A device-side adaptive knowledge transfer method and system adapted for knowledge-based books

Abstract

This application discloses a device-side adaptive knowledge transfer method and system adapted to knowledge-based books. The method includes: structurally decomposing the content of the knowledge book to obtain a structured knowledge representation; generating a user cognitive state vector based on user interaction information related to the target knowledge book; performing knowledge boundary detection on all knowledge atoms and dividing all knowledge atoms into a mastered set, an accessible set, and a temporarily inaccessible set; responding to interactive response commands, identifying the current content intent, and determining the target knowledge atom; performing knowledge access judgment on each target knowledge atom, determining the set type, and defining a corresponding controlled response strategy; and generating target content based on the current content intent. This method enables novel knowledge transfer adapted to knowledge-based books, operates under predetermined knowledge boundary constraints, can identify the user's current cognitive state, control the knowledge release order, and execute constrained adaptive knowledge transfer on the device side.

Description

Technical Field

[0001] This disclosure relates to the field of artificial intelligence technology, and in particular to an edge-side adaptive knowledge transfer method and system adapted to knowledge-based books. Background Technology

[0002] Generative artificial intelligence technology is driving the transformation of knowledge-based content acquisition towards interactivity. In reading technical and professional books, users' demands for personalized interaction, such as instant explanations and layered explanations, are becoming increasingly prominent, but existing knowledge delivery methods struggle to meet these needs. Traditional paper books and e-books have fixed content structures, cannot recognize users' cognitive states, and have low knowledge acquisition efficiency; video courses can only track playback progress, with weak interactivity and no ability to dynamically adjust content; while online documents support updates, they remain static, easily leading to gaps in user understanding and lacking explicit control over knowledge boundaries and release order.

[0003] In the context of technical professional books, the shortcomings of existing methods are even more apparent. These books have clearly defined knowledge boundaries and strong prerequisites, requiring content depth to be adapted to different roles and avoiding cognitive biases. However, current technologies cannot accurately identify users' cognitive states, making it difficult to control content access and explicitly regulate knowledge boundaries. They also cannot proactively address or progressively explain questions that exceed the user's level. Furthermore, existing solutions struggle to balance personalization, controllability, and device-side feasibility. Static books lack adaptability, while large cloud-based models suffer from issues such as scope drift, high deployment costs, and poor offline applicability. Summary of the Invention

[0004] In view of this, embodiments of this disclosure provide an edge-side adaptive knowledge transfer method and system adapted to knowledge-based books. This method can construct an adaptive knowledge transfer method for knowledge-based books that operates under predetermined knowledge boundary constraints, can identify the user's current cognitive state, control the order of knowledge release, handle out-of-bounds requests, and execute restricted generation on the edge.

[0005] In a first aspect, embodiments of this disclosure provide an edge-side adaptive knowledge transfer method for knowledge-based books, including: The target knowledge books are decomposed into a structured content structure to obtain a structured knowledge representation; Generate a user cognitive state vector based on user interaction information related to the target knowledge book; Based on the user cognitive state vector and the structured knowledge representation, knowledge boundary detection is performed on all the knowledge atoms, and all the knowledge atoms are divided into a mastered set, an accessible set, and a temporarily inaccessible set. In response to interactive response commands, identify the intent of the current content, and determine the target knowledge atom based on the intent of the current content; Based on the known set, the accessible set, and the temporarily inaccessible set, a knowledge access determination is performed on each of the target knowledge atoms to determine the corresponding set type and the controlled response strategy corresponding to the set type. Based on the current content intent and the controlled response strategy, generate target content.

[0006] Secondly, embodiments of this disclosure provide an edge-side adaptive knowledge transfer system adapted to knowledge-based books, including: The decomposition unit is used to decompose the target knowledge book into a structured content structure to obtain a structured knowledge representation; The user cognitive state vector generation unit is used to generate user cognitive state vectors based on user interaction information related to target knowledge books. The knowledge boundary detection unit is used to perform knowledge boundary detection on all the knowledge atoms according to the user cognitive state vector and the structured knowledge representation, and divide all the knowledge atoms into a set that has been mastered, a set that can be entered, and a set that cannot be entered for the time being. The target knowledge atom determination unit is used to respond to an interactive response command, identify the current content intent, and determine the target knowledge atom based on the current content intent; The knowledge access determination unit is used to perform knowledge access determination on each of the target knowledge atoms based on the already mastered set, the accessible set, and the temporarily inaccessible set, to determine the corresponding set type, and to determine the controlled response strategy corresponding to the set type. The generation unit is used to generate target content based on the current content intent and the controlled response strategy.

[0007] The edge-side adaptive knowledge transfer method for knowledge-based books provided in this disclosure first performs a structured decomposition of the knowledge book's content to obtain a structured knowledge representation. This breaks down fragmented knowledge content into standardized, identifiable knowledge atoms, overcoming the limitations of traditional linear reading and the inability to break down knowledge, allowing knowledge transfer to shift from holistic instruction to precise matching. Based on user interaction information related to the target knowledge book, a user cognitive state vector is generated to accurately capture the user's initial cognitive level, points of interest, and weaknesses. Then, knowledge boundary detection is performed on all knowledge atoms, dividing them into a mastered set, an accessible set, and a temporarily inaccessible set, clearly defining the boundaries of knowledge transfer and preventing users from accessing content beyond their current cognitive level. This method avoids repeatedly pushing already mastered content, ensuring controllable knowledge transfer. Responding to interactive commands, it identifies the current content intent, determines the target knowledge atom, and accurately locates the knowledge content the user needs. For each target knowledge atom, it performs knowledge access judgment, determines the set type, and establishes a corresponding controlled response strategy. Combined with the current content intent, it generates target content. Under predetermined knowledge boundary constraints, it develops differentiated response methods for knowledge atoms at different cognitive stages, ensuring both the orderliness and controllability of knowledge transfer while avoiding cognitive gaps for users. The generated content is both relevant to the user's current needs and cognitive level. Furthermore, it performs constrained generation on the device side, without relying on large amounts of cloud computing resources, improving response speed and balancing personalization, controllability, and device-side feasibility. This method enables novel knowledge transfer adapted to knowledge-based books, operating under predetermined knowledge boundary constraints. It can identify the user's current cognitive state, control the order of knowledge release, and execute constrained adaptive knowledge transfer on the device side. Attached Figure Description

[0008] To more clearly illustrate the technical solutions of the embodiments of this disclosure, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this disclosure. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0009] Figure 1 A schematic flowchart illustrating the edge-side adaptive knowledge transfer method for knowledge-based books provided in this embodiment of the disclosure.

[0010] Figure 2 This is a flowchart illustrating a method for generating a user cognitive state vector based on user interaction information related to target knowledge books, as provided in an embodiment of this disclosure.

[0011] Figure 3 This is a flowchart illustrating the partitioning method provided in an embodiment of the present disclosure. Detailed Implementation

[0012] The embodiments of this disclosure will now be described in detail with reference to the accompanying drawings.

[0013] Reference Figure 1 This application discloses an edge-side adaptive knowledge transfer method adapted for knowledge-based books. It is a specific edge-side adaptive knowledge transfer method based on knowledge boundary detection and knowledge release rhythm control for knowledge-based books. The method specifically includes: S100 involves structurally decomposing the target knowledge books to obtain a structured knowledge representation.

[0014] It should be noted that the target knowledge books in this application specifically refer to knowledge-type books with a clear chapter structure, consistent authorial stance, clear prerequisite dependencies, and clear knowledge boundaries. They do not include non-book knowledge carriers such as generalized course resources and multi-source popular science content, nor do they include non-knowledge-type books without a consistent stance or clear knowledge boundaries.

[0015] The target knowledge book can be a pre-generated intelligent book content package or a structured content package; in other implementations, it can also be obtained from electronic text, PDF, EPUB, Markdown, Word, or scanned and recognized text after structured conversion.

[0016] The structured knowledge representation includes at least knowledge atoms, a prerequisite dependency graph, a chapter skeleton, misconception rules, and knowledge release rules. Prerequisite dependencies record the dependencies between knowledge atoms, including at least the source knowledge atom, the target knowledge atom, the dependency type, and the dependency strength. The chapter skeleton records the book's chapter structure, chapter progression order, and the mapping relationship between chapters and knowledge atoms. Misconception rules record common misunderstandings, typical error paths, corresponding correction strategies, and prerequisite correction paths. Knowledge release rules record the release conditions, release order, compression conditions, expansion conditions, and delay conditions for each knowledge atom under different user cognitive states.

[0017] In this step, fragmented knowledge content is broken down into standardized, identifiable knowledge atoms, breaking the limitations of traditional linear reading and the inability to break down knowledge. This allows knowledge transfer to shift from holistic indoctrination to precise matching, while reducing the processing difficulty of knowledge content on the device side, improving data processing efficiency, and solving the drawbacks of technical professional books being complex and difficult to accurately locate core knowledge.

[0018] S200 generates a user cognitive state vector based on user interaction information related to the target knowledge book.

[0019] This step can accurately capture the user's initial level of understanding of the book, points of interest, and weaknesses, avoiding a one-size-fits-all knowledge delivery model, and making subsequent knowledge push and response more in line with the user's personalized needs.

[0020] S300 performs knowledge boundary detection on all knowledge atoms based on the user's cognitive state vector and structured knowledge representation, and divides all knowledge atoms into a set that has been mastered, a set that can be entered, and a set that cannot be entered for the time being.

[0021] This step clarifies the boundaries of knowledge transfer, effectively preventing users from accessing content beyond their current cognitive level and avoiding repeated pushes of content they have already mastered, thus achieving controllability in knowledge transfer. At the same time, the categorized knowledge set can reduce the real-time computing pressure on the device side, taking into account the lightweight deployment requirements of the device side, and is especially suitable for the step-by-step learning pattern of technical books.

[0022] S400 responds to interactive response commands, identifies the intent of the current content, and determines the target knowledge atom based on the intent of the current content.

[0023] This step can quickly capture the user's current learning needs, accurately locate the knowledge content the user needs, avoid ineffective push notifications, and improve the relevance and smoothness of the interaction.

[0024] S500 performs knowledge access determination on each target knowledge atom based on the known set, the accessible set, and the temporarily inaccessible set, determines the corresponding set type, and executes the corresponding controlled response strategy according to the set type.

[0025] This step, under the constraints of established knowledge boundaries, can formulate differentiated response methods for knowledge atoms at different cognitive stages. For example, content that has been mastered can be briefly reviewed, content that can be accessed can be explained in detail, and content that cannot be accessed at the moment can be prompted with prerequisite knowledge. This ensures the orderliness and controllability of knowledge transfer, avoids cognitive gaps for users, and the controlled strategy can reduce the complexity of content generated on the device side, ensuring that the device side can execute efficiently and balancing personalization and device side feasibility.

[0026] S600 generates target content based on the current content intent and controlled response strategy.

[0027] This step enables a closed-loop adaptation of needs, cognition, strategy, and content. The generated content not only meets the user's current needs but also aligns with their cognitive level. Furthermore, it completes limited generation on the device side without relying on a large amount of cloud computing resources, thus improving response speed. It balances personalization, controllability, and device-side feasibility, effectively addressing the core shortcomings of poor knowledge transfer adaptability and weak interactivity in technical professional books.

[0028] Specifically, the method for S100 to structurally decompose the content of target knowledge books includes: S110: Segment the target knowledge book according to paragraphs and sentence groups to determine several candidate segments.

[0029] Each candidate segment includes at least a book identifier, a chapter identifier, a paragraph identifier, and a sentence / segment range identifier.

[0030] Specifically, based on the inherent paragraph divisions of the target knowledge book, each independent paragraph is used as the initial segmentation unit. Here, "paragraph" refers to the natural paragraphs in the book's layout, that is, text blocks explicitly divided by the author or editor and separated by line breaks. In knowledge books, paragraph layout typically follows a single semantic focus principle; a paragraph usually revolves around a core knowledge point, a viewpoint, or a logical link, meeting the basic requirement of semantic integrity for candidate segments.

[0031] If a paragraph is too long, such as exceeding 500 words, and contains multiple independent semantics, such as simultaneously involving different content like concept definitions, examples, and precautions, then the paragraph should first be split into several sub-paragraphs, each sub-paragraph corresponding to an independent semantic, and then the sub-paragraphs should be used as the initial segmentation units.

[0032] If a paragraph is too short, such as less than 50 words, and its meaning is incomplete, such as being merely a transitional sentence or supplementary annotation and unable to independently carry a knowledge fragment, then it should be merged with adjacent paragraphs that are semantically closely related to form an initial segmentation unit to avoid semantic breaks caused by fragmented segmentation.

[0033] Based on basic paragraph segmentation, precise optimization is performed using sentence group logic to ultimately determine candidate segments. A sentence group is a collection of semantically related and logically coherent sentences that collectively express a complete meaning, such as concept explanation sentence groups, principle derivation sentence groups, and case analysis sentence groups. During identification, sentences are used as the basic unit. Connective words such as "firstly," "secondly," "again," "because," "therefore," "for example," and "in addition," semantically repetitive words, and logical relationships such as parallel, progressive, causal, and adversative relationships are used to determine the degree of connection between sentences and thus identify independent sentence groups.

[0034] The identified sentence groups are matched with the basic segmented paragraphs. If a sentence group corresponds to a paragraph, the paragraph is directly identified as a candidate segment.

[0035] If a sentence group spans multiple paragraphs, such as sentences in two adjacent paragraphs that are semantically closely related and together form a complete sentence group, then these multiple paragraphs are merged to form a candidate segment. If a paragraph contains multiple independent sentence groups, such as a paragraph that contains both a concept definition and an example illustration, and the two parts are semantically independent and not strongly related, then the paragraph is split into multiple candidate segments, each candidate segment corresponding to an independent sentence group.

[0036] Furthermore, the adjusted candidate segments are verified by combining the knowledge boundaries and chapter framework of the target knowledge book to ensure that each candidate segment belongs to the core knowledge category of the book and does not contain irrelevant or redundant content, such as acknowledgments or publication notes in the preface that are not related to the core knowledge, or irrelevant summaries and reflections in the postscript; at the same time, it is ensured that there is no semantic overlap between candidate segments and that they cover the core content of the corresponding chapters.

[0037] S120, perform trigger mode analysis on each candidate fragment, determine the trigger type, and determine the atom type corresponding to each candidate fragment based on the trigger type.

[0038] S130: Extract the corresponding knowledge atoms based on the atom type, and determine the knowledge release rules for each knowledge atom.

[0039] If the fragment contains phrases such as "refers to", "defined as", "represents", or "that is", the trigger type is determined to be the first type of trigger mode, the corresponding atom type is definition type, and it is extracted as a definition type knowledge atom.

[0040] If the fragment contains phrases such as "should," "must not," "needs to be satisfied," or "only when," the trigger type is determined to be the second type of trigger mode, and the corresponding atom type is boundary type. Then, it is extracted as a rule type or boundary type knowledge atom.

[0041] If the fragment contains phrases such as "step one," "first," "then," or "finally," the trigger type is determined to be the third type of trigger mode, and the corresponding atom type is step-type. In this case, the knowledge atom is extracted as a step-type atom.

[0042] If words such as "for example," "like," "example," or "case" appear in the fragment, the trigger type is determined to be the fourth type of trigger mode, and the corresponding atom type is example type. Then, the example type knowledge atom is extracted.

[0043] If the fragment contains phrases such as "common errors, misconceptions, easily misunderstood, do not assume", the trigger type is determined to be the fifth type of trigger mode, and the corresponding atom type is misconception type. Then, the knowledge atom is extracted as misconception type.

[0044] Among them, the knowledge release rules are used to record the release conditions, skip conditions, compression conditions, expansion conditions, backfill conditions, delay conditions, and upper limit of output depth for each knowledge atom.

[0045] Furthermore, the release condition is the trigger condition for determining whether a knowledge atom can be output. It is the core prerequisite for a knowledge atom to be presented and transmitted. The core of the determination method revolves around pre-dependencies and scenario adaptation. Specifically, it focuses on pre-dependencies. If knowledge atom A has pre-dependent knowledge atoms B and C, then the release condition of knowledge atom A must include the successful release of pre-dependent knowledge atoms B and C, meaning the user has mastered B and C. For basic knowledge atoms without pre-dependent dependencies, the release condition can be set to entering the corresponding chapter's knowledge scenario, such as entering the function definition chapter, which will release the basic knowledge atom of functions. The release condition can be supplemented according to the purpose of the knowledge atom. For example, for practical knowledge atoms such as formula application and step breakdown, the release condition can be supplemented by entering a practical scenario; for theoretical knowledge atoms, the release condition can be supplemented by completing the corresponding theoretical background scenario.

[0046] Skip conditions are used to determine which knowledge atoms can be skipped from output, avoiding redundant output and ensuring efficient knowledge transfer. Specifically, if a knowledge atom belongs to content that the user has already clearly mastered, it is set as a skip condition, such as when the user can accurately describe the core content of the knowledge atom.

[0047] If the knowledge atom belongs to the extended level of content, and the current knowledge transfer scenario only requires mastering the basic level of content, such as the introductory stage, then a skip condition is set. For example, if the current learning scenario is basic, there is no need to master the extended knowledge.

[0048] If a knowledge atom is not directly related to the current core learning objective, such as a supplementary annotation knowledge atom within a chapter, then a skip condition is set. For example, if the current learning objective is the core principle, supplementary annotation knowledge atoms can be skipped.

[0049] Compression conditions are used to determine the conditions under which the output of knowledge atoms needs to be simplified. These conditions are used to simplify the output content of knowledge atoms without affecting core understanding, and the methods for determining this are based on scenario requirements and knowledge priorities. Specifically, if the current knowledge delivery scenario is to quickly organize a knowledge framework or for emergency review, compression conditions are set for non-core knowledge atoms. For example, when it is necessary to quickly build a chapter's knowledge framework, the examples and extended content of that knowledge atom are compressed, retaining only the core definitions.

[0050] If the knowledge atom is non-key content, and the current focus should be on key knowledge such as in exam preparation scenarios, then compression conditions should be set, such as prioritizing the output of key knowledge and compressing non-key knowledge atoms into core keywords.

[0051] Expanding conditions determine the detailed output required for a knowledge atom. These conditions, in contrast to compression conditions, are used in scenarios requiring a deeper understanding of the knowledge atom. Specifically, expanding conditions are set for core knowledge atoms such as key concepts or core principles within a chapter. If a user has questions about the understanding of a knowledge atom (e.g., the pre-test reveals the user's lack of grasp), expanding conditions are set; if the user's understanding of the knowledge atom is insufficient, detailed explanations and supporting examples are provided. If a knowledge atom is a crucial foundation for subsequent advanced knowledge and the user is currently in an advanced learning scenario, expanding conditions are set; if further advanced knowledge is required, the details and application scenarios of the preceding knowledge atom are expanded.

[0052] The supplementary condition is the condition for determining the supplementary output of a knowledge atom. It is used to make up for omissions in knowledge transmission and ensure the integrity of the knowledge system. The determination method includes: if it is found that the user cannot understand the subsequent content because he has not mastered the knowledge atom during the release of subsequent knowledge atoms such as advanced knowledge, then a supplementary condition is set. For example, when releasing advanced knowledge atoms, if it is detected that the user has not mastered the knowledge atom, the core content of the knowledge atom is supplemented and output.

[0053] If a knowledge atom contains supplementary content, such as corrections or extensions added by the author in subsequent chapters, a supplementary condition is set. If the supplementary content related to that knowledge atom is triggered, the supplementary information is output.

[0054] The delay condition is the condition for determining the delayed output of a knowledge atom. It is used to adjust the output order of knowledge atoms to ensure that knowledge transfer is logical. The determination method includes: if the preceding dependent knowledge atom of a knowledge atom has not been released, i.e., the user has not mastered it, then the delay condition is set. If the preceding knowledge atom has not been released, the output of this knowledge atom is delayed.

[0055] If the current scenario is not suitable for outputting the knowledge atom, such as outputting the core concept first and then delaying the output of related cases, then a delay condition is set. For example, if the current scenario is a concept definition scenario, the case analysis content of the knowledge atom is output later.

[0056] The upper limit of output depth is used to determine the boundary of the level of detail of the knowledge atom output. It is used to control the output range of the knowledge atom and avoid over-expansion or insufficient output. The determination method is based on the knowledge level and scenario requirements.

[0057] Specifically, for basic knowledge atoms, the upper limit of output depth should preferably be set to the core definition and simple examples, without excessive extension.

[0058] For advanced knowledge atoms, the upper limit of output depth is preferably set to the core definition, derivation process, typical cases and precautions.

[0059] For extended-level knowledge atoms, the upper limit of output depth should preferably be set to the core content, brief extensions, and reference directions, so as not to occupy too much transmission resources.

[0060] Furthermore, different upper limits for output depth can be set according to different scenarios. For example, in the introductory scenario, the upper limit for the output depth of all knowledge atoms is controlled at the basic core content; in the in-depth learning scenario, the upper limit for the output depth of core knowledge atoms can be increased to detailed analysis and extension, while non-core knowledge atoms maintain the basic depth.

[0061] In this application, a knowledge atom is the smallest unit of knowledge after the target book has been broken down. Each knowledge atom includes at least an atom identifier, atom content, its chapter, difficulty level, topic tag, and term tag.

[0062] S140, based on the preceding and following relationships between different knowledge atoms, determine the dependency type and dependency strength of the dependency edges between related different knowledge atoms, and generate a preceding dependency relationship graph and corresponding misconception rules.

[0063] Specifically, strong dependency edges are generated based on explicit expressions such as "based on...", "after understanding...", "the premise is...", and "must first master...". The core of strong dependency edges is that the source knowledge atoms must be mastered before the target knowledge atoms can be understood, which is entirely based on the explicit expressions in the original text of the book.

[0064] Furthermore, the process begins by screening for explicit association cases. From the previously labeled explicit association clues, all knowledge atom pairs that conform to the preceding-following logic are extracted. For example, if the original text states that understanding the 'function definition' (knowledge atom A) is necessary to grasp the 'monotonicity of a function' (knowledge atom B), then the knowledge atom pair (A→B) is extracted. Similarly, if the original text states that "mastering the 'Pythagorean theorem' (knowledge atom C) is a prerequisite for applying 'right triangle area calculation' (knowledge atom D)," then the knowledge atom pair (C→D) is extracted. Then, the source / target knowledge atoms, dependency type, and dependency strength of strongly dependent edges are determined.

[0065] The corresponding minimum mastery threshold can be set to complete mastery. For example, if the mastery level is 1-10, the threshold is 10, meaning that the source knowledge atom must be fully understood and accurately expressed in order to trigger the release of the target knowledge atom.

[0066] Term dependency edges are generated based on term coverage relationships. The core of term dependency edges is the core terms contained in the target knowledge atom, which need to be explained or defined by the source knowledge atom. That is, the source knowledge atom is the terminology basis of the target knowledge atom. Specifically, knowledge atom pairs of term explanation and term application can be extracted from the term association clues annotated in the early stage. The criterion for judgment is that the source knowledge atom clearly explains / defines the core terms that appear in the target knowledge atom.

[0067] The corresponding minimum mastery threshold is preferably set to mastering the core definition. For example, if the mastery level is 1-10, the threshold is 8 points. That is, the core definition of the term in the source knowledge atom needs to be mastered, and there is no need to master the extended content to support the understanding of the target knowledge atom.

[0068] Based on the misconception rule that a lack of understanding of A leads to misunderstanding of B, a corrective dependency edge is generated. The core of the corrective dependency edge is that the source knowledge atom can correct common misunderstandings of the target knowledge atom. This is done by working backward from the misconception rule that a lack of understanding of A leads to misunderstanding of B. Specifically, the misconception rule is first improved by combining the common mistakes and suggestive statements in the original text of the book. Each misconception rule needs to clearly define the knowledge atom that is not understood (A), the knowledge atom that is easily misunderstood (B), and the specific content of the misunderstanding. For example, misconception rule 1 is that a lack of understanding of the basic properties of fractions (knowledge atom I) and a misunderstanding of the principle of reducing fractions (knowledge atom J) lead to the misunderstanding that reducing fractions changes the size of the fraction. Misconception rule 2 is that a lack of understanding of the three elements of force (knowledge atom K) and a misunderstanding of the effect of force (knowledge atom L) lead to the misunderstanding that only the magnitude of the force affects the effect.

[0069] Based on the premise that a lack of understanding of A easily leads to misunderstanding of B, we can deduce that mastering A can correct misunderstandings of B. Therefore, the source knowledge atom of the corrective dependency edge is A, and the target knowledge atom is B, i.e., A→B. For example, based on misunderstanding rule 1, we can deduce the corrective dependency edge: I→J, meaning that mastering I can correct misunderstandings of J. Based on misunderstanding rule 2, we can deduce the corrective dependency edge: K→L, meaning that mastering K can correct misunderstandings of L.

[0070] The corresponding minimum mastery threshold can be set as mastering the core points. For example, if the mastery level is 1-10 points, the threshold is 6 points. That is, the core corrective points of the source knowledge atom need to be mastered, so as to avoid common misunderstandings of the target knowledge atom.

[0071] In this embodiment, each dependency edge preferably includes at least the source knowledge atom, the target knowledge atom, the dependency type, the dependency strength, whether it is mandatory to prioritize, and the minimum mastery threshold.

[0072] S150: Extract the table of contents of the target knowledge book and generate chapter skeletons based on the prior dependency graph.

[0073] Specifically, the table of contents for the target knowledge book is compiled, divided into chapters according to hierarchy, clearly defining chapter identifiers, parent chapter identifiers, and chapter order to ensure consistency with the book's layout and to ensure no chapter is omitted. Based on the dependency graph and the generated dependency edges, the chapter entry point, key points, and summarized knowledge are identified. Considering the depth of knowledge in the book, the maximum allowable depth for each chapter is set, aligning with the chapter's knowledge level; basic chapters are set to basic depth, and advanced chapters to intermediate or higher depth. The chapter objectives are clarified. All the above information is organized and integrated to form a complete chapter framework, ensuring that all elements are consistent with the table of contents hierarchy and dependency relationships, and that the logic is coherent.

[0074] S160, extract author expression constraint information.

[0075] Specifically, the author's expression of constraint information includes at least a glossary and preferred usage of terms, definition sentence templates, explanation order preferences, prohibited alternative expressions, common example styles, and risk warning tone patterns.

[0076] Through the above steps, a structured knowledge representation can be obtained for direct invocation during knowledge boundary detection and restricted generation triggering by the readers of this application.

[0077] Reference Figure 2 The method for generating user cognitive state vectors based on user interaction information related to target knowledge books, as described in S200, specifically includes: S210 collects user interaction information related to the target knowledge book.

[0078] The user interaction information includes at least one or more of the following: exploratory question-and-answer results, historical reading records, historical answer results, historical question-and-answer trajectory, user-selected familiarity tags, current reading trajectory, number of times the content was reviewed, confirmation of the explanation, denial of the explanation, and repeated follow-up questions about the explanation.

[0079] S220 maps user interaction information to corresponding knowledge atoms and generates atomic-level state features for each knowledge atom.

[0080] Specific mapping methods include: if a user reads a paragraph, the mapping is to the knowledge atom bound to that paragraph; if a user answers a question, the mapping is to the knowledge atom associated with that question; if a user asks a question, the question is mapped to one or more target knowledge atoms through terminology matching, sentence vector similarity matching, or rule matching; if a user actively selects tags such as "familiar" or "unfamiliar," the mapping is to the knowledge atom corresponding to the current reading position or the current question.

[0081] The atomic-level state features corresponding to each knowledge atom include at least: whether it has been read, the score of the related questions, the user's self-assessment of familiarity, the degree of confirmation of the explanation, the number of repeated follow-up questions around the atom, whether the misconception rule has been hit, the average mastery of its preceding knowledge atoms, and the time decay value since the last effective interaction.

[0082] S230 determines the user's depth of understanding and interaction stability based on user interaction information.

[0083] The level of understanding is defined as any one of the following levels: Level 1 (untouched class), Level 2 (able to restate the definition class), Level 3 (able to explain the principle or steps class), Level 4 (able to apply to the example class), and Level 5 (able to identify boundary errors and counterexamples class).

[0084] Specifically, if there is no reading and no related interaction records such as no answers, no Q&A, and no familiarity tag selection, it means that the user has not been exposed to the relevant content of this knowledge atom. The user's understanding depth is determined to be in the "unexposed" category, and the corresponding score is preferably 0.

[0085] If a user has read and self-assessed their familiarity with the knowledge and can accurately paraphrase the core definition in probing questions about the knowledge atom, but can only answer basic definition-related questions and cannot explain the principles behind the definition or apply it, then the user's understanding depth is determined to be able to paraphrase the definition, and the corresponding score is 0.2.

[0086] If all conditions for restating the definition class are met and the core principle, derivation process, or operation steps can be clearly explained in the questions and answers for that knowledge atom, without repeated follow-up questions about the principle / step, the user's understanding depth is determined to be able to explain the principle or step class, and the corresponding score is preferably 0.4.

[0087] If all conditions for explaining the principle or steps are met and the relevant answers can apply the knowledge atomically to specific examples or cases, and the accuracy rate of completing application-type questions is ≥80%, the user's understanding depth is determined to be able to apply it to example types, and the corresponding score is preferably 0.7.

[0088] If all conditions for application to the example class are met, and the user can actively identify the boundary conditions and common misconceptions of the knowledge atom in the question and answer process, accurately judge and explain the reasons for counterexample questions, and there is no repeated confusion around the knowledge atom, the user's understanding depth is determined to be able to identify boundary misconceptions and counterexample classes, and the corresponding score is 1.

[0089] It should be noted that if a user's interaction behavior meets the judgment criteria of multiple levels at the same time, the highest level shall prevail; if there are contradictions in the interaction behavior, such as self-assessment of familiarity but extremely low accuracy in answering questions, the objective interaction results, such as answering questions or Q&A, shall prevail.

[0090] Interaction stability is calculated based on whether the results of the most recent interactions with the knowledge atom are consistent, and its value ranges from 0 to 1, where 0 represents completely unstable and 1 represents completely stable.

[0091] Specifically, the method for determining interaction stability includes: selecting the most recent preset number of valid interaction results for a knowledge atom. Valid interactions include: answers to related questions, confirmation / negation of explanations, user self-assessment of familiarity, and Q&A feedback surrounding the atom. Invalid interactions, such as erroneous operations and duplicate submissions, are excluded. For the selected interaction samples, the consistency coefficients for answer accuracy, confirmation of explanations, and self-assessment of familiarity are determined, and the average value is used to obtain the final interaction stability.

[0092] The accuracy consistency coefficient is calculated by calculating the accuracy of the most recent preset number of questions, and the difference between the maximum and minimum values ​​is calculated. The smaller the difference, the higher the consistency. The difference is normalized to between 0 and 1 to obtain the accuracy consistency coefficient. Accuracy consistency coefficient = 1 - normalized difference.

[0093] The consistency coefficient of the interpretation is calculated as follows: if in multiple consecutive interactions, the user's interpretation of the knowledge atom is either affirmative or negative, the consistency coefficient is 1; if there are two or more feedbacks, the coefficient is calculated based on the proportion of consistent feedbacks, i.e., the number of consistent feedbacks / the total number of samples.

[0094] To calculate the self-assessed familiarity consistency coefficient, users convert their self-assessed familiarity level (e.g., 1-5 points) into numerical values, calculate the difference between the maximum and minimum values, normalize the result to obtain the self-assessed familiarity consistency coefficient, which is 1 - the normalized difference.

[0095] The final interaction stability is obtained by averaging the above consistency coefficients, which is (accuracy consistency coefficient + confirmation consistency coefficient + familiarity consistency coefficient) / number of indicators.

[0096] Furthermore, if the number of effective interactions for the knowledge atom is less than the preset number, for example, less than 3 times, the interaction stability is preferably 0.5, that is, the default is medium stability, and the calculation is recalculated after the number of interactions reaches the target.

[0097] S240 determines the mastery level of each knowledge atom based on atomic-level state characteristics, understanding depth, and interaction stability.

[0098] Specifically, firstly, the ratio of the reading time of the paragraph corresponding to the knowledge atom to the standard reading time of that paragraph (i.e., the preset reasonable reading time) is calculated and recorded as the normalized reading completion value. Then, the scores of the relevant questions are used as the basis for the calculation. Normalized reading completion rate User self-assessment of familiarity Degree of confirmation of the explanation The average mastery of prerequisite knowledge atoms Interaction stability Whether the misconception rules are hit The number of repeated follow-up questions about this atom Time decay value Determine the level of mastery corresponding to each knowledge atom.

[0099] No. i The mastery level corresponding to each knowledge atom is: : in, This means cropping the result to between 0 and 1. , ... These are the corresponding weights, and their sum is 1.

[0100] Furthermore, for whether something has been read, a binary score is preferred: read means the user has viewed the paragraph corresponding to the knowledge atom, which scores 1; unread means there is no browsing record, which scores 0.

[0101] For the relevant questions, the score is 0 if there is no relevant answer record; if there is an answer record, the score = the correct answer rate of the relevant knowledge atom = the number of correct answers / the total number of questions, no additional conversion is required.

[0102] For users' self-assessment of familiarity, the preferred scale is 4 levels: 0 for very unfamiliar, 0.3 for somewhat unfamiliar, 0.7 for fairly familiar, and 1 for very familiar.

[0103] For the degree of confirmation of the explanation, a binary score is preferred, which is determined based on user feedback. If the user confirms and understands (e.g., clicks "understand" or "confirm"), the score is 1. If the user does not confirm or deny (e.g., no feedback or clicks "do not understand"), the score is 0.

[0104] For the number of repeated follow-up questions around this atom, the score is 0 if there are no repeated follow-up questions; if there are repeated follow-up questions, the score is equal to the actual number of follow-up questions, such as 1 time = 1, 2 times = 2, with a maximum of 3. Any number exceeding 3 will be counted as 3.

[0105] Regarding whether a user's behavior matches the misconception rule, if it does, the user's behavior matches the misconception rule corresponding to that atom and the score is 1; if it does not match, the user's behavior does not match the relevant misconception rule and the score is 0.

[0106] Regarding the average mastery of prerequisite knowledge atoms, if there are no prerequisite knowledge atoms, the score is 1, which means there are no prerequisite dependencies and does not affect subsequent calculations; if there are prerequisite knowledge atoms, the score is the average mastery score of all prerequisite knowledge atoms.

[0107] For the time decay value since the last effective interaction, it is preferable to simplify the assignment according to the time range and normalize it to 0-1. Specifically, the score is 0.1 for effective interaction within 1 day, 0.3 for effective interaction within 1-7 days, 0.6 for effective interaction within 7-30 days, and 0.9 for no effective interaction for more than 30 days. The higher the score, the more obvious the time decay.

[0108] It should be noted that the above formula is only a preferred embodiment. Those skilled in the art can flexibly determine the specific formula for the mastery level corresponding to each knowledge atom according to their needs. As long as it can reflect the mastery level corresponding to each knowledge atom, it is within the protection scope of this application.

[0109] S250 determines the probability of a misconception for each knowledge atom based on a misconception rule base and user interaction performance.

[0110] No. i The probability of error corresponding to each knowledge atom is: : ,in, , ... These are the corresponding weights, and their sum is 1.

[0111] The score corresponds to whether the user's incorrect answer falls into a typical misconception pattern of the atom; specifically, if the user's incorrect answer completely matches a typical misconception pattern of the atom, the score is 1; if the user has no incorrect answer or the incorrect answer does not fall into any typical misconception pattern, the score is 0.

[0112] The score represents whether the user interaction matches any of the corresponding misconception rules. Specifically, if the user interaction matches any of the misconception rules corresponding to the atom, the score is 1; if the user interaction does not match any of the misconception rules corresponding to the atom, the score is 0.

[0113] The score corresponds to whether the user repeatedly expresses confusion about the same knowledge point. Specifically, if the user asks the same question repeatedly for the same atom ≥2 times, reviews the corresponding paragraph for the atom ≥2 times, or answers the related questions for the same atom ≥2 times with the same reason for the error, then the score is 1. If the user does not express confusion multiple times as mentioned above, or only expresses confusion once, then the score is 0.

[0114] The score corresponds to whether the user's expression conflicts with the rules in the book; specifically, if the user's expression clearly conflicts with the rules in the book, such as denying the definitions in the book or misunderstanding the principle statements, then it is 1; if the user has no relevant expression, or the expression is consistent with the rules in the book and does not conflict, then it is 0.

[0115] The score corresponds to whether the user has confirmed their understanding after the correction process. Specifically, if the user confirms their understanding after the correction process, such as by actively providing feedback that they understand, or by providing correct answers or Q&A regarding that atom without similar errors, the score is 1. If the user has not undergone correction, or has not confirmed their understanding after correction, such as by providing feedback but still not understanding, or by continuing to make similar errors, the score is 0.

[0116] S260 generates a user cognitive state vector corresponding to each knowledge atom based on mastery, probability of misconception, and interaction stability.

[0117] In this embodiment, the user's cognitive state vector preferably includes mastery level, probability of misconception, and interaction stability.

[0118] Reference Figure 3 For S300, knowledge boundary detection is performed on all knowledge atoms based on the user's cognitive state vector and structured knowledge representation, and all knowledge atoms are divided into a mastered set, an accessible set, and a temporarily inaccessible set. Specifically, the division method includes: S310: Obtain the knowledge atoms in the user's cognitive state vector that satisfy the first preset condition, denot them as the mastered class, and classify them into the mastered set.

[0119] The first preset conditions include: the level of mastery is not lower than the mastery threshold, the probability of misconception is not higher than the misconception threshold, and the interaction stability is not lower than the stability threshold, or the user actively marks it as familiar and it has passed verification by at least one exploratory question.

[0120] In this embodiment, when the user's cognitive state vector satisfies any one of the first preset conditions, the corresponding knowledge atom can be assigned to the mastered set.

[0121] S320, determine the prerequisite satisfaction of knowledge atoms of non-mastered classes based on the mastery degree of knowledge atoms of non-mastered classes and the dependency strength of corresponding dependency edges.

[0122] The prerequisite satisfaction degree of the i-th non-already-mastered knowledge atom is : ,in, This represents the dependency strength of the corresponding dependent edge.

[0123] Furthermore, if there is a mandatory prerequisite dependency and the mastery of the corresponding prerequisite atom is lower than the minimum mastery threshold, it is recorded as a mandatory prerequisite not satisfied, and the prerequisite satisfaction of the corresponding non-mastered knowledge atom is recorded as 0.

[0124] S330, determine the release verification status of knowledge atoms that are not already mastered.

[0125] Among them, the release verification status preferably includes chapter permission, release permission, and misconception blocking status. All three are to verify whether the knowledge atom can be released normally and output to the user. They together constitute the verification conditions before the knowledge atom is released from the three dimensions of chapter progress constraints, release depth restrictions, and misconception risks.

[0126] The chapter permissibility indicates whether the presentation of the knowledge atom conforms to the progression constraints of the current chapter's framework, i.e., whether the current chapter is allowed to present the knowledge atom, avoiding premature / delayed output. If the chapter to which the knowledge atom belongs is the same as the chapter the user is currently reading or is a prerequisite chapter for the current chapter, then the chapter permissibility is determined to be permissible, preferably denoted as 1. If the chapter to which the knowledge atom belongs is unrelated to the chapter the user is currently reading or belongs to a chapter that has not yet been progressed, then the chapter permissibility is determined to be permissible, preferably denoted as 0.

[0127] Release permissibility determines whether the release level of a knowledge atom does not exceed the maximum release depth allowed by the current system. Specifically, first, the release level of the corresponding knowledge atom is determined, which can correspond to the comprehension depth; for example, 1 for no exposure, 2 for retelling, and so on. Then, the maximum release depth allowed by the current system is determined, which can be derived from the chapter skeleton; for example, the maximum depth of the current chapter is set to 3. If the release level is less than or equal to the maximum allowed release depth, the release permissibility is determined to be permissible, preferably denoted as 1. If the release level is greater than the maximum allowed release depth, the release permissibility is determined to be disallowed, preferably denoted as 0.

[0128] The "Misconception Blocking Status" is used to determine whether a knowledge atom cannot be released normally temporarily because the current misconception probability is too high, requiring correction to resolve the misconception or supplementation of prior knowledge. Specifically, a simple threshold is set: if the misconception probability is greater than or equal to the preset threshold, it indicates that the misconception is too high, and the misconception blocking status is set to 1; if the misconception probability is less than the preset threshold, it indicates that the misconception is controllable, and the misconception blocking status is set to 0. If there is no misconception probability data, the default misconception blocking status is 0.

[0129] S340: Obtain the knowledge atom that satisfies the prerequisite satisfaction and releases the verification state that satisfies the second preset condition, denoted as the enterable class and assigned to the enterable set.

[0130] The second preset conditions include: the prerequisite satisfaction is not lower than the preset threshold, there is no forced prerequisite non-satisfaction, the chapter permission is true (i.e., 1), the release permission is true (i.e., 1), and the misconception blocking status is false (i.e., 0).

[0131] S350, the remaining knowledge atoms are assigned to a set that cannot be entered for the time being.

[0132] In this embodiment, knowledge atoms are divided into three sets: mastered, accessible, and temporarily inaccessible. This enables precise and efficient knowledge delivery, aligning with users' cognitive patterns. Mastered knowledge is filtered using a first preset condition, avoiding redundant pushes, reducing ineffective interference, and alleviating learning fatigue. The pre-satisfaction level is calculated by combining the mastery and dependence strength of non-mastered knowledge, and the release verification status is checked to filter accessible knowledge. This ensures that the content matches the user's cognitive foundation and chapter pace, avoiding problems such as premature learning and misconceptions. Simultaneously, temporarily inaccessible knowledge is clearly identified, prioritizing learning and guiding users to learn step-by-step. The division into three sets provides a clear basis for subsequent knowledge release, facilitating personalized instruction, effectively avoiding cognitive confusion, reducing comprehension obstacles, helping users solidly master knowledge, and improving the overall knowledge delivery effect.

[0133] The S400 method of "responding to interactive response commands, identifying the current content intent, and determining the target knowledge atom based on the current content intent" specifically includes: responding to user interactive response commands by first identifying the current content intent, which includes at least explanation requests, question-and-answer requests, example requests, prompt requests, and supplementary requests. Then, based on the current reading position, the user's question text, the recent interaction context, and the chapter skeleton, the current request is mapped to one or more target knowledge atoms.

[0134] For S500, "Based on the already mastered set, the accessible set, and the temporarily inaccessible set, perform knowledge access judgment on each target knowledge atom, determine the corresponding set type, and determine the controlled response strategy corresponding to the set type." If the target knowledge atom belongs to the already mastered set, the controlled response strategy corresponding to the set type is the first strategy, which prioritizes generating compressed summaries, transfer applications, or quick review content. If the target knowledge atom belongs to the accessible set, the controlled response strategy corresponding to the set type is the second strategy, which allows the generation of content within the boundary. If the target knowledge atom belongs to the temporarily inaccessible set, the controlled response strategy corresponding to the set type is the third strategy, which prohibits the direct output of the target answer and switches to the out-of-bounds processing flow.

[0135] For the target knowledge atoms that are allowed to enter, a restricted generative material package can be constructed.

[0136] The material package may only contain: the standard expression of the current target knowledge atom, the source fragment of the current target knowledge atom in the book, the related knowledge atoms adjacent to the target knowledge atom in the currently accessible set, the summary of the preceding knowledge atoms in the already mastered set that can serve as connecting bridges, the context that can be cited in the chapter skeleton, the knowledge release rules corresponding to the target knowledge atom, author style constraint information, and one or more of the following in the user's cognitive state related to the target knowledge atom: mastery, depth of understanding, probability of misconception, and readiness for release.

[0137] When the target knowledge atom is in a set that cannot be entered temporarily, and the reason for the block is that the prior knowledge is not mastered or the probability of misconception is too high, supplementary content is generated.

[0138] Specifically, firstly, the target knowledge atom is searched backward along the prior dependency graph to find all prior knowledge atoms that are not yet mastered and whose dependency strength is higher than the threshold; then, the minimum necessary backfill set is selected by combining the chapter skeleton and the dependency topology order; then, backfill explanations, backfill questions, backfill examples or backfill hints are generated in sequence; after the user completes the backfill, the cognitive state of the relevant knowledge atoms is recalculated and the set division is updated.

[0139] If the target knowledge atom is located in the set that cannot be entered temporarily, the target answer will not be output directly. Instead, at least one out-of-bounds handling strategy will be executed according to the reason for the blockage. The out-of-bounds handling strategy is to refuse to answer directly, output a lower-level explanation, guide the user to supplement the prior knowledge, ask follow-up questions to clarify the user's background, indicate that the content will be released in a later stage, or provide a safe explanation that does not exceed the current boundary.

[0140] The method of S600 for "generating target content based on current content intent and controlled response strategy" includes: determining the intent type of the current content intent; and generating target content based on the intent type and controlled response strategy.

[0141] When the content intent is a request for explanation, the explanation content is generated based on the target knowledge atoms and the user's cognitive state.

[0142] If the user has a low level of understanding of the atom, a definition-level or concept-level explanation will be generated first; if the user's understanding reaches the principle level, a principle-level or step-level explanation will be generated.

[0143] If the probability of user misconception is high, corrective reminders will be inserted into the explanation content; if the knowledge release rules require compression, only the core conclusions and necessary conditions of the atom will be output.

[0144] If the knowledge release rules require expansion, then the definition, function, applicable boundaries, and connections with prerequisite knowledge should be output sequentially. The sentence structure, terminology, and organizational order of the explanation content must conform to the author's style constraints.

[0145] When the content intent is a question-and-answer request, first determine whether all the target knowledge atoms mapped to the user's question are located in the accessible or mastered set. If all are located in the accessible or mastered set, then generate the question-and-answer content; if some are located in the accessible or mastered set and others are located in the temporarily inaccessible set, then only answer the part in the accessible or mastered set, and do not directly answer the part in the temporarily inaccessible set, but output the explanation within the boundary.

[0146] If all values ​​are in the set that cannot be entered temporarily, the target answer will not be given directly, and the process will proceed to out-of-bounds handling.

[0147] In one implementation, the question-and-answer generation template includes at least: the knowledge points in the book corresponding to the current question, the connection relationship with the content already mastered by the user, the current allowed explanation level, hints for parts that will not be expanded, and the next supplementary question if necessary.

[0148] When the content intent is an example request, only the following candidate examples are selected: in-book examples that are bound to the current target knowledge atom itself; low-level examples of related atoms in the currently accessible set; and bridging examples in the already mastered set that can help understand the current knowledge atom.

[0149] If users have a low level of understanding, simple examples should be generated first; if users are likely to make mistakes, comparative or counterexamples should be generated first; if the chapter skeleton prohibits cross-chapter calls, the examples must not introduce knowledge atoms from subsequent chapters.

[0150] When the content is intended to prompt a request, instead of providing a complete answer directly, it generates prompts to help the user continue thinking or locate the prerequisites. These prompts must include at least one of the following: a prompt to review a prerequisite definition; a prompt to confirm a rule condition; a prompt to compare two easily confused concepts; or a prompt to first understand a specific prerequisite knowledge atom.

[0151] The method disclosed in this embodiment simultaneously monitors user questions and content access requests in real time, and quickly identifies requests that exceed the current cognitive scope or knowledge boundaries by combining knowledge boundaries with the user's cognitive level. For requests that exceed the boundaries, a response approach of pre-emptive supplementation, gradual explanation, and rhythm control is adopted. First, the user's missing prior knowledge is supplemented, and then the target content is explained step by step to avoid leaps in understanding. For requests that exceed the knowledge boundaries of the book, the boundary range is clearly stated, and no generalized output is performed.

[0152] After the client-side model generates content, output validation is performed. Validation includes at least: whether it references core conclusions that are not currently included in the set; whether it exceeds the output depth limit stipulated by the current knowledge release rules; whether it violates the context allowed by the chapter skeleton; whether it uses prohibited alternative expressions under author style constraints; and whether it hits prohibited answer patterns in the misconception rules. If the validation fails, at least one corrective method is implemented: reducing the explanation level, compressing it into hint content, rewriting it into supplementary content, rewriting it into a lower-level explanation, or delaying release.

[0153] Furthermore, after receiving explanations, Q&A, examples, prompts, or supplementary content, the system continues to collect the user's confirmation, denial, follow-up questions, pauses, revisits, and answers, and updates the cognitive state vector of the relevant knowledge atoms. Then, it re-divides the mastered set, the accessible set, and the temporarily inaccessible set, forming a closed-loop operation process of knowledge boundary detection, boundary generation, boundary crossing handling, and state updating.

[0154] The edge-side adaptive knowledge transfer method for knowledge-based books disclosed in this application does not focus on recommending the next learning node, but rather on content access control within the knowledge boundaries of the book. Even if the user actively raises a high-level question, the system still determines whether the question is accessible based on the user's cognitive state and the dependencies between knowledge atoms. If the question is inaccessible, the system prohibits the direct release of the target answer and proceeds to a process of pre-emptive supplementation, low-level explanation, clarification and follow-up questions, or delayed release, thereby achieving controlled knowledge transfer for the target book.

[0155] Secondly, this application discloses an edge-side adaptive knowledge transfer system adapted to knowledge-based books, used to execute the edge-side adaptive knowledge transfer method for knowledge-based books disclosed in the first aspect of this application. The system includes: The decomposition unit is used to decompose the target knowledge book into a structured content structure to obtain a structured knowledge representation; The user cognitive state vector generation unit is used to generate user cognitive state vectors based on user interaction information related to target knowledge books. The knowledge boundary detection unit is used to perform knowledge boundary detection on all knowledge atoms based on the user's cognitive state vector and structured knowledge representation, and divide all knowledge atoms into a set that has been mastered, a set that can be entered, and a set that cannot be entered for the time being. The target knowledge atom determination unit is used to respond to interactive response commands, identify the current content intent, and determine the target knowledge atom based on the current content intent; The knowledge access determination unit is used to perform knowledge access determination on each target knowledge atom based on the already mastered set, the accessible set, and the temporarily inaccessible set, to determine the corresponding set type and the controlled response strategy corresponding to the set type. The generation unit is used to generate target content based on the current content intent and controlled response strategy.

[0156] The basic principles of this disclosure have been described above with reference to specific embodiments. However, it should be noted that the advantages, benefits, and effects mentioned in this disclosure are merely examples and not limitations, and should not be considered as essential features of each embodiment of this disclosure. Furthermore, the specific details disclosed above are for illustrative and facilitative purposes only, and are not limitations. These details do not limit the scope of this disclosure to the necessity of employing the aforementioned specific details for implementation.

[0157] In this disclosure, relational terms such as "first" and "second" are used merely to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. The block diagrams of devices, apparatuses, devices, and systems involved in this disclosure are merely illustrative examples and are not intended to require or imply that they must be connected, arranged, or configured in the manner shown in the block diagrams. As those skilled in the art will recognize, these devices, apparatuses, devices, and systems can be connected, arranged, and configured in any manner. Words such as "comprising," "including," "having," etc., are open-ended terms meaning "including but not limited to," and are used interchangeably with them. The terms "or" and "and" as used herein refer to the terms "and / or," and are used interchangeably with them unless the context clearly indicates otherwise. The term "such as" as used herein refers to the phrase "such as but not limited to," and is used interchangeably with it.

[0158] It should also be noted that in the systems and methods of this disclosure, the components or steps can be decomposed and / or recombined. These decompositions and / or recombinations should be considered as equivalent solutions to this disclosure.

[0159] Various changes, substitutions, and modifications can be made to the technology described herein without departing from the teachings defined by the appended claims. Furthermore, the scope of the claims of this disclosure is not limited to the specific aspects of the processes, machines, manufactures, events, means, methods, and actions described above. Currently existing or later-developed processes, machines, manufactures, events, means, methods, or actions that perform substantially the same function or achieve substantially the same result as the corresponding aspects described herein can be utilized. Therefore, the appended claims include such processes, machines, manufactures, events, means, methods, or actions within their scope.

[0160] The above description of the disclosed aspects is provided to enable any person skilled in the art to make or use this disclosure. Various modifications to these aspects will be readily apparent to those skilled in the art, and the general principles defined herein may be applied to other aspects without departing from the scope of this disclosure. Therefore, this disclosure is not intended to be limited to the aspects shown herein, but rather to be carried out within the widest scope consistent with the principles and novel features disclosed herein.

[0161] The above description has been given for purposes of illustration and description. Furthermore, this description is not intended to limit the embodiments of this disclosure to the forms disclosed herein. Although numerous exemplary aspects and embodiments have been discussed above, those skilled in the art will recognize certain variations, modifications, alterations, additions, and sub-combinations therein.

Claims

1. An edge-side adaptive knowledge transfer method adapted to knowledge-based books, characterized in that, include: The target knowledge books are decomposed into a structured content structure to obtain a structured knowledge representation; Generate a user cognitive state vector based on user interaction information related to target knowledge books; Based on the user cognitive state vector and the structured knowledge representation, knowledge boundary detection is performed on all the knowledge atoms, and all the knowledge atoms are divided into a mastered set, an accessible set, and a temporarily inaccessible set. In response to interactive response commands, identify the intent of the current content, and determine the target knowledge atom based on the intent of the current content; Based on the known set, the accessible set, and the temporarily inaccessible set, a knowledge access determination is performed on each of the target knowledge atoms to determine the corresponding set type and the controlled response strategy corresponding to the set type. Based on the current content intent and the controlled response strategy, generate target content.

2. The edge-side adaptive knowledge transfer method for adapting to knowledge-based books according to claim 1, characterized in that, The process of structurally decomposing the target knowledge books to obtain structured knowledge representations includes: The target knowledge book is segmented based on paragraphs and sentence groups to determine several candidate segments; For each candidate fragment, a trigger mode analysis is performed to determine the trigger type, and the atomic type corresponding to each candidate fragment is determined based on the trigger type; Based on the atomic type, extract the corresponding knowledge atom and determine the knowledge release rule for each knowledge atom; Based on the preceding and following relationships between different knowledge atoms, determine the dependency type and dependency strength of the dependency edges between related different knowledge atoms, and generate a preceding dependency relationship graph and corresponding misconception rules; Extract the table of contents of the target knowledge book and generate chapter skeletons based on the prior dependency graph; The structured knowledge representation includes at least the knowledge atoms, the prerequisite dependency graph, the chapter skeleton, the misconception rules, the knowledge release rules, and author expression constraint information; The target knowledge books are knowledge-based books with a clear chapter structure, consistent authorial stance, clear prerequisite dependency diagrams, and clear knowledge boundaries.

3. The edge-side adaptive knowledge transfer method for knowledge-based books according to claim 1, characterized in that, The generation of the user cognitive state vector based on user interaction information related to the target knowledge book includes: Collect user interaction information related to target knowledge books; The user interaction information is mapped to the corresponding knowledge atom, and an atomic-level state feature is generated for each knowledge atom. The user's depth of understanding and interaction stability are determined based on the user interaction information. Based on the atomic-level state characteristics, the depth of understanding, and the interaction stability, the mastery level corresponding to each knowledge atom is determined. Based on the misconception rule base and user interaction performance, the probability of misconception corresponding to each knowledge atom is determined; Based on the mastery level, the probability of misconception, and the interaction stability, a user cognitive state vector is generated corresponding to each knowledge atom.

4. The edge-side adaptive knowledge transfer method for knowledge-based books according to claim 3, characterized in that, Determining the user's depth of understanding based on the user interaction information includes: Determine the preset depth level corresponding to the depth of understanding based on the user interaction information; The score corresponding to the preset depth level is used as the score corresponding to the user's understanding depth.

5. The edge-side adaptive knowledge transfer method for knowledge-based books according to claim 3, characterized in that, Determining the user's interaction stability based on the user interaction information includes: Select the most recent preset number of valid interaction results for knowledge atoms; Based on the interaction samples corresponding to the effective interaction results, determine the consistency coefficient of the correct answer rate, the consistency coefficient of the explanation confirmation, and the consistency coefficient of the self-assessment familiarity. The stability of user interactions is determined based on all consistency coefficients.

6. The edge-side adaptive knowledge transfer method for knowledge-based books according to claim 3, characterized in that, The step of performing knowledge boundary detection on all knowledge atoms based on the user's cognitive state vector and the structured knowledge representation, and dividing all knowledge atoms into a mastered set, an accessible set, and a temporarily inaccessible set, includes: Obtain the knowledge atoms in the user cognitive state vector that satisfy the first preset condition, denot them as mastered classes and classify them into the mastered set; Based on the mastery degree of knowledge atoms of non-mastered classes and the dependency strength of corresponding dependency edges, determine the pre-satisfaction degree of knowledge atoms of non-mastered classes; Determine the release verification status of knowledge atoms that are not of the previously mastered class; The knowledge atoms that obtain the prerequisite satisfaction degree and the release verification state satisfying the second preset condition are denoted as enterable classes and assigned to enterable sets; The remaining knowledge atoms are assigned to a set that cannot be entered for the time being.

7. The edge-side adaptive knowledge transfer method for knowledge-based books according to claim 1, characterized in that, The step of performing knowledge access determination on each target knowledge atom based on the known set, the accessible set, and the temporarily inaccessible set to determine the corresponding set type and the controlled response strategy corresponding to the set type includes: If the target knowledge atom belongs to the already mastered set, the controlled response strategy corresponding to the set type is determined to be the first strategy. If the target knowledge atom belongs to the accessible set, the controlled response strategy corresponding to the set type is determined to be the second strategy. If the target knowledge atom belongs to the set that cannot be entered temporarily, the controlled response strategy corresponding to the set type is determined to be the third strategy.

8. The edge-side adaptive knowledge transfer method for adapting knowledge-based books according to claim 7, characterized in that, The step of generating target content based on the current content intent and the controlled response strategy includes: Determine the intent type of the current content intent; Based on the intent type and the controlled response strategy, generate target content.

9. The edge-side adaptive knowledge transfer method for adapting to knowledge-based books according to claim 8, characterized in that, Also includes: The target content is verified according to preset verification conditions; If the verification fails, at least one correction method is executed, including reducing the interpretation level, compressing into prompt content, rewriting into supplementary content, rewriting into a low-level interpretation, and delaying release.

10. An edge-side adaptive knowledge transfer system adapted to knowledge-based books, characterized in that, include: The decomposition unit is used to decompose the target knowledge book into a structured content structure to obtain a structured knowledge representation; The user cognitive state vector generation unit is used to generate user cognitive state vectors based on user interaction information related to target knowledge books. The knowledge boundary detection unit is used to perform knowledge boundary detection on all the knowledge atoms according to the user cognitive state vector and the structured knowledge representation, and divide all the knowledge atoms into a set that has been mastered, a set that can be entered, and a set that cannot be entered for the time being. The target knowledge atom determination unit is used to respond to an interactive response command, identify the current content intent, and determine the target knowledge atom based on the current content intent; The knowledge access determination unit is used to perform knowledge access determination on each of the target knowledge atoms based on the already mastered set, the accessible set, and the temporarily inaccessible set, to determine the corresponding set type, and to determine the controlled response strategy corresponding to the set type. The generation unit is used to generate target content based on the current content intent and the controlled response strategy.

Images