software's "Revision" or "Annotation" function to manually enter their own revisions and evaluations in the appropriate locations. These comments are entirely based on Teacher B's personal knowledge and experience.

[0027] 5. Revision and Finalization: Teacher A revises the manuscript based on Teacher B's comments and finalizes the manuscript.

[0028] The review of this teaching document is limited to correcting literal errors. The core professional evaluation process relies entirely on manual labor, which has inherent limitations in terms of efficiency, objectivity, and depth, and cannot meet the needs for large-scale, high-quality, and timely optimization of teaching documents.

[0029] To address the problems of low efficiency, insufficient depth, and strong subjectivity in the review of teaching documents in existing technologies, this invention provides a teaching document annotation system that enables automated, intelligent, and in-depth analysis and annotation of teaching documents, assisting teachers in optimizing instructional design.

[0030] Figure 1 and Figure 2 A schematic diagram of an annotation system for teaching documents provided in an embodiment of the present invention is described in detail below: This invention provides an annotation system for teaching documents, the core of which is to combine a user terminal for file uploading and interactive display, an application server responsible for business logic and data management, and an analysis engine for performing in-depth analysis in the teaching field, thereby realizing a complete and efficient intelligent annotation process.

[0031] The annotation system for the teaching materials includes: user terminal 10, application server 20, and analysis engine 30; here, analysis engine 30 can be an AI analysis engine.

[0032] User terminal 10 is connected to application server 20 for receiving teaching documents to be annotated uploaded by users and uploading them to application server 20. Application server 20 is also connected to analysis engine 30 to obtain plain text content and structured analysis instructions corresponding to the teaching manuscript to be annotated based on the received teaching manuscript to be annotated and the preset teaching professional evaluation indicators, and to send a call request to analysis engine 30, and synchronously transmit plain text content and structured analysis instructions to analysis engine 30. Analysis engine 30 is used to perform deep semantic analysis in the teaching field based on the received plain text content and structured analysis instructions, generate annotation data, and return the annotation data to application server 20; Application server 20 is also used to associate and match the received annotation data with the specific location of the original text of the teaching document to be annotated, and send the annotated teaching document to user terminal 10 for user viewing.

[0033] Optionally, before receiving the teaching document to be annotated uploaded by the user, the user terminal 10 is also used to provide a user web interface for the user to log in to the annotation system of the teaching document. The user terminal 10 verifies whether the current user is a legitimate user by using the saved user information, that is, verifies whether the current user information is consistent with the saved user information. If the current user information is consistent with the saved user information, then the current user is a legitimate user, and the relevant information on successful user login is displayed on the user web interface.

[0034] After successfully logging in, users can upload teaching documents that need annotation, such as... Figure 3 As shown in the user interface diagram, users can upload teaching materials via the "Upload from Local" option on the left. Teaching materials can be in the form of text, formulas, charts, code, etc.

[0035] After receiving the uploaded teaching document to be annotated, user terminal 10 sends it to application server 20.

[0036] In one embodiment, such as Figure 4 As shown, the application server 20 includes: a file processing module 21, a quality analysis module 22, and an API interface module 23; The file processing module 21 is used to extract text from the received teaching manuscript to be annotated, and obtain the plain text content corresponding to the teaching manuscript to be annotated. The analysis quality module 22 is used to construct structured analysis instructions for plain text content based on preset teaching professional evaluation indicators. API interface module 23 is used to establish communication links between the application server, user terminal, and analysis engine.

[0037] Optionally, the API interface module 23 is used to receive the teaching manuscript to be annotated sent by the user terminal 10 and send it to the file processing module 21 for plain text content extraction.

[0038] In one embodiment, the file processing module 21 is used to scan the teaching document to be annotated, which contains mixed content including text, formulas, charts or code, to obtain a scanned document, and to extract text based on the scanned document using Optical Character Recognition (OCR) technology, while preserving the chapter, paragraph and list hierarchy of the teaching document to be annotated during the extraction process, to obtain plain text content.

[0039] Enhanced OCR technology refers to the use of enhanced optical character recognition technology to convert text content in carriers such as images and documents into editable and searchable text.

[0040] Optionally, since the teaching manuscript to be annotated may contain different types of content such as text, formulas, tables, or code, different parsing units are used in this embodiment to extract the content of various types of manuscripts in order to accurately extract the content of various types of manuscripts.

[0041] In one embodiment, see Figure 5 and Figure 6 As shown, the file processing module 21 includes: a scanning unit 211, a detection unit 212, a semantic structure parsing unit 213, a formula parsing unit 214, a code parsing unit 215, and a summarizing unit 216. Scanning unit 211 is used to scan a teaching manuscript to be annotated containing mixed content such as text, formulas, charts or codes, to obtain a scanned document; The detection unit 212 is used to detect the document type corresponding to the content in different positions in the scanned document. The document type includes text type, mathematical formula type and programming code type. The text type content is sent to the semantic structure parsing unit 213 for plain text content extraction, the mathematical formula type content is sent to the formula parsing unit 214 for plain text content extraction, and the programming code type content is sent to the code parsing unit 215 for plain text content extraction. The summarization unit 216 is used to receive the parsing results sent by the semantic structure parsing unit 213, the formula parsing unit 214 and the code parsing unit 215, and integrate the parsing results according to the chapter, paragraph and list hierarchy of the teaching manuscript to be annotated to obtain plain text content.

[0042] Optionally, different parsing units can be composed of different functional modules in Deepseek. The semantic structure parsing unit 213 can be Deepseek-Reader, which is used to parse the semantic structure of text content. The formula parsing unit 214 can be Deepseek-Math, which is used to parse mathematical formulas. The code parsing unit 215 can be Deepseek-Code, which is used to parse programming code content.

[0043] After extracting the plain text content from the teaching document to be annotated, an analysis quality module 22 is constructed. This module is used to construct structured analysis instructions for the plain text content based on preset teaching professional evaluation indicators. In one embodiment, the analysis quality module 22 is constructed to obtain the grade information corresponding to the teaching document to be annotated. The module obtains the curriculum standards corresponding to the grade information by calling the get_curriculum_standards interface, and generates structured analysis instructions by combining preset teaching professional evaluation indicators, user historical preferences, and real-time context. The preset teaching professional evaluation indicators include: teaching objectives, teaching key points and difficulties, adaptability of teaching activity design, and student learning analysis. The structured analysis instructions require the analysis engine to analyze the plain text content from the dimensions of curriculum standard compliance and teaching logic rationality.

[0044] It should be noted that when the analysis quality module 22 generates structured analysis instructions, the relevant nodes in the Dify workflow will construct a highly structured and goal-oriented instruction based on the preset teaching evaluation indicators. This instruction will instruct the analysis engine 23 to play the role of a "senior teaching expert" and perform analysis and evaluation from specific dimensions.

[0045] Dify is an application development platform for Large Language Models (LLMs), which can be understood as a "factory for AI applications." Using this factory and sophisticated "blueprints"—that is, our structured analysis instructions (Prompts)—we produce a "robot" specifically for "teaching document analysis." Therefore, the core technology of this invention lies not in manufacturing the LLM itself, but in how to "control" and "apply" the LLM to solve specific problems in the education field.

[0046] Prompt is a "template" we've distilled from extensive pedagogical knowledge and experience, guiding AI to think like a true expert. This is the "recipe" for achieving professional analysis and a key technical difference between this invention's embodiments and simply calling AI interfaces.

[0047] See Figure 4 As shown, when constructing the analysis quality module 22 to generate structured analysis instructions, it is necessary not only to rely on the teaching syllabus knowledge base, but also to refer to the user's historical preferences and real-time context for analysis, so as to provide personalized and domain-adapted instructions.

[0048] Optional, based on an educational scenario, an example of embedding syllabus requirements in the Prompt: def build_teaching_prompt(grade_level): standards = get_curriculum_standards(grade_level) return f""" You are annotating the teaching materials for grade {grade_level}. The following course standards must be met: {standards} """ After receiving the plain text content and structured analysis instructions corresponding to the teaching document to be annotated, application server 20 sends a call request to analysis engine 30, and synchronously transmits the plain text content and structured analysis instructions to analysis engine 30. Analysis engine 30 performs deep semantic understanding and pedagogical analysis on the plain text content according to the received structured analysis instructions, generating annotations. As required by the structured analysis instructions, analysis engine 30 returns structured data (such as a JSON array), rather than a piece of free text. This facilitates subsequent programmatic processing.

[0049] In one embodiment, the analysis engine 30 is built on the Dify large-scale language model application platform. It is used to decompose the uploaded teaching materials based on the dynamic model routing function of the Dify workflow, perform in-depth semantic analysis of the plain text content in the teaching field based on the decomposed content, and generate annotation data. At the same time, through the multi-turn dialogue memory function of the Dify workflow, the continuity of the annotation context during the interaction with the application server is controlled.

[0050] like Figure 7 As shown, the analysis engine includes: a Dify workflow and a knowledge retrieval module, a knowledge reasoning module, and a knowledge logic module that are communicatively connected to the Dify workflow; The Dify workflow is used to receive the plain text content and teaching materials, and send the teaching materials to the knowledge retrieval module; the teaching materials are uploaded by the user terminal 10, and can be in Word or PDF format.

[0051] The knowledge retrieval module is used to match the plain text content with the textbook, and return the related exercises and / or related knowledge points in the corresponding content of the matched textbook catalog to the Dify workflow; The Dify workflow is also used to send the associated exercises and / or the associated knowledge points to the knowledge reasoning module; The knowledge reasoning module is used to sort out the related exercises and / or related knowledge points based on the plain text content, identify key and difficult points and mark them, and return the marked key and difficult points to the Dify workflow. The Dify workflow is also used to generate a teaching process chain based on the associated exercises and / or the associated knowledge points, the marked key and difficult points, and the matching degree between teaching objectives and learning situation, and the correspondence between teaching activities and key and difficult points, and send it to the knowledge logic module; the teaching process chain may include a series of questions or obtain a design.

[0052] The knowledge logic module is used to perform logical processing on the teaching process chain based on the plain text content, generate annotation data, and pass it to the Dify workflow; the annotation data is structured data in JSON format, and the structured data contains the original text fragment index, question type, improvement suggestions, and educational theory basis corresponding to each annotation.

[0053] The Dify workflow is also used to send the annotation data to the application server.

[0054] See Figure 7 In the process, after receiving the plain text content and teaching materials, the Dify workflow also saves the plain text content, for example, by storing it on Volcano Cloud.

[0055] Optionally, the knowledge reasoning module can be the Deepseek-V3 model.

[0056] In one embodiment, the analysis engine 30 is also used to generate hierarchical annotations of different depths based on the importance of the plain text content, and to propose teaching improvement suggestions based on learning theory.

[0057] Optionally, when generating hierarchical annotations of different depths based on the importance of the plain text content, the analysis engine 30 performs text segmentation on the plain text content to obtain multiple semantic analysis units. Then, it extracts features for each semantic analysis unit, determines the concept density of the extracted features in the subject's core concept vocabulary, and determines the structural weight based on the position of each semantic analysis unit in the document. It also determines the semantic similarity of each semantic analysis unit with a pre-set curriculum standard knowledge base to determine its relevance to the syllabus of core knowledge points. Based on concept density, structural weight, and syllabus relevance, it determines the content criticality index of each semantic analysis unit and determines the annotation level based on the content criticality index. According to the defined annotation levels, different annotation generation strategy templates are called and the specific analysis results are filled in to obtain hierarchical annotations of different depths. Specifically: 1. Generation of the depth level to be annotated.

[0058] Text segmentation: Based on paragraphs, headings, and semantic integrity, plain text content is segmented into multiple semantic analysis units.

[0059] Feature extraction: Extracting features from each semantic analysis unit.

[0060] Concept density: Using a pre-trained subject-domain word vector model, based on the features extracted from each semantic analysis unit, the ratio of the number of features in each semantic analysis unit that match the subject's core concept vocabulary to the total number of features is calculated.

[0061] Structural weights: Weights assigned based on the position of each semantic analysis unit in the document. For example: chapter title (weight=1.0), section title (weight=0.7), paragraph first sentence (weight=0.5), and body text (weight=0.2).

[0062] Syllabus relevance: Each semantic analysis unit is matched with a pre-built curriculum standard knowledge base for semantic similarity, and its relevance score with core knowledge points is calculated.

[0063] Scoring Calculation: A weighted summation model is used to calculate the content criticality index for each semantic analysis unit. Optional, based on... Calculate the content criticality index for each semantic analysis unit, where, This represents the content criticality index of the current semantic analysis unit. , , The weights are assigned based on the grade level and subject, and the sum of the three is 1. Indicates concept density, Represents structural weights, Indicates the degree of relevance of the outline.

[0064] Hierarchical division: when If the value is greater than or equal to the first threshold, the current semantic analysis unit is determined to be the core layer to be annotated. The first threshold can be set according to requirements; for example, the first threshold can be 0.7. The content of the core layer to be annotated involves core concepts, principles, or higher-order cognitive objectives.

[0065] when If the value is greater than or equal to the second threshold and less than the first threshold, then the current semantic analysis unit is determined to be an explanation layer to be annotated. The second threshold can be set according to requirements; for example, the first threshold can be 0.4. This layer contains explanations, descriptions, or transitions of core knowledge.

[0066] when If the value is less than the second threshold, the current semantic analysis unit is determined to be an extension layer to be annotated; the content of this layer is examples, extended materials, or auxiliary information.

[0067] II. Generation of layered annotations.

[0068] Based on the defined depth level, different annotation generation strategy templates are invoked and the specific analysis results are populated: For the core layer, in-depth analysis is performed, which requires the application model to perform "precise definition of concepts", "clarification of common misunderstandings" and "association analysis with predecessor / follower knowledge" for the content of the semantic analysis unit.

[0069] Application Template: Populate the above deep analysis results into the deep annotation JSON template, with the type field marked as "core_concept", the depth field as "high", and containing structured fields such as definition, misconception, and knowledge_graph_link.

[0070] For the explanatory layer, standard analysis is performed, which requires the application model to perform "key point summarization", "clarity check", and "logical coherence verification".

[0071] Application Template: Fill the above standard analysis results into the standard annotation JSON template, with the type field being "explanation" and the depth field being "medium", including fields such as summary, clarity_score, and suggestion.

[0072] For the extended layer, perform basic analysis, which requires the application model to perform "fact accuracy checks" or "provide additional resource links".

[0073] Application Template: Populate the above basic analysis results into the prompt annotation JSON template, with the type field set to "example_or_extension" and the depth field set to "low", mainly including the accuracy_check or reference_link fields.

[0074] Optionally, the analysis engine 30, based on the teaching theory rule engine, generates teaching improvement suggestions, and these suggestions are executed in parallel with the content analysis during the hierarchical annotation generation process.

[0075] Teaching Problem Pattern Detection: The macroscopic analysis results of the entire plain text content or its semantic analysis units are matched with a predefined teaching problem pattern library. This pattern library is built based on learning theory. For example: Cognitive Overload Pattern: Detects whether too many new concepts are introduced in a short period of time without sufficient examples; Construction Connection Discontinuity Pattern: Detects whether there is a lack of review or linking of necessary prerequisite knowledge before the introduction of new knowledge; Feedback and Practice Deficiency Pattern: Detects whether there is a lack of formative exercises or summarizing questions after the explanation of key knowledge points.

[0076] Suggestion Generation and Mapping: Once a matching pattern is detected, the corresponding improvement suggestion generation rule is triggered. For example: If a "cognitive overload pattern" is detected, the following suggestions are generated: "X new concepts have been densely introduced in [specific chapter]. Based on cognitive load theory, the suggestions are: 1) Move some concepts to subsequent chapters; 2) Add an analogy or diagram for each core concept." If a "constructional connection discontinuity pattern" is detected, the following suggestion is generated: "It was detected that [Concept A] was not explicitly linked to previously learned knowledge [Concept B]. According to constructivist theory, it is recommended to add a connecting statement at the beginning: 'As we have learned [Concept B] before...'" Structured output: The generated teaching improvement suggestions are also returned in a structured format (such as a JSON array). Each suggestion contains fields such as pattern_detected (detected problem pattern), theory_basis (based teaching theory), location (location in the document), and concrete_suggestion (specific suggestion text).

[0077] Optionally, the analytics engine integrates the hierarchical annotation array and the teaching improvement suggestion array into a complete, structured JSON response, which is returned to the application server for subsequent presentation, storage, and interaction.

[0078] See Figure 6 As shown, after the annotation data is generated, quality verification is required. Optionally, sensitive words in the annotations can be filtered, educational appropriateness checks can be performed, and academic integrity verification can be conducted. If the quality verification passes, the annotation data is output. If the quality verification fails, the process is downgraded to Deepseek-V3 and annotations are generated again.

[0079] For example, a customized example of validation for educational scenarios: graph TD A [Original Comments] --> B [Sensitive Word Filtering] B --> C [Educational Appropriateness Check] C --> D [Academic Standards Verification] D --> E [Final Output] like Figure 4 As shown, after the analysis engine 30 outputs annotation data, it returns the annotation data to the application server 20. The application server 20 also includes an annotation management module 24. The annotation management module 24 is used to bind each annotation in the annotation data to the corresponding segment using the paragraph number and character offset of the teaching manuscript to be annotated as the positioning identifier.

[0080] like Figure 7As shown, the application server 20 is also used to store annotation data, teaching documents and other information using different storage components. For example, it uses Volcano Cloud Storage with massive storage capacity of object storage to store large-capacity, unstructured raw annotation data (such as complete annotation content in JSON format), and uses MySQL to store the structured core information of the annotations (such as annotation ID, associated objects, creation time, etc.) for long-term persistence and complex queries. It also uses a remote dictionary server (Redis) to store the "current version snapshot" of the annotations and sets a 24-hour expiration time for fast reading in high-frequency access scenarios.

[0081] After completing the above storage processing, the application server 20 returns the annotation ID and access link to the user terminal 10 through the Dify workflow, so that the user can query and access the corresponding content based on the annotation ID and access link.

[0082] In one embodiment, the user terminal 10 is also used to receive the annotated teaching text sent by the application server and display it through a web interface, marking the original text fragments bound with annotation data; and displaying a "comment" input box below the annotation data display area of ​​the web interface.

[0083] The original text (teaching manuscript) is displayed normally on the web interface of user terminal 10, while text segments with AI annotations are marked with highlights or underlines to make them more prominent and easier for users to view.

[0084] For example, when a user hovers their mouse over or clicks on the highlighted text, the system will display detailed AI annotations (problem type, suggestion, reason) in the form of cards and sidebars.

[0085] Below the AI ​​annotation card, a "Comment" input box is provided, allowing users to express their opinions on the AI ​​annotation (e.g., "This suggestion does not apply to my class") so that the analysis engine 30 can update the annotation data based on user comments or suggestions. Users can also actively create their own annotations by selecting text anywhere in the document with the mouse.

[0086] See Figure 7 As shown, user terminal 10 can also send the teaching document containing user annotations to application server 20. The annotation management module 24 in application server 20 then sends the teaching document containing user annotations to analysis engine 30 to re-verify and modify the annotation content. It also receives update suggestions from analysis engine 30, updates the document, returns the updated version to user terminal 10, and stores a snapshot of the current version in Redis. In this way, a teaching document simultaneously carries AI annotations and multi-layered user interaction information, forming a complete and traceable optimization record.

[0087] This invention provides a teaching document annotation system. A user terminal communicates with an application server to receive and upload teaching documents to be annotated. The application server also communicates with an analysis engine to obtain the plain text content and structured analysis instructions corresponding to the received teaching document and preset teaching professional evaluation indicators. The system also sends a request to the analysis engine and simultaneously transmits the plain text content and structured analysis instructions to the analysis engine. The analysis engine performs deep semantic analysis of the teaching domain based on the received plain text content and structured analysis instructions, generates annotation data, and returns the annotation data to the application server. The application server further matches the received annotation data with the specific location of the original text in the teaching document and sends the annotated teaching document to the user terminal for viewing, achieving "instant upload and review," greatly improving teachers' work efficiency. This invention, through sending structured analysis instructions generated based on preset teaching professional evaluation indicators to the analysis engine, enables the engine to perform professional and in-depth semantic analysis of the teaching field. Its analytical depth far surpasses traditional software and is more systematic and comprehensive than ordinary manual review. The analysis in this invention, based on a unified model and algorithm, provides objective and consistent evaluation standards, contributing to the standardized management of teaching quality within a region or school.

[0088] This invention supports user comments on AI annotations, treating AI as a "virtual teaching expert" that can engage in dialogue and communication. This creates a new model of human-computer collaboration for jointly optimizing teaching materials, which can better promote teachers' professional reflection and growth.

[0089] The annotation system for teaching documents improves annotation quality by 35% and reduces processing time by 40% compared to the single-model approach. Based on real-world test data from educational scenarios, the amount of manual modification is reduced by 60%.

[0090] It should be understood that the sequence number of each step in the above embodiments does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present invention.

[0091] 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 the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention, and should all be included within the protection scope of the present invention.

Claims

1. A commenting system for teaching documents, characterized in that, include: User terminals, application servers, and analytics engines; The user terminal is communicatively connected to the application server and is used to receive teaching documents to be annotated uploaded by the user and upload them to the application server. The application server is also connected to the analysis engine for obtaining the plain text content and structured analysis instructions corresponding to the teaching manuscript to be annotated based on the received teaching manuscript to be annotated and the preset teaching professional evaluation indicators, as well as sending a call request to the analysis engine and synchronously transmitting the plain text content and the structured analysis instructions to the analysis engine. The analysis engine is used to perform deep semantic analysis in the teaching domain based on the received plain text content and the structured analysis instructions, generate annotation data, and return the annotation data to the application server. The application server is also used to associate and match the received annotation data with the specific location of the original text of the teaching document to be annotated, and send the annotated teaching document to the user terminal for the user to view.

2. The annotation system for teaching manuscripts according to claim 1, characterized in that, The application server includes: a file processing module, a quality analysis module, and an API interface module; The file processing module is used to perform text extraction processing on the received teaching manuscript to be annotated, and obtain the plain text content corresponding to the teaching manuscript to be annotated. The construction and analysis quality module is used to construct structured analysis instructions for the plain text content based on preset teaching professional evaluation indicators; The API interface module is used to establish communication links between the application server, the user terminal, and the analysis engine.

3. The annotation system for teaching manuscripts according to claim 2, characterized in that, The file processing module is used to scan the teaching manuscript to be annotated, which contains a mixture of text, formulas, charts or code, to obtain a scanned document. Based on the scanned document, enhanced OCR technology is used to extract the text, and the chapter, paragraph and list hierarchy of the teaching manuscript to be annotated are preserved during the extraction process to obtain plain text content.

4. The annotation system for teaching manuscripts according to claim 3, characterized in that, The file processing module includes: a scanning unit, a detection unit, a semantic structure parsing unit, a formula parsing unit, a code parsing unit, and a summarizing unit; The scanning unit is used to scan the teaching manuscript to be annotated, which contains a mixture of text, formulas, charts or codes, to obtain a scanned document; The detection unit is used to detect the document type corresponding to the content at different locations in the scanned document. The document type includes text type, mathematical formula type and programming code type. The text type content is sent to the semantic structure parsing unit for plain text content extraction, the mathematical formula type content is sent to the formula parsing unit for plain text content extraction, and the programming code type content is sent to the code parsing unit for plain text content extraction. The summarization unit is used to receive the parsing results sent by the semantic structure parsing unit, the formula parsing unit, and the code parsing unit, and to integrate the parsing results according to the chapter, paragraph, and list hierarchy of the teaching manuscript to be annotated to obtain plain text content.

5. The annotation system for teaching manuscripts according to claim 2, characterized in that, The construction and analysis quality module is used to obtain the grade information corresponding to the teaching manuscript to be annotated, obtain the curriculum standards corresponding to the grade information by calling the get_curriculum_standards interface, and generate structured analysis instructions by combining preset teaching professional evaluation indicators, user historical preferences and real-time context. The preset teaching professional evaluation indicators include: teaching objectives, teaching focus and difficulties, adaptability of teaching activity design and student learning analysis. The structured analysis instructions require the analysis engine to analyze the plain text content from the dimensions of curriculum standard compliance and teaching logic rationality.

6. The annotation system for teaching manuscripts according to any one of claims 2-5, characterized in that, The application server also includes: a comment management module; The annotation management module is used to bind each annotation in the annotation data to the corresponding segment using the paragraph number and character offset of the teaching manuscript to be annotated as positioning identifiers.

7. The annotation system for teaching manuscripts according to any one of claims 1-5, characterized in that, The analysis engine, built on the Dify large-scale language model application platform, is used to decompose the uploaded teaching materials based on the dynamic model routing function of the Dify workflow. Based on the decomposed content, it performs in-depth semantic analysis of the plain text content in the teaching field and generates annotation data. At the same time, through the multi-turn dialogue memory function of the Dify workflow, it controls the coherence of the annotation context during the interaction with the application server.

8. The annotation system for teaching manuscripts according to claim 7, characterized in that, The analysis engine includes: a Dify workflow and a knowledge retrieval module, a knowledge reasoning module, and a knowledge logic module that are communicatively connected to the Dify workflow; The Dify workflow is used to receive the plain text content and teaching materials, and send the teaching materials to the knowledge retrieval module; The knowledge retrieval module is used to match the plain text content with the textbook, and return the related exercises and / or related knowledge points in the corresponding content of the matched textbook catalog to the Dify workflow; The Dify workflow is also used to send the associated exercises and / or the associated knowledge points to the knowledge reasoning module; The knowledge reasoning module is used to sort out the related exercises and / or related knowledge points based on the plain text content, identify key and difficult points and mark them, and return the marked key and difficult points to the Dify workflow. The Dify workflow is also used to generate a teaching link chain based on the associated exercises and / or the associated knowledge points, the marked key and difficult points, and the matching degree between teaching objectives and learning situation, and the correspondence between teaching activities and key and difficult points, and send it to the knowledge logic module. The knowledge logic module is used to perform logical processing on the teaching process chain based on the plain text content, generate annotation data, and pass it to the Dify workflow; the annotation data is structured data in JSON format, and the structured data contains the original text fragment index, question type, improvement suggestions, and educational theory basis corresponding to each annotation. The Dify workflow is also used to send the annotation data to the application server.

9. The annotation system for teaching manuscripts according to claim 8, characterized in that, The analysis engine is also used to generate hierarchical annotations of different depths based on the importance of the plain text content, and to propose teaching improvement suggestions based on learning theory.

10. The annotation system for teaching manuscripts according to any one of claims 1-5, characterized in that, The user terminal is also used to receive the annotated teaching text sent by the application server and display it through a web interface, mark the original text fragments bound with the annotation data, and display a "comment" input box below the annotation data display area of ​​the web interface.

Images