Literature learning note point prompting method, system and device and storage medium

Abstract

The invention relates to the technical field of computers, and discloses a literature learning note point prompting method, system and device and a storage medium, and the method comprises the steps: obtaining a sequence fragment of a historical learning literature of a current user; constructing a note classification system of the user based on the sequence fragments and the existing notes of the user; based on the sequence fragments and a note classification system, training and learning the potentially labeled note points and the classification of the note points when the user learns by establishing a prediction model; processing the current to-be-learned literature of the user based on the prediction model, and outputting a prediction result which comprises the note points and the classification of the note points; and generating candidate annotation contents based on the prediction result, presenting the note point and the candidate annotation contents to the user for interactive operation, and confirming a final note point in response to the interactive operation. The note taking efficiency can be improved, so that the learning speed and depth of the user are improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of computer, in particular to a literature learning note point prompting method, system, device and storage medium. BACKGROUND

[0002] At present, learning notes are a series of operations performed by users in the process of learning literature, such as marking (highlighting, underlining, selecting) and quoting the content of the original text, and adding comments (their own insights, ideas or custom tags) to the quoted content.

[0003] User learning notes have high personalization features, which are not only reflected in the content of the original text, but also in the structure, expression method, symbols and marks used in the comment content. Influenced by multiple factors such as user knowledge background, learning habit, thinking mode and interest, the degree and mode of personalization of each user are unique. Therefore, even for the same learning literature, the notes recorded by different users will present completely different attention points, styles and characteristics. Such personalized learning notes not only help users better understand and remember knowledge, but also reflect their unique cognitive process and learning strategies.

[0004] The conventional note mode is that the user first reads the literature word by word and sentence by sentence, filters effective information fragments, and then marks and comments. This manual filtering and commenting mode not only consumes time, but also easily breaks the coherence of thinking, and is extremely low in efficiency in the face of information overload.

[0005] Therefore, the present application provides a literature learning note point prompting method to solve the above technical problems. SUMMARY

[0006] The purpose of the present application is to provide a literature learning note point prompting method, system, device and storage medium, which solves the technical problems of time-consuming, easy to break the coherence of thinking and low efficiency of the manual filtering and commenting mode in the prior art.

[0007] In order to solve the above technical problems, the present application provides a literature learning note point prompting method, comprising:

[0008] responding to the obtained sequence fragments of the current user's historical learning literature, wherein the sequence fragments include the text content, structure information and associated knowledge metadata of the literature;

[0009] based on the sequence fragments and the user's existing notes, constructing a note classification system of the user, wherein the note classification system includes user personalized note classification tags and codes;

[0010] Based on the sequence fragments and the note classification system, a prediction model is established to train and learn the note points that users may potentially annotate during learning and the categories to which the note points belong.

[0011] The prediction model is used to process the current literature to be studied by the user and output prediction results, wherein the prediction results include note points and the category to which the note points belong;

[0012] Based on the prediction results, candidate annotation content is generated, and the note points and candidate annotation content are presented to the user for interactive operation. The final note points are confirmed in response to the interactive operation.

[0013] The sequence fragments are used as input data for constructing a note classification system and for the prediction model to learn. The note classification system is used for the prediction model to learn and to provide a classification basis for generating candidate annotation content. The prediction model converts the sequence fragments into note points and performs classification predictions. The prediction results are used as input for generating candidate annotation content.

[0014] Compared with existing technologies, its advantages are as follows:

[0015] This invention discloses a method, system, device, and storage medium for providing note-taking suggestions for literature study. By constructing a user-specific note classification system and predicting note points and classifications based on machine learning models, it achieves a paradigm shift from "manual information screening" to "intelligent assisted thinking." It can proactively identify literature content segments that users may be interested in and generate personalized annotation suggestions, freeing users from tedious information screening and allowing them to focus more on in-depth thinking and knowledge digestion. By analyzing users' historical learning behavior and note characteristics, it can capture each user's unique focus, thinking habits, and expression style, providing completely personalized note suggestions even for the same document. This improves the accuracy and practicality of notes, making the generated notes more in line with the user's actual needs and cognitive characteristics. Through intelligent prediction and annotation generation, users can quickly locate key content in the literature without reading word for word, effectively alleviating information overload. The multi-dimensional candidate annotations provided can inspire user thinking, promoting deeper knowledge understanding and absorption, thus improving learning speed while ensuring learning depth. By constructing a hierarchical note classification system and a prediction model that integrates multiple features, it can adapt to the needs of users from different disciplines and knowledge backgrounds, supporting a variety of learning scenarios from basic concept understanding to complex research analysis. Attached Figure Description

[0016] Other features, objects, and advantages of this application will become more apparent from the following detailed description of non-limiting embodiments with reference to the accompanying drawings:

[0017] Figure 1This is a flowchart illustrating some specific embodiments of the document study note-taking method of the present invention;

[0018] Figure 2 This is one of the flowcharts of another embodiment of the document study note prompting method of the present invention;

[0019] Figure 3 This is a second flowchart illustrating another embodiment of the document study note-taking method of the present invention;

[0020] Figure 4 This is a third flowchart illustrating another embodiment of the document study note-taking method of the present invention;

[0021] Figure 5 This is the fourth flowchart of another embodiment of the document study note prompting method of the present invention;

[0022] Figure 6 This is a schematic diagram of the structure of a document study note-taking system according to some specific embodiments of the present invention;

[0023] Figure 7 This is a schematic diagram of the structure of an electronic device according to some specific embodiments of the present invention;

[0024] In the diagram, 710 is the processor; 720 is the memory; 730 is the input device; and 740 is the output device. Detailed Implementation

[0025] To make the objectives, technical solutions, and advantages of this application clearer, the application will be further described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments in this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0026] It should be noted that any symbols and / or numbers present in the specification that are not marked in the accompanying drawings are not reference numerals.

[0027] Reference Figure 1 A method for providing prompts in literature study notes, including:

[0028] S101, in response to the acquisition of a sequence fragment of the current user's historical learning documents, wherein the sequence fragment includes the text content, structural information and associated knowledge metadata of the documents;

[0029] S102, Based on the sequence fragment and the user's existing notes, construct the user's note classification system, wherein the note classification system includes the user's personalized note classification tags and codes;

[0030] S103, Based on the sequence fragment and the note classification system, a prediction model is established to train and learn the note points that users may potentially annotate during learning and the categories to which the note points belong;

[0031] S104, Based on the prediction model, process the current literature to be studied by the user and output the prediction result, wherein the prediction result includes note points and the category to which the note points belong;

[0032] S105, Based on the prediction results, candidate annotation content is generated, the note points and candidate annotation content are presented to the user for interactive operation, and the final note points are confirmed in response to the interactive operation.

[0033] The sequence fragments are used as input data for constructing a note classification system and for the prediction model to learn. The note classification system is used for the prediction model to learn and to provide a classification basis for generating candidate annotation content. The prediction model converts the sequence fragments into note points and performs classification predictions. The prediction results are used as input for generating candidate annotation content.

[0034] Specifically, in this embodiment of the invention, in response to the acquisition of a sequence of historical learning documents of the current user, wherein the sequence of documents includes the text content, structural information, and associated knowledge metadata, the structural information covers the title, keywords, abstract, author, institution, chapter title, and chapter content sentences, and the knowledge metadata includes conceptual knowledge elements, definitional knowledge elements, and methodological knowledge elements, and each sequence of documents is appended with detailed information such as the chapter title, knowledge element type, text feature identifier, and contextual information including similar data of preceding and following segments; based on the sequence of documents and the user's existing notes, a user's note classification system is constructed, and the user's history is analyzed. The marked fragments and corresponding annotations in the notes are sequentially replaced with knowledge elements, keywords, and multilingual representations in the annotations to form corresponding note classification codes. Then, the annotation content of all user notes is analyzed to extract global and local common note classifications, which are then converted into classification codes. Finally, knowledge element classification, keyword classification, multilingual representation classification, and common classification are integrated to form a user note classification system. Based on the sequence fragments and the note classification system, a predictive model is built to train and learn the potential note points and their classifications during learning. The training samples include the text content, structural features, contextual information, and corresponding note classification codes of the sequence fragments. The system embeds text content and vectorizes non-text features, fusing text embeddings, structured features, and contextual features to form a unified feature representation. Based on this unified feature representation, a multi-label classification model predicts note classification codes, using a binary cross-entropy loss function for model training. The system then processes the user's current learning document using the prediction model, outputting prediction results, including obtaining sequence fragments of the learning document, inputting them into the model to obtain predicted note classification codes, and extracting the sequence fragments corresponding to non-note point classification codes as note points and their respective categories. Based on the prediction results, candidate annotation content is generated by converting note classification codes into note category names and classifying knowledge elements and keywords. Keyword and multilingual representation classification are used to reconstruct the actual text representation. Based on the note points and category names, prompt word engineering is used to generate multiple candidate annotations. The note points and candidate annotations are presented to the user for interactive operation, including displaying predicted note points and candidate annotations, receiving user filtering and modification instructions, and responding to the user's confirmation to save the final note points. The sequence fragments serve as input data for building the note classification system and the prediction model learning. The note classification system provides the classification basis for the prediction model learning and generating candidate annotations. The prediction model converts the sequence fragments into note points and category predictions, and the prediction results serve as input for generating candidate annotations.

[0035] For example, taking the content of user A's learning literature, "If a specific band with the same characteristics as the positive control can be amplified, it may be a correct positive clone," we obtain the sequence fragments of user A's historical learning literature, including text content and structural information such as title, keywords, and abstract, and add detailed information such as the chapter title being "Experimental Methods," the inclusion of the conceptual knowledge elements "positive control" and "positive clone," and whether it contains formulas. Based on user A's historical notes, we construct a note classification system. For example, we analyze the annotation information from user A's 1432 notes, replacing knowledge elements in the annotations, such as "quantification of policy texts," with the conceptual knowledge element code ncly_cku, and keywords, such as "insufficient manpower," with the keyword code ncly_kws, and multilingual representations such as "Collection." The words "Tube" and "Receiver" are encoded using the multilingual encoding ncly_lantrans. Global common note categories, such as "Research Results," are extracted and encoded as ncly_gcomm_98, while local common note categories, such as "What does it refer to," are encoded as ncly_lcomm_4, forming a note classification system for user A. A prediction model is built based on the sequence fragments and the note classification system. Training samples are constructed, including the text content, structural features, and contextual information of the sequence fragments, along with their corresponding note classification codes. The text content is embedded using a pre-trained model to obtain semantic vectors, and non-textual features, such as whether a formula is included, are encoded using multi-hot encoding. Quantization involves fusing text embeddings, structured features, and contextual features through weighted summation to form a unified feature vector. Based on this unified feature vector, a multilayer perceptron is used for multi-label classification prediction, and a binary cross-entropy loss function is used to train the model. The trained model is then used to process the sequence fragments of the literature currently being studied by user A. The input sequence fragment text content is "If the same specific band as the positive control can be amplified, it may be a correct positive clone." The model outputs the note classification code prediction as [ncly_kws,ncly_lcomm_4]. The sequence fragments corresponding to the non-note point classification codes are extracted as note points, i.e., the text content itself.Based on the prediction results, candidate annotations are generated. The note classification codes ncly_kws and ncly_lcomm_4 are converted into note classification names "keywords" and "refers to". The keywords are then used to restore the content, resulting in "positive control" and "positive clone". Based on the note content and classification name, candidate annotations are generated using prompt word engineering. For example, the prompt word is "Please generate 3 possible annotations for the user based on the original text and content tags". The original text is "If a band with the same specificity as the positive control can be amplified, it may be a correct positive clone". The content tags are ["positive control", "positive clone", "refers to"]. Candidate annotations such as "What does the positive control specifically refer to here? Is it a plasmid or other standard?", "Why is it said 'may be correct' instead of 'definite'? Is there a risk of false positives?", and "What is the relationship between the positive clone and the positive control? How can the correctness of the clone be confirmed by band comparison?" are generated. The note points and candidate annotations are presented to user A. User A filters and modifies the annotations, and finally confirms and saves the note points and annotations.

[0036] In some applications, sequence fragment information of users' historical learning documents is extracted based on full-text structured databases and knowledge element databases. This sequence fragment information is then organized into structured data containing titles, keywords, abstracts, authors, institutions, chapter titles, and chapter content sentences, maintaining the hierarchical organization of the documents. Detailed information, including the chapter title, knowledge element type, and text features, is appended to each sequence fragment. The chapter title uses a multi-level heading structure, and the knowledge element types include conceptual, definitional, and methodological knowledge elements. The text features include multi-dimensional feature identifiers such as whether formulas, charts, and cited content are included. Contextual information, including features of preceding and following fragments, is established for each sequence fragment. This contextual information uses the same structured representation as the current sequence fragment, forming a complete environmental semantic framework. The sequence fragment information serves as the foundation for constructing structured data, which is used to generate detailed and contextual information. This detailed and contextual information together constitute the input features for modeling and learning, providing comprehensive feature support for subsequent personalized note prediction.

[0037] For example, taking an academic paper in the field of policy research as an example, the sequence fragment information of the user's historical learning documents is extracted from a full-text structured database. The title field is "Research on Methods of Quantitative Analysis of Policy Texts," the keyword fields are ["Policy Text", "Quantitative Methods", "Content Analysis"], and the abstract field contains multiple sentences and their location information. The sequence fragment information is organized into structured data, including organizing the chapter titles into a three-level structure of ["Introduction", "Research Methods", "Results Analysis"], and the chapter content sentences containing specific argumentative text. Detailed information is added to each sequence fragment; for example, the chapter title of a certain sequence fragment is ["Research Methods", "Data Collection"]. The set of knowledge elements includes the conceptual knowledge element "Policy Text Quantification" and the methodological knowledge element "Content Analysis Method". The text feature is identified as not containing formulas or charts but containing references. Contextual information is established for this sequence segment. The preceding segment contains the background introduction, and the following segment contains the description of specific implementation steps. The contextual information uses the same structured representation. The sequence segment information serves as the basis for constructing structured data. The generated structured data is used to generate detailed information and contextual information. The final complete feature set containing the chapter structure, knowledge element distribution, and contextual semantics serves as the input features for modeling and learning, establishing a data foundation for subsequent personalized note prediction.

[0038] In some applications, marked segments and corresponding annotations from a user's historical notes are retrieved. The marked segments are original text selected by the user during literature review, and the annotations are personalized interpretations and comments added by the user. Based on the annotation information, knowledge elements, keywords, and multilingual representations are identified and encoded. Conceptual, definitional, and methodological knowledge elements involved in the annotations are identified and converted into corresponding classification codes. Literature keywords contained in the annotations are identified and converted into keyword classification codes. The multilingual correspondence between annotations and the original text is identified and converted into multilingual representation codes. Common features in the annotation content are analyzed to extract global common note categories and local common features. Common note classification is achieved by statistically analyzing all user note annotations to obtain high-frequency global common classifications, and by analyzing current user note annotations to obtain personalized high-frequency local common classifications. Various note classification codes are integrated to form a complete user note classification system. Knowledge element classification, keyword classification, multilingual representation classification, and common classification are organized according to a unified coding standard, establishing a mapping relationship between classification names and classification codes. The marked fragments and annotation information serve as inputs for recognition and encoding conversion. The encoding conversion results are used for common feature analysis, and the common feature analysis results participate in constructing a complete user note classification system, forming a hierarchical and structured personalized classification framework.

[0039] For example, taking user A's study of medical literature as an example, the system retrieves marked fragments and corresponding annotations from the user's historical notes. One marked fragment is "Specifically, after the introduction of some equipment and facilities, due to insufficient personnel, related medical and health services cannot be carried out, resulting in a generally high idle rate of equipment and facilities," and the corresponding annotation is "insufficient manpower." Based on this annotation, the system identifies and encodes knowledge elements, keywords, and multilingual representations. It identifies that the annotation does not contain conceptual, definitional, or methodological knowledge elements, keeping the annotation content unchanged. It identifies "insufficient manpower" as a keyword in the original text and converts it into the keyword classification code ncly_kws. Since the annotation and the original text are both in Chinese and there is no multilingual correspondence, no multilingual encoding conversion is performed. The system analyzes the common characteristics of the annotation content. Statistical analysis reveals that "research conclusions" appears more frequently than the average in all user notes, classifying it as a global common note and encoding it as n. The code `cly_gcomm_98` was used to identify the phrase "referring to what" which appeared frequently in the current user notes. This phrase was then categorized as a local common note and encoded as `ncly_lcomm_4`. A complete user note classification system was formed by integrating various note classification codes, including the default category "non-note point" code `ncly_none`, the knowledge element category "concept knowledge element" code `ncly_cku`, the keyword classification code `ncly_kws`, the multilingual representation classification code `ncly_lantrans`, and the common category codes `ncly_gcomm_98` and `ncly_lcomm_4`. Tag fragments and annotation information served as input for recognition and encoding conversion. The resulting classification codes were used for common feature analysis, and the global and local common classifications obtained from this analysis contributed to the construction of the final user note classification system, forming a classification foundation that supports personalized note prediction.

[0040] In some applications, a training sample set is constructed, containing sequence fragment features and corresponding note classification codes. This training sample set consists of the text content, structured features, contextual information, and the note classification code corresponding to the sequence fragment. The text content in the training samples is embedded, and a pre-trained language model is used to convert the text content into a semantic vector representation. Non-textual features are vectorized, converting binary features in the structured features into multi-hot encoded vectors. The hierarchical title features are obtained by weighted summation after obtaining the title embeddings through a pre-trained model to obtain the title representation vector. The text embedding representation, structured feature encoding, and contextual features are fused to form a unified feature representation, which is then calculated... The attention weights of context fragments enable dynamic context selection. After concatenating the text embedding vector, structured feature vector, and context feature vector, the contribution of different feature types is learned through a weight matrix. A multi-label classification model is trained based on a unified feature representation to obtain a prediction model. A multi-layer perceptron architecture is used for multi-label classification prediction. Each classification encoding predicts independently and outputs a probability value through an activation function. The binary cross-entropy loss function is used to optimize the model parameters. The training sample set provides basic data for text embedding and feature encoding. The embedding and encoding results are used for feature fusion. The feature fusion result serves as the training input for the multi-label classification model, forming an end-to-end note-point prediction model training process.

[0041] For example, taking user A's study of computer science literature as an example, a training sample set is constructed containing sequence fragment features and corresponding note classification codes. One training sample contains the sequence fragment text content "Application of Deep Learning Models in Natural Language Processing". The structured features include the chapter title ["Introduction", "Background Technology"], contain conceptual knowledge elements ["Deep Learning", "Natural Language Processing"], and do not contain formulas or charts. The context features include the preceding fragment "Development of Artificial Intelligence Technology" and the following fragment "Specific Application Cases". The corresponding note classification codes are [ncly_cku, ncly_kws]. The text content is embedded and represented. A pre-trained language model is used to convert "Application of Deep Learning Models in Natural Language Processing" into a 384-dimensional semantic vector. The non-text features are vectorized and encoded. The six binary features are converted into a 6-dimensional multi-hot encoded vector [1,0,1,0,1,0]. The two-level heading features are converted into... The semantic vector is weighted and summed with weights of 0.6 and 0.4 to obtain the title representation vector. A unified feature representation is formed by fusing text embedding representation, structured feature encoding, and contextual features. Contextual segment attention weights are calculated to be 0.7 and 0.3. The text embedding vector, structured feature vector, and contextual feature vector are concatenated into a 456-dimensional vector and then fused using a weight matrix. A multi-label classification model is trained based on the unified feature representation, using a three-layer perceptron architecture for multi-label classification prediction. The output layer uses a sigmoid activation function to obtain a prediction probability of 0.92 for concept knowledge elements and 0.87 for keywords. A prediction label is determined with a threshold of 0.5, and the model is optimized using a binary cross-entropy loss function. The training sample set provides the basic data for text embedding and feature encoding. The vectors obtained from embedding and encoding are used for feature fusion. The unified representation after feature fusion serves as the training input for the multi-label classification model, completing the training process of the prediction model.

[0042] In some applications, sequence fragment data of the literature to be studied is acquired by extracting all text units of the literature from a full-text structured database and a knowledge element database. The sequence fragment data is input into a prediction model to obtain note classification code prediction results. The prediction model outputs multiple note classification codes and their corresponding prediction probability values ​​for each sequence fragment. Valid note points and their corresponding classification codes are selected from the prediction results. Low-probability prediction results are filtered out by setting a probability threshold, and sequence fragments corresponding to non-note point classification codes are excluded. The correspondence between valid note points and classification codes is output to form a structured set of prediction results. The sequence fragment data serves as the input to the model prediction, and the model prediction results are used to select valid note points. The selection results constitute the correspondence between note points and classification codes, providing basic data for subsequent annotation generation.

[0043] For example, taking user A's study of biomedical literature as an example, the sequence fragment data of the literature to be studied is obtained, including the text unit "If a specific band with the same as the positive control can be amplified, it may be a correct positive clone" and its related structural information and knowledge metadata; the sequence fragment data is input into a pre-trained prediction model, and the model outputs the note classification code prediction results, obtaining the prediction probability of the corresponding concept knowledge element of the sequence fragment as 0.76, the prediction probability of the keyword as 0.82, the prediction probability of the common category "what does it refer to" as 0.68, and the prediction probability of non-note points as 0.15; from the prediction results, effective note points and their corresponding classification codes are screened out, the probability threshold is set to 0.5, the keyword classification code ncly_kws and the common category code ncly_lcomm_4 with the prediction probability exceeding the threshold are retained, and the non-note point classification code ncly_none is excluded; the correspondence between effective note points and classification codes is output, establishing the sequence fragment "If a specific band with the same as the positive control can be amplified, it may be a correct positive clone" and the classification code [ncly_kws, The mapping relationship of ncly_lcomm_4]; where the sequence fragment data is used as the input of the model prediction, the probability result obtained by the model prediction is used for effective note point screening, and the classification code retained after screening together with the original sequence fragment constitutes the correspondence between note points and classification codes, providing structured input for the next step of generating candidate annotations.

[0044] In some applications, note classification codes are converted into corresponding note classification names. The mapping between classification codes and classification names is achieved by querying an existing user note classification system. For specific types of classification names, content restoration is performed to obtain the actual text representation. For knowledge element classification names, the restoration is to the specific conceptual, definitional, or methodological knowledge elements contained in the note points. For keyword classification names, the restoration is to the core keyword content contained in the note points. For multilingual representation classification names, the restoration is to the corresponding language identifier and translation content. Based on the note point content and classification names, candidate annotation content is generated through prompt word engineering. A prompt word template containing the original document content and content tags is constructed. A pre-trained language model is used to generate multiple personalized candidate annotations based on the prompt words. A formatted set of candidate annotation content is output, organized according to a unified output standard. The note classification code conversion result is used for content restoration processing, and the content restoration result participates in the generation of candidate annotation content. The generated candidate annotation content forms a formatted output, providing a complete annotation suggestion scheme for subsequent user interaction.

[0045] For example, taking user A's study of biomedical literature as an example, the note classification code is converted into the corresponding note classification name. By querying user A's note classification system, the classification code ncly_kws is converted into "keywords", and ncly_lcomm_4 is converted into "what it refers to". For specific types of classification names, content restoration is performed to obtain the actual text representation. For keyword classification names, the core keywords "positive control" and "positive clone" contained in the note points are restored. For common classification names, the original "what it refers to" expression is maintained. Based on the note point content and classification names, candidate annotation content is generated through prompt word engineering. A prompt word template is constructed containing the original text content "If a specific band with the same characteristics as the positive control can be amplified, it may be a correct positive clone" and... The content tags ["positive control", "positive clone", "what does it refer to"] are used to generate three candidate annotations using a pre-trained language model. The output is a formatted set of candidate annotations, listed in a numbered list format as follows: "1. What specifically does the positive control refer to here? Is it a plasmid or another standard? 2. Why is it said 'possibly correct' instead of 'definite'? Is there a risk of false positives? 3. What is the relationship between the positive clone and the positive control? How can the correctness of the clone be confirmed through band alignment?". The category names obtained from the note classification encoding are used for content restoration. The keywords and common category content obtained from content restoration participate in the generation of candidate annotations. The three generated candidate annotations form a formatted output result, providing users with directly usable annotation suggestions.

[0046] In some applications, the note points and candidate annotations are presented to the user for interactive operation, and the final note point is confirmed in response to the interactive operation. This includes displaying the note point and corresponding candidate annotation content options, and simultaneously displaying the predicted note point text and multiple candidate annotations for the user to choose from through the user interface; receiving the user's filtering instructions for the note points and annotations, allowing the user to select appropriate annotation content from multiple candidate annotations through point-and-click operations; processing the user's modification input for the note points and annotations, providing text editing functions to allow users to personalize and supplement the selected annotation content; and saving the user's finally confirmed note points and annotation content, storing the correspondence between the user's confirmed note points and annotations in the user's note database. The displayed note points and candidate annotation options provide the basis for user interaction, the filtering instructions and modification inputs generated by user interaction are used for content updates, and the updated content is confirmed by the user to form the final saved result, completing the complete interactive process from intelligent prediction to user confirmation.

[0047] For example, taking user A's study of biomedical literature as an example, the system displays note points and corresponding candidate annotation options. The user interface shows the original note point text, "If a specific band with the same characteristics as the positive control can be amplified, it may be a correct positive clone," and three candidate annotation options, numbered 1, 2, and 3, for the user to choose from. It receives the user's filtering instructions for the note points and annotations. The user selects the second annotation option, "Why does it say 'may be correct' instead of 'confirm'? Is there a risk of false positives?" The system processes the user's modification input for the note points and annotations. The user adds modifications to the selected annotation, such as "Further verification is needed to confirm the risk of false positives; this can be confirmed through repeated experiments." The system saves the user's final confirmed note points and annotations, establishing a link between the original note point text and the user's final confirmed annotations and storing them in the database, while also recording the operation timestamp and document identification information. The displayed note points and candidate annotations load within 2 seconds, the user's filtering operation response time is less than 1 second, modified inputs are saved as drafts in real time, and the data is persistently stored within 5 seconds after final confirmation, forming a complete interactive loop from display to saving.

[0048] The following is combined with Figures 2-5 Another embodiment of the document study note-taking method of the present invention is described below:

[0049] like Figure 2 and Figure 3As shown, this module retrieves the sequence fragments of the user's historical learning documents. It primarily obtains the sequence fragment information (doc_history_seqs_infos) from the full-text structured database and the knowledge element database. The document's sequence fragment information consists of a set of sequence fragments [seq_1, seq_2,....,seq_i,...,seq_n]. The types of sequence fragments include: title, keywords, abstract, author, institution, chapter title, and chapter content sentences. The sequence fragment seq_i = {"segment sequence":"","document title":"","subject to":"","details of document":"","details of document":{"chapter title":["first-level title","second-level title","third-level title",.....],"language":"","conceptual knowledge elements":["","",...],"conceptual knowledge elements"} ":["","",...],"Contains methodological knowledge elements":["","",...],"Does it contain formulas":"Yes|No,"Does it contain charts":"Yes|No","Does it contain cited content":"Yes|No","Does it contain chapter titles":"Yes|No","Does it contain article titles":"Yes|No","Does it contain keywords":"Yes|No","Does it contain abstracts":"Yes|No","Does it contain authors":"Yes|No","Does it contain author affiliations":"Yes|No"},"Fragment sequence context information":{"Fragment sequence preceding information":{"Fragment sequence preceding fragment sequence List Information: "", "Chapter Title": ["First-level Heading", "Second-level Heading", "Third-level Heading", ...], "Language of the Fragment": "", "Conceptual Knowledge Elements Contained": ["", "", ...], "Definition Knowledge Elements Contained": ["", "", ...], "Method Knowledge Elements Contained": ["", "", ...], "Does it contain formulas": "Yes|No", "Does it contain charts": "Yes|No", "Does it contain cited content": "Yes|No", "Does it contain chapter titles": "Yes|No", "Does it contain article titles": "Yes|No", "Does it contain keywords": "Yes|No", "Does it contain abstracts": "Yes|No", "Does it contain authors": "Yes|No", "Does it contain author affiliations": "Yes|No"}, "Fragment Sequence Information": {"Fragment Sequence Information": "", "Chapter Title": ["First-level Heading", "Second-level Heading", "Third-level Heading", ...], "Language of the Fragment": "", "Conceptual Knowledge Elements Contained": ["", "", ...], "Definition Knowledge Elements Contained": ["", "", ...], "Method Knowledge Elements Contained": ["", "", ...[,"Does it contain formulas":"Yes|No,"Does it contain figures and tables":"Yes|No","Does it contain cited content":"Yes|No","Does it contain chapter titles":"Yes|No","Does it contain article titles":"Yes|No","Does it contain keywords":"Yes|No","Does it contain abstracts":"Yes|No","Does it contain authors":"Yes|No","Does it contain author affiliations":"Yes|No"}}}. Database description: The full-text structured database stores the structured information of documents. The structured information of a document is as follows: {"Article title":"","Author":"","Affiliation":"","Keywords":"","Abstract":[{"Sentence content":"","Sentence start and end positions":[]},{"Sentence content" :"","Sentence Start and End Position":[]},......],"Subject":"","First-Level Chapter":[{"Chapter Title":"","Chapter Content":[{"Sentence Content":"","Sentence Start and End Position":[],"Formulas Contained in the Sentence":"","Figures Contained in the Sentence":"","References Cited in the Sentence":["Reference 1",......]},{"Sentence Content":"","Sentence Start and End Position":[],"Formulas Contained in the Sentence":"","Figures Contained in the Sentence":"","References Cited in the Sentence":["Reference 1",......]},......],"Sub-Chapter":[{"Chapter Title":" {"Chapter Content":[{"Sentence Content":"","Sentence Start and End Position":[],"Formulas Contained in the Sentence":"","Graphs Contained in the Sentence":"","References Cited in the Sentence":["Reference 1",......]},{"Sentence Content":"","Sentence Start and End Position":[],"Formulas Contained in the Sentence":"","Graphs Contained in the Sentence":"","References Cited in the Sentence":["Reference 1",......]},......],"Sub-Chapter":[{}]},.......]},{"Chapter Title":"","Chapter Content":[{"Sentence Content":"","Sentence Start and End Position":[],"Sentence Sub-section:[{"Sentence Content":"","Sentence Chart":"","Sentence References":["Reference 1",......]},{"Sentence Content":"","Sentence Start and End Position":[],"Sentence Content":"","Sentence Chart":"","Sentence References":["Reference 1",......]},......],"Sub-section":[{"Chapter Title":"","Chapter Content":[{"Sentence Content":"","Sentence Start and End Position":[],"Sentence Content":"","Sentence Chart":"","Sentence References":["Reference 1",......]}The knowledge element database stores knowledge element pieces such as concepts, methods, and definitions contained in a document. A document contains the following knowledge elements: [{"Knowledge Element":"","Knowledge Element Start and End Position":[],"Knowledge Element Category":"Concept Knowledge Element|Method Knowledge Element|Definition Knowledge Element"},{"Knowledge Element":"","Knowledge Element Start and End Position":[],"Knowledge Element Category":"Concept Knowledge Element|Method Knowledge Element|Definition Knowledge Element"},{"Knowledge Element":"","Knowledge Element Start and End Position":[],"Knowledge Element Category":"Concept Knowledge Element|Method Knowledge Element|Definition Knowledge Element"},......].

[0050] like Figure 4 As shown, this module constructs the user's note classification system: It primarily builds the user's note classification system, `user_note_cly_set`, based on the user's historical notes. The process of constructing this system involves analyzing the global personalization, local commonalities, and global commonalities of note annotations. The construction process is as follows:

[0051] The user_note_cly_set contains, by default, category names and category codes for non-note points:

[0052] #Note category name --Note category code;

[0053] Non-note point -- ncly_none;

[0054] Obtain the marked fragments and annotation information for the marked fragments, quote_note_set, quote_note_set=[{"quote":"Specifically, after the introduction of some equipment and facilities, due to insufficient personnel, related medical and health services cannot be carried out, resulting in a generally high idle rate of equipment and facilities.","note":"Insufficient manpower"},{"quote":"Policy text quantification is a common policy research tool, mainly through...","note":"Policy text quantification"},{"quote":"First, from the conclusion analysis, digital transformation relying on government environmental protection subsidies and green technology innovation as the path has a significant improvement on corporate environmental performance. This conclusion remains valid after a series of robustness tests, such as changing the measurement method of explanatory variables and laging the explained variable by one period...","note":"Research conclusion"},{"quote" The co-marketing model refers to sales personnel forming teams to conduct sales, with each person only responsible for a specific part of the sales process. "What is the co-marketing model?", {"quote":"Collection Tube","note":"Receiver Tube"}......], where "quote" is the original text marked by the user, and "note" is the user's annotation on the original text.

[0055] Analyzing this user's note categorization system: The processing flow is as follows:

[0056] Extracting the user's knowledge element note categories: By sequentially replacing the knowledge elements in the notes with note category codes, the user's knowledge element note categories are obtained. The processing procedure is as follows:

[0057] Retrieve all concept knowledge elements cku_lst, definition knowledge elements dku_lst, and method knowledge elements mku_lst contained in the quote corresponding to the note; replace all concept knowledge elements cku_lst, definition knowledge elements dku_lst, and method knowledge elements mku_lst contained in the note with note classification codes to obtain the note after the first stage of preprocessing, denoted as prehand_1_note. The correspondence between the corresponding knowledge elements and note classification codes is as follows:

[0058] #Note Category Name--Note Category Code

[0059] Conceptual knowledge element -- ncly_cku; Definition knowledge element -- ncly_dku; Method knowledge element -- cly_mku; Add discovered knowledge element category tags to user_note_cly_set.

[0060] For example:

[0061] The `note` in `{"quote":"Policy text quantification is a common policy research tool, mainly used for...", "note":"Policy text quantification"}`, after the first stage of preprocessing, is `prehand_1_note=ncly_cku`. Because "Policy text quantification" is a conceptual knowledge element, knowledge element classification tags are added to `user_note_cly_set`, and `user_note_cly_set["Conceptual Knowledge Element"]=ncly_cku`.

[0062] The phrase {"quote":"Specifically, after the introduction of some equipment and facilities, due to a shortage of personnel, related medical and health services cannot be carried out, resulting in a generally high idle rate of equipment and facilities.","note":"Insufficient manpower"} is obtained as prehand_1_note="Insufficient manpower" after the first stage of preprocessing, because note does not contain the knowledge elements in quote.

[0063] {"quote":""First, from the conclusion analysis, digital transformation relying on government environmental protection subsidies and green technology innovation as the path has a significant improvement on corporate environmental performance. This conclusion remains valid after a series of robustness tests, such as changing the measurement method of explanatory variables and laging the explained variable by one period...","note":"Research Conclusion"} The note in "note" is obtained after the first stage of preprocessing, prehand_1_note="Research Conclusion", because note does not contain the knowledge elements in quote;

[0064] The `note` in `{"quote":"Cooperative marketing model refers to sales personnel forming teams to conduct sales, with each person only responsible for a specific part of the sales process.", and `note":"What is cooperative marketing model?"}`, after the first stage of preprocessing, the resulting `prehand_1_note=ncly_cku` is determined because "cooperative marketing model" is a conceptual knowledge element. Simultaneously, knowledge element category tags are added to `user_note_cly_set`, where `user_note_cly_set["conceptual knowledge element"]=ncly_cku`. Category tags that have already been added are not added again.

[0065] The note in {"quote":"Collection Tube","note":"Receiver Tube"} is prehand_1_note="Receiver Tube" after the first stage of preprocessing, because note does not contain the knowledge elements in quote.

[0066] Extracting the user's keyword note categories: By sequentially replacing the keywords in prehand_1_note with note category codes, the user's keyword note categories are obtained. The processing steps are as follows:

[0067] To obtain all keywords (kws_lst) contained in the quote corresponding to the note: There are multiple keyword extraction methods. Keyword extraction can be performed using jieba segmentation, or large models (such as Deepseek, Yuanbao, etc.) can be used to extract core keywords through prompts. Replace all keywords (kws_lst) contained in prehand_1_note with note classification codes to obtain the note after the second stage of preprocessing, denoted as prehand_2_note. The correspondence between keywords and note classification codes is as follows:

[0068] Keyword -- ncly_kws; Add keyword category tag to user_note_cly_set. #Note category name --Note category code; Keyword -- ncly_kws;

[0069] For example:

[0070] The prehand_1_note in {"quote":"Policy text quantification is a common policy research tool, mainly through...","prehand_1_note":"ncly_cku"} is processed in the second stage of preprocessing to obtain prehand_2_note=ncly_cku, because prehand_1note does not contain the keywords in the quote.

[0071] The quote {"quote":"Specifically, after the introduction of some equipment and facilities, due to a shortage of personnel, related medical and health services cannot be carried out, resulting in a generally high idle rate of equipment and facilities.", "prehand_1_note":"Insufficient manpower"} is used to obtain prehand_2_note=cly_kws after the second stage of preprocessing. Since "insufficient manpower" is a keyword in the quote, keyword category tags are added to user_note_cly_set, and user_note_cly_set["keyword"]=ncly_kws;

[0072] {"quote":""First, from the conclusion analysis, digital transformation relying on government environmental protection subsidies and green technology innovation as the path has a significant improvement in corporate environmental performance. This conclusion remains valid after a series of robustness tests, such as changing the measurement method of explanatory variables and laging the explained variable by one period...","prehand_1_note":"Research Conclusion"} The prehand_1_note in the second stage of preprocessing is now prehand_2_note ="Research Conclusion", because prehand_1_note does not contain the keywords in the quote;

[0073] The prehand_1_note in {"quote":"Cooperative marketing model refers to sales personnel forming teams to conduct sales, with each person only responsible for a specific part of the sales process.", "prehand_1_note":"What is ncly_cku?"} becomes prehand_2_note = "What is ncly_cku?" after the second stage of preprocessing, because prehand_1_note does not contain the keywords in the quote;

[0074] The prehand_1_note in {"quote":"Collection Tube","prehand_1_note":"receiver tube"} is prehand_2_note=receiver tube after the second stage of preprocessing, because prehand_1_note does not contain the keywords in the quote.

[0075] Extracting the user's multilingual note categories: By replacing the category tags for multilingual representations in prehand_2_note, the user's multilingual note categories are obtained. The extraction method is as follows:

[0076] Determine the language of the quote: quote_lan;

[0077] Determine the language of prehand_2_note: prehand_2_note_lan;

[0078] Determine whether quote and pre_hand_2_note are multilingual representations:

[0079] If quote_lan and prehand_2_note_lan are different and can be translated into each other, then it is considered a multilingual representation;

[0080] Otherwise, it does not fall under the category of multilingual representation;

[0081] Replace the multilingual note classification codes in prehand_2_note to obtain the second-stage preprocessed note, denoted as prehand_3_note. The correspondence between the multilingual representations and note classification codes is as follows:

[0082] Multilingual representation -- ncly_lantrans;

[0083] Add multilingual category tags to user_note_cly_set.

[0084] #Note Category Name--Note Category Code

[0085] Multilingual representation -- ncly_lantrans;

[0086] For example:

[0087] The prehand_2_note in {"quote":"Policy text quantification is a common policy research tool, mainly through...","prehand_2_note":"ncly_cku"} is prehand_3_note=ncly_cku after the third stage of preprocessing, because prehand_2_note does not contain multilingual representations;

[0088] The specific manifestation of this is that after the introduction of some equipment and facilities, due to a shortage of personnel, related medical and health services cannot be carried out, resulting in a generally high rate of idle equipment and facilities. The prehand_2_note in "prehand_2_note":cly_kws} is prehand_3_note=cly_kws after the third stage of preprocessing, because prehand_2_note does not contain multilingual representations;

[0089] {"quote":"First, based on the conclusion analysis, digital transformation relying on government environmental protection subsidies and green technology innovation significantly improves corporate environmental performance. This conclusion remains valid even after a series of robustness tests, such as changing the measurement method of explanatory variables and laging the explained variable by one period…","prehand_2_note":"Research Conclusion"} The prehand_2_note in the first part of the prehand_note is the result of the third-stage preprocessing, which is prehand_3_note="Research Conclusion", because prehand_2_note does not contain multilingual representations;

[0090] The prehand_2_note in {"quote":"Cooperative marketing model refers to sales personnel forming teams to sell, with each person only responsible for a specific part of the sales process.", "prehand_2_note":"What is ncly_cku?"} becomes prehand_3_note = ncly_cku after the third stage of preprocessing because prehand_2_note does not contain multilingual representations;

[0091] After the third stage of preprocessing, `prehand_2_note` in `{"quote":"Collection Tube","prehand_2_note":"Receiver Tube"}` becomes `prehand_3_note=ncly_lantrans`. Because `prehand_2_note` and `quote` are multilingual representations, multilingual note classification codes are added to `ser_note_cly_set`, and `user_note_cly_set["Multilingual Representation"]=ncly_lantrans`.

[0092] Extracting common note categories for this user: By replacing category tags with global and local commonalities in prehand_2_note, the common note categories for this user are obtained. The extraction method is as follows:

[0093] Extracting global common category tags: Analyze the note data of all users to extract common category tags. The process is as follows:

[0094] The knowledge meta-fragments, keyword fragments, and multilingual representation fragments in the corresponding quotes contained in all user notes are removed in turn to obtain the note set prehand_3_note_set=[{"quote":"","prehand_3_note":""},{"quote":"","prehand_3_note":""},......];

[0095] Perform grouped statistics on the remaining fragments of prehand_3_note in prehand_3_note_set:

[0096] prehand_3_note_set_group=[{"count":"","prehand_3_note":""},{"count":"","prehand_3_note":""},......], where count is the frequency of occurrence of prehand_3_note;

[0097] Take the prehand_3_notes whose frequency count is greater than the average frequency in the grouping results as the global common note set:

[0098] gloal_comm_note_set=[{"prehand_3_note":"","order_num":""},{"prehand_3_note":"","order_num":""},.....], where order_num is the reverse order of the frequency of prehand_3_note, starting from 1;

[0099] The correspondence between the global common note set `gloal_comm_note_set` and the note category codes is as follows:

[0100] #Note category name--Note category code:

[0101] prehand_3_note -- ncly_gcomm_{order_num};

[0102] For example, if the ranking of "Research Results" in prehand_3_note is 98, then the note category code for "Research Results" would be: ncly_gcomm_98

[0103] Extracting Local Common Category Tags: Analyze the note data of all the user's notes to extract local common category tags. The process is as follows:

[0104] After sequentially removing knowledge element fragments, keyword fragments, multilingual representation fragments, and global common fragments from the corresponding quotes in all user notes, the fourth stage of processing is obtained:

[0105] prehand_4_note_set=[{"quote":"","prehand_4_note":""},{"quote":"","prehand_4_note":""},...];

[0106] Perform grouped statistics on the remaining fragments of prehand_4_note in prehand_4_note_set:

[0107] prehand_4_note_set_group=[{"count":"","prehand_4_note":""},{"count":"","prehand_4_note":""},......], where count is the frequency of occurrence of prehand_4_note;

[0108] Take the prehand_4_notes in the grouping results whose frequency count is greater than the average frequency as the local common note set:

[0109] local_comm_note_set=[{"prehand_4_note":"","order_num":""},{"prehand_4_note":"","order_num":""},.....], where order_num is the reverse order of the frequency of prehand_4_note, starting from 1;

[0110] The correspondence between the local common note set (local_comm_note_set) and the note classification codes is as follows:

[0111] #Note category name--Note category code:

[0112] prehand_4_note -- ncly_lcomm_{order_num};

[0113] For example, if the ranking of "prehand_4_note" is 3, then the note category code for "what is it" would be: ncly_lcomm_3;

[0114] Replace all common notes (gloal_comm_note_set and local_comm_note_set) in prehand_3_note and prehand_4_note with note classification codes in sequence to obtain the note after the fifth stage of preprocessing, denoted as prehand_5_note:

[0115] Add common note category tags to user_note_cly_set.

[0116] For example:

[0117] {"quote":"Policy text quantification is a common policy research tool, mainly used to...","prehand_3_note":"ncly_cku"} The prehand_5_note obtained after the fourth and fifth stages of preprocessing is ncly_cku, because prehand_3note does not contain common notes.

[0118] The specific manifestation of this is that after the introduction of some equipment and facilities, due to a shortage of personnel, related medical and health services cannot be carried out, resulting in a generally high rate of idle equipment and facilities. The prehand_3_note in "prehand_3_note":cly_kws} is obtained after the fourth and fifth stages of preprocessing, resulting in prehand_5_note=cly_kws, because prehand_3_note does not contain common notes.

[0119] {"quote":""First, from the conclusion analysis, digital transformation relying on government environmental protection subsidies and green technology innovation as the path has a significant improvement on corporate environmental performance. This conclusion remains valid after a series of robustness tests, such as changing the measurement method of explanatory variables and laging the explained variable by one period...","prehand_3_note":"Research Conclusion"} The prehand_3_note in the first part, after the fourth and fifth stages of preprocessing, becomes prehand_5_note=ncly_comm_98. This is because prehand_3_note contains the global common note "Research Results," and the note classification code corresponding to "Research Results" is ncly_gcomm_98. At the same time, the common note classification code is added to user_note_cly_set, so user_note_cly_set["Research Results"]=ncly_gcomm_98;

[0120] The prehand_3_note in {"quote":"Cooperative marketing model refers to sales personnel forming teams to conduct sales, with each person only responsible for a specific part of the sales process.", "prehand_3_note":"What is ncly_cku?"}, after the fourth and fifth stages of preprocessing, becomes prehand_5_note = ncly_cku ncly_lcom_3. This is because prehand_3_note contains the local common note "What is it?", and the note classification code corresponding to "What is it?" is ncly_lcomm_3. At the same time, the common note classification code is added to user_note_cly_set, so user_note_cly_set["What is it?"] = ncly_lcomm_3;

[0121] The prehand_3_note in {"quote":"Collection Tube","prehand_3_note":"ncly_lantrans"} is processed through the fourth and fifth stages of preprocessing to obtain prehand_5_note=ncly_lantrans. Prehand_3_note does not contain any common notes.

[0122] like Figure 5 As shown, the model learns the note points that the user might annotate during learning and the categories to which those note points belong. This module mainly models by analyzing the note classification codes corresponding to the sequence segments of all the user's notes, and analyzes the relationship between sequence segments and note categories. The note classification code corresponding to the sequence segment of the document is the note classification code in the note classification system contained in the user's annotations for that sequence segment. If the user's quoted notes correspond to multiple sequence segments, they will be split into multiple sequence segments during modeling. The modeling process is as follows:

[0123] Constructing training predictions: The format of training predictions is as follows,

[0124] Training samples = {

[0125] "enter":{

[0126] "Title of the document":"text string",

[0127] "Segment Sequence":"Text Content",

[0128] "Discipline":"Discipline Name",

[0129] "Fragment Details":{

[0130] Chapter Title: ["Level 1 Heading", "Level 2 Heading", ...],

[0131] "Relevance of Indexed Sentences": "Description of Relevance Level",

[0132] "Conceptual knowledge elements included":["Concept 1","Concept 2",...],

[0133] "Contains definitional knowledge elements":["Definition 1","Definition 2",...],

[0134] "Contains method knowledge elements":["Method 1","Method 2",...],

[0135] "Does it contain formulas": "Yes / No"

[0136] "Include charts": "Yes / No"

[0137] #...Other binary features

[0138] },

[0139] "Context information":{

[0140] "Previous information":{similar structure},

[0141] "Following information":{similar structure}

[0142] }

[0143] },

[0144] Output: ["Classification Code 1", "Classification Code 2", ...] #Multiple Tags

[0145] }

[0146] The working principle of the overall structure of the model:

[0147] Text embedding: This module primarily converts various text fields into semantic vectors. The conversion formula is as follows:

[0148] ;

[0149] Each text field is encoded using BERT individually. Other models, such as SentenceBERT, can also be used. Structured feature encoding: This module primarily vectorizes non-text fields, including binary feature vectors and hierarchical heading vectorization. The vectorization formula is as follows:

[0150] ;

[0151] in:

[0152] : The negation / affirmation vector for all binary features;

[0153] : Convert features into multi-hot codes;

[0154] Weight matrix: learns the importance of each binary feature;

[0155] The activation function introduces nonlinearity.

[0156] ;

[0157] in:

[0158] : The title text of the i-th level;

[0159] Use pre-trained models such as BERT and SentenceBERT to obtain title embeddings;

[0160] : Specific weights for each level, learning the importance of different levels;

[0161] The summation operation merges information from all levels.

[0162] Contextual Feature Fusion: This module mainly fuses contextual fragment information. The fusion formula is as follows:

[0163] ;

[0164] ;

[0165] in:

[0166] : Calculate similarity using zoomed clicks;

[0167] Attention weights represent the importance of contextual fragments;

[0168] Weighted summation: Important context contributes more, while unimportant context contributes less;

[0169] Enables dynamic, adaptive context selection.

[0170] Feature fusion: This module mainly fuses existing features. The fusion formula is as follows.

[0171] ;

[0172] in:

[0173] Vector concatenation operation;

[0174] : Fuse weight matrices to learn the contributions of different feature types;

[0175] The output is a unified feature representation.

[0176] Multilayer perceptron classification: Multi-label classification prediction is performed based on the fused features. The prediction formula and loss function are as follows:

[0177] Category label prediction formula:

[0178] ;

[0179] The sigmoid function outputs probabilities between 0 and 1, with a default threshold of 0.5 for binary classification.

[0180] Each classification code is predicted independently, supporting multiple labels;

[0181] : Classification weight matrix.

[0182] Loss function:

[0183] ;

[0184] in:

[0185] : Represents the true label; binary labels are 0 or 1.

[0186] : Predicted probability, if greater than or equal to 0.5, the true label is 1; otherwise, the corresponding true label is 0.

[0187] Model learning objective: Minimize the prediction error for all labels.

[0188] Predicting the note points and their categories in the current user's study document: This module mainly uses a trained model to predict the note points and their categories in the current user's study document. The prediction supports prediction of the entire document and prediction of selected text. The steps are as follows:

[0189] Get the sequence segments to be predicted (predict_seqs): Get sequence segments for different prediction needs;

[0190] Full document prediction: Retrieves all sequence fragment information (doc_study_docs_infos) of the document the user wants to study. The sequence fragment information is ordered, and predict_seqs=doc_study_docs_infos.

[0191] Selected text prediction: Obtain the sequence fragments corresponding to the currently selected text from the full-text structured database and knowledge element database. There may be one or more sequence fragments, predict_seqs=[seq1,......];

[0192] Input the sequence segments to be predicted (predict_seqs) into the model to obtain the note classification code corresponding to each sequence segment;

[0193] Extracting note classification codes: The note classification code ncly_none corresponds to the note classification name "non-note point". Here, the note classification codes of non-ncly_none and the corresponding sequence segments are extracted as the predicted note classification codes and note points.

[0194] This module analyzes a user's possible annotations based on predicted note points and note point classification codes. The process is as follows:

[0195] Convert all note category codes to corresponding note category names: This is done by converting the predicted note category codes into corresponding note category names based on the constructed note category system user_note_cly_set;

[0196] Further transformation of category names: This step only involves restoring and transforming knowledge elements, keywords, and multilingual representations. The transformation rules are as follows.

[0197] Conceptual knowledge elements -- the conceptual knowledge elements contained in the notes;

[0198] Define knowledge elements -- the defined knowledge elements contained in the notes;

[0199] Method knowledge elements -- the method knowledge elements contained in the notes;

[0200] Keywords -- The core keywords contained in the notes;

[0201] Multilingual representation -- [Note point language, user-preferred language];

[0202] Generating possible user annotations for note points: This section uses prompt word engineering to generate possible user annotations for note points. The steps are as follows.

[0203] To create a personalized user-generated comment prompt, the following structure is required:

[0204] #Task Description;

[0205] Please generate three possible user comments based on the original text and content tags.

[0206] ##Original text of the document;

[0207] {$Original content};

[0208] ##Content tags;

[0209] [{$content tag},{$content tag},.......];

[0210] ## Output format;

[0211] 1. XXXXX;

[0212] 2. XXXXX;

[0213] 3. XXXXX;

[0214] The user can further filter, modify, and confirm the note points and annotations: This module mainly displays the predicted note points and n possible annotations for each note point to the user, allowing the user to further filter, modify, and confirm the note points and the required annotations.

[0215] For example, for user A on a learning platform who selected the literature content "If a specific band with the same characteristics as the positive control can be amplified, it is likely a correct positive clone," the following intelligent note-taking suggestions are provided:

[0216] The intelligent suggestion process is as follows:

[0217] Retrieve a sequence of historical learning documents for the current user: When user A has studied 263 documents and has 1432 notes;

[0218] Construct a note classification system for this user: Based on all sequence fragments of the 263 documents studied by user A and the annotations corresponding to each fragment, construct the user's note classification system: user_note_cly_set={”Non-note points":"ncly_none",”Keywords":"ncly_kws",”Conceptual knowledge elements":"ncly_cku",”Definition knowledge elements":"ncly_dku",”Method knowledge elements":"cly_mku",”Multilingual representation":"ncly_lantrans",......,"What does it refer to":"ncly_lcomm_4",......,Research results":"ncly_gcomm_98",......};

[0219] Modeling the learning process involves identifying the user's note points and their categories during the learning process: based on... Figure 5 Perform modeling;

[0220] Predict the notes and categories of the current user's historical learning documents: Predict the note category code for the content selected by user A as: [ncly_kws,ncly_lcomm_4];

[0221] Based on note points and note point classification, analyze the user's possible annotations:

[0222] The classification code, after conversion, is: ["positive control", "positive clone", "what does it refer to"].

[0223] The prompt is as follows:

[0224] #Task Description;

[0225] Please generate three possible user comments based on the original text and content tags.

[0226] ##Original text of the document;

[0227] If the same specific band as the positive control can be amplified, it may be a correct positive clone.

[0228] ##Content tags;

[0229] ["Positive control", "positive clone", "what does it refer to"];

[0230] ## Output format;

[0231] 1. XXXXX; 2. XXXXX; 3. XXXXX;

[0232] Three possible comments generated:

[0233] 1. What exactly does "positive control" refer to here? Is it a plasmid or other standard? 2. Why is it said "possibly correct" instead of "definitely correct"? Is there a risk of false positives? 3. What is the relationship between the positive clone and the positive control? How can the correctness of the clone be confirmed through band comparison? Further filtering, modification, and confirmation of notes and annotations are pending from user A: Show the three annotations to user A, awaiting further filtering, modification, and confirmation of notes and annotations.

[0234] like Figure 6 As shown, the present invention also provides a document study note-taking system, comprising:

[0235] The sequence fragment acquisition module 201 is configured to respond to the acquisition of sequence fragments of the current user's historical learning documents, wherein the sequence fragments include the text content, structural information and associated knowledge metadata of the documents;

[0236] The note classification system construction module 202 is configured to construct the user's note classification system based on the sequence fragment and the user's existing notes, wherein the note classification system includes the user's personalized note classification tags and codes;

[0237] The prediction model training module 203 is configured to train a prediction model based on the sequence fragment and the note classification system to learn the note points that the user may have labeled during learning and the category to which the note points belong.

[0238] The prediction result output module 204 is configured to process the user's current study literature based on the prediction model and output prediction results, wherein the prediction results include note points and the category to which the note points belong;

[0239] The note point confirmation interaction module 205 is configured to generate candidate annotation content based on the prediction result, present the note point and candidate annotation content to the user for interactive operation, and confirm the final note point in response to the interactive operation.

[0240] The sequence fragments are used as input data for constructing a note classification system and for the prediction model to learn. The note classification system is used for the prediction model to learn and to provide a classification basis for generating candidate annotation content. The prediction model converts the sequence fragments into note points and performs classification predictions. The prediction results are used as input for generating candidate annotation content.

[0241] like Figure 7As shown, the present invention also provides an electronic device, including: a processor, a communication interface, a memory, and a communication bus, wherein the processor, the communication interface, and the memory communicate with each other through the communication bus; the memory stores a computer program, and when the computer program is executed by the processor, the processor performs the steps of a document study note-taking method.

[0242] Figure 7 This is a schematic diagram of the structure of an electronic device provided in an embodiment of the present invention. For example... Figure 7 The structure shown in this embodiment of the invention includes an electronic device comprising one or more processors 710 and a memory 720; the processors 710 in this electronic device may be one or more. Figure 7 Taking a processor 710 as an example; the memory 720 is used to store one or more programs; the one or more programs are executed by the one or more processors 710, so that the one or more processors 710 implement a document study note prompting method as described in any one embodiment of the present invention.

[0243] The electronic device may also include an input device 730 and an output device 740.

[0244] The processor 710, memory 720, input device 730, and output device 740 in this electronic device can be connected via a bus or other means. Figure 7 Taking the example of a connection between China and Israel via a bus.

[0245] The memory 720 in this electronic device serves as a computer-readable storage medium, capable of storing one or more programs. These programs can be software programs, computer-executable programs, or modules, such as the program instructions / modules corresponding to the document study note-taking method provided in this embodiment of the invention. The processor 710 executes various functional applications and data processing of the electronic device by running the software programs, instructions, and modules stored in the memory 720, thereby implementing the document study note-taking method described in the above embodiment.

[0246] The memory 720 may include a program storage area and a data storage area. The program storage area may store the operating system and applications required for at least one function; the data storage area may store data created based on the use of the electronic device. Furthermore, the memory 720 may include high-speed random access memory and may also include non-volatile memory, such as at least one disk storage device, flash memory device, or other non-volatile solid-state storage device. In some instances, the memory 720 may further include memory remotely located relative to the processor 710, which can be connected to the device via a network. Examples of such networks include, but are not limited to, the Internet, intranets, local area networks, mobile communication networks, and combinations thereof.

[0247] Input device 730 can be used to receive input digital or character information, and to generate key signal inputs related to user settings and function control of the electronic device. Output device 740 may include display devices such as a display screen.

[0248] The present invention also provides a computer-readable storage medium storing a computer program executable by an electronic device, which, when run on the electronic device, causes the electronic device to perform the steps of a document study note-taking method.

[0249] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; 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 or all of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for providing prompts in literature study notes, characterized in that, include: In response to the acquisition of a sequence of historical learning documents of the current user, wherein the sequence of documents includes the text content, structural information and associated knowledge metadata of the documents; Based on the sequence fragments and the user's existing notes, a note classification system for the user is constructed, wherein the note classification system includes personalized note classification tags and codes for the user; Based on the sequence fragments and the note classification system, a prediction model is established to train and learn the note points that users may potentially annotate during learning and the categories to which the note points belong. The prediction model is used to process the current literature to be studied by the user and output prediction results, wherein the prediction results include note points and the category to which the note points belong; Based on the prediction results, candidate annotation content is generated, and the note points and candidate annotation content are presented to the user for interactive operation. The final note points are confirmed in response to the interactive operation. The sequence fragments are used as input data for constructing a note classification system and for the prediction model to learn. The note classification system is used for the prediction model to learn and to provide a classification basis for generating candidate annotation content. The prediction model converts the sequence fragments into note points and performs classification predictions. The prediction results are used as input for generating candidate annotation content.

2. The method for providing prompts in document study notes according to claim 1, characterized in that, In response to the acquisition of a sequence of historical learning documents of the current user, wherein the sequence of documents includes the text content, structural information, and associated knowledge metadata, further including: Extracting sequence fragment information of users' historical learning documents based on full-text structured databases and knowledge element databases; The sequence fragment information is organized into structured data containing article title, keywords, abstract, author, institution, chapter title, and chapter content sentences; Each sequence segment is appended with detailed information including the chapter title, knowledge element type, and text features. Contextual information containing features of the preceding and following segments is established for each sequence segment; The sequence fragment information serves as the basis for constructing structured data, which is used to generate detailed information and contextual information. The detailed information and contextual information together constitute the input features for modeling and learning.

3. The method for providing prompts in document study notes according to claim 1, characterized in that, Based on the sequence fragments and the user's existing notes, a note classification system is constructed for the user, wherein the note classification system includes personalized note classification tags and codes for the user, and further includes: Retrieve marked segments and corresponding annotation information from the user's historical notes; Based on the annotation information, the identification and encoding conversion of knowledge elements, keywords, and multilingual representations are performed; Analyze the common features in the annotation content to extract global common note categories and local common note categories; Integrate various note classification codes to form a complete user note classification system; The marked fragments and annotation information serve as inputs for recognition and encoding conversion. The encoding conversion results are used for common feature analysis, and the results of the common feature analysis participate in the construction of a complete user note classification system.

4. The method for providing prompts in document study notes according to claim 1, characterized in that, Based on the sequence fragments and the note classification system, a prediction model is established and trained to learn the note points that users may potentially annotate during learning and the categories to which those note points belong. This further includes: Construct a training sample set containing sequence fragment features and corresponding note classification codes; The text content in the training samples is embedded and represented, and the non-text features are vectorized and encoded. By integrating text embedding representation, structured feature encoding, and contextual features, a unified feature representation is formed; A prediction model is obtained by training a multi-label classification model based on a unified feature representation; The training sample set provides basic data for text embedding and feature encoding. The embedding and encoding results are used for feature fusion, and the feature fusion results serve as the training input for the multi-label classification model.

5. The method for providing prompts in document study notes according to claim 1, characterized in that, Based on the prediction model, the user's current study document is processed, and a prediction result is output. The prediction result includes note points and the category to which the note points belong, and further includes: Obtain sequence fragment data from the literature to be studied; Inputting sequence fragment data into the prediction model yields note classification and coding prediction results; Valid note points and their corresponding classification codes are selected from the prediction results; Output the correspondence between valid note points and category codes; The sequence fragment data serves as the input for model prediction, and the model prediction results are used to filter valid note points. The filtering results constitute the correspondence between note points and classification codes.

6. The method for providing prompts in document study notes according to claim 1, characterized in that, Based on the prediction results, candidate annotation content is generated, further including: Convert the note category codes into the corresponding note category names; Content restoration of specific category names yields actual text representations; Based on the note points and category names, candidate annotation content is generated through prompt word engineering; Output a formatted collection of candidate comments; The note classification and encoding conversion results are used for content restoration processing. The content restoration results participate in the generation of candidate annotation content, and the generated candidate annotation content forms a formatted output.

7. The method for providing prompts in document study notes according to claim 1, characterized in that, Presenting the note points and candidate annotations to the user for interactive operation, and confirming the final note points in response to the interactive operation, further includes: Displays note points and corresponding candidate annotation options; Receive user filtering instructions for note points and annotations; Process user input for modifying notes and annotations; Save the user's final confirmed notes and annotations.

8. A document study note-taking system, characterized in that, include: The sequence fragment acquisition module is configured to respond to the acquisition of sequence fragments of the current user's historical learning documents, wherein the sequence fragments include the text content, structural information and associated knowledge metadata of the documents; The note classification system construction module is configured to construct the user's note classification system based on the sequence fragments and the user's existing notes, wherein the note classification system includes the user's personalized note classification tags and codes; The prediction model training module is configured to learn, based on the sequence fragments and the note classification system, the potential annotations of users during learning and the categories to which the note points belong by establishing a prediction model; The prediction result output module is configured to process the user's current study literature based on the prediction model and output prediction results, wherein the prediction results include note points and the category to which the note points belong; The note point confirmation interaction module is configured to generate candidate annotation content based on the prediction result, present the note point and candidate annotation content to the user for interactive operation, and confirm the final note point in response to the interactive operation. The sequence fragments are used as input data for constructing a note classification system and for the prediction model to learn. The note classification system is used for the prediction model to learn and to provide a classification basis for generating candidate annotation content. The prediction model converts the sequence fragments into note points and performs classification predictions. The prediction results are used as input for generating candidate annotation content.

9. An electronic device, characterized in that, include: The system includes a processor, a communication interface, a memory, and a communication bus, wherein the processor, the communication interface, and the memory communicate with each other via the communication bus; the memory stores a computer program, which, when executed by the processor, causes the processor to perform the steps of the method according to any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, It stores a computer program executable by an electronic device, which, when run on the electronic device, causes the electronic device to perform the steps of the method according to any one of claims 1 to 7.

Images