A teaching resource knowledge point boundary segmentation method and system

Abstract

The application discloses a teaching resource knowledge point boundary segmentation method and system, and relates to the field of teaching resource warehousing.The method comprises the following steps: constructing a paragraph sequence carrying position marks, and extracting teaching scene information; according to the paragraph order and the scene information, marking the knowledge point attribution state of each paragraph, marking and checking the adjustment amount of the teaching function transfer direction after the difference of the adjacent states, and calculating the adjustment amount of the teaching function transfer; introducing state transition sample precision calculation, using the state transition sample precision to constrain the posterior parameter updating process of the Koopman operator, and finally segmenting the paragraph sequence according to the judgment result, matching the original text position marks, completing the knowledge point boundary segmentation, and improving the reliability of the teaching resource.

Description

Technical Field

[0001] This invention relates to the field of teaching resource database processing, and in particular to a method and system for dividing the boundaries of knowledge points in teaching resources. Background Technology

[0002] With the widespread application of online education, smart teaching platforms, and the RAG knowledge base in the input of teaching resources, teaching resources such as textbooks, courseware, questions, and solutions need to be transformed into the smallest units of knowledge that are searchable, reusable, and locatable. These units must not only retain the original text's location markers but also maintain content continuity within the sequence of teaching paragraphs. Accurate segmentation of knowledge point boundaries is a crucial prerequisite for achieving this transformation.

[0003] While existing solutions can organize teaching content using layout structure, table of contents hierarchy, keywords, or statistical features, in the segmentation of knowledge points, these solutions tend to focus on static semantics or external structural clues and fail to incorporate changes in teaching functions (such as from definition to example) as a basis for state transition into model updates. This leads to frequent misjudgments and affects the accuracy of matching knowledge point segmentation with the actual teaching boundaries.

[0004] Therefore, how to effectively divide the boundaries of knowledge points in teaching resources and improve teaching quality has become a technical problem that urgently needs to be solved by those in the field. Summary of the Invention

[0005] This invention provides a method and system for dividing the boundaries of knowledge points in teaching resources, in order to improve the accuracy of dividing the boundaries of knowledge points in teaching resources by introducing the transfer information of teaching resources into the Bayesian posterior update stage of the Koopman operator in MetaKoopman.

[0006] To address the aforementioned technical problems, embodiments of the present invention provide a method for segmenting knowledge point boundaries in teaching resources, including: Obtain the original text of the teaching resources to be added to the database, and construct a sequence of paragraph units carrying the original text location markers; Teaching scenario information is extracted from the paragraph unit sequence to mark the relationship between adjacent paragraphs in the paragraph unit sequence, thereby obtaining the knowledge point affiliation status of each paragraph; A state transition description is generated based on the difference in the attribution status of adjacent knowledge points, and the teaching scenario information is integrated to obtain the teaching function transition adjustment amount; The teaching function transfer adjustment is introduced into the Koopman operator Bayes posterior update loop of MetaKoopman, and the state transition sample accuracy is output to update the Koopman operator posterior parameters. Based on the updated posterior parameters of the Koopman operator, the state evolution of the knowledge point attribution status is performed to obtain the knowledge point attribution status evolution result. The adjacent paragraphs are determined based on the evolution of the knowledge point's attribution status, and the paragraph unit sequence is segmented according to the determination result.

[0007] Furthermore, the step of obtaining the original text of the teaching resources to be added to the database, in order to construct a sequence of paragraph units carrying the original text's location markers, includes: Boundary identification is performed on various paragraphs in the original text of the teaching resources to be added to the database to form initial teaching paragraph units; Extract the original text location markers from the original text of the teaching resources to be added to the database, and align the original text location markers with the initial teaching paragraph units to form candidate teaching paragraph units; The candidate teaching paragraph units are arranged in order to obtain the paragraph unit sequence.

[0008] Furthermore, the step of extracting teaching scenario information from the paragraph unit sequence to annotate the relationships between adjacent paragraphs in the paragraph unit sequence and obtain the knowledge point attribution status of each paragraph includes: Based on the connection relationships between adjacent paragraphs in the paragraph unit sequence, a paragraph order relationship is constructed; Extract the heading level, question type hints, terminology context, and layout separation information of each paragraph from the paragraph unit sequence; Using the heading hierarchy, the paragraph order is labeled with chapter affixation to form a heading hierarchy affiliation relationship; Using the question type hints, the hierarchical relationship of the titles is marked with teaching functions to form the question type hint hierarchy; Using the terminology context, the attribution relationship of the question type prompts is labeled with conceptual orientation, thus forming a terminology context attribution relationship; Using the layout separation information, paragraph boundaries are marked for the terminology context attribution relationship to form a layout separation attribution relationship; The relationship between paragraph order, heading level, question type, terminology context, and layout separation is integrated to generate the attribute status of the knowledge points.

[0009] Furthermore, the step of generating a state transition description based on the differences in the attribution states of adjacent knowledge points, and integrating the teaching scenario information to obtain a teaching function transition adjustment amount, includes: Based on the attribution status of the knowledge points, the direction of state change between adjacent paragraphs is calculated to obtain the state transition description. Based on the title level and the question type prompts, the teaching function transfer direction is marked on the state transition description to form a teaching function transfer description; Based on the terminology context and the layout separation information, the description of teaching function transfer is checked for conceptual orientation and paragraph separation to form the meaning of teaching function transfer. Based on the meaning of the teaching function transfer, the adjustment amount of the teaching function transfer is calculated.

[0010] Furthermore, updating the posterior parameters of the Koopman operator includes: Based on the state transition sample accuracy, the degree of participation of the state transition sample formed by the adjacent knowledge point attribution states is calculated; Based on the participation level of the state transition samples, the state transition samples in the Koopman operator Bayes posterior update stage are allocated to obtain the posterior update input. The posterior parameters of the Koopman operator are updated based on the posterior update input.

[0011] Furthermore, the step of determining adjacent paragraphs based on the evolution result of the knowledge point attribution status, and segmenting the paragraph unit sequence according to the determination result, includes: Based on the evolution results of the knowledge point attribution status, calculate the state evolution deviation between adjacent paragraphs; The state evolution deviation is compared with a preset condition threshold, and a list of marked paragraphs is generated based on the comparison result between adjacent paragraphs. Extract knowledge point boundaries from the list of marked paragraphs, and use the knowledge point boundaries to segment the paragraph unit sequence.

[0012] Furthermore, the step of comparing the state evolution deviation with a preset condition threshold and marking adjacent paragraphs based on the comparison result includes: When the state evolution deviation exceeds a preset threshold, adjacent paragraphs are marked as the same knowledge point unit; When the state evolution deviation does not exceed the preset condition threshold, the adjacent paragraphs are marked as knowledge point boundaries.

[0013] Another embodiment of the present invention provides a knowledge point boundary segmentation system for teaching resources, comprising: The sequence construction module is used to obtain the original text of the teaching resources to be added to the database, so as to construct a sequence of paragraph units carrying the original text position markers; The paragraph association annotation module is used to extract teaching scenario information from the paragraph unit sequence to annotate the association relationship between adjacent paragraphs in the paragraph unit sequence, and obtain the knowledge point attribution status of each paragraph; The teaching function transfer analysis module is used to generate a state transfer description based on the differences in the ownership status of adjacent knowledge points, and to obtain the teaching function transfer adjustment amount by integrating the teaching scenario information. The posterior update module is used to introduce the teaching function transfer adjustment into the MetaKoopman operator Bayes posterior update loop, and output the state transition sample accuracy to update the Koopman operator posterior parameters. The state evolution module is used to perform state evolution on the adjacent knowledge point ownership states based on the updated Koopman operator posterior parameters, and obtain the knowledge point ownership state evolution results. The sequence segmentation module is used to determine adjacent paragraphs based on the evolution result of the knowledge point attribution status, and to segment the paragraph unit sequence according to the determination result.

[0014] Another embodiment of the present invention provides a computer device, including a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor, wherein the processor executes the computer program to implement the teaching resource knowledge point boundary segmentation method as described above.

[0015] Another embodiment of the present invention provides a computer-readable storage medium storing a computer program, wherein when the device containing the computer-readable storage medium executes the computer program, it implements the teaching resource knowledge point boundary segmentation method as described above.

[0016] Compared with the prior art, the beneficial effects of the embodiments of the present invention are at least one of the following: This invention first obtains the original text of the teaching resources to be added to the database and constructs a sequence of paragraph units carrying the original text location markers, providing a traceable original text mapping basis for subsequent boundary localization. Then, it extracts teaching scenario information from the sequence to label the relationships between adjacent paragraphs, enhancing the semantic richness of the state objects and avoiding misjudgment based on a single feature. Based on state differences, it fuses scenario information to calculate the teaching function transfer adjustment amount and introduces Koopman posterior update, enabling the sample precision to carry the meaning of teaching function transfer, thereby constraining the direction of posterior parameter update and enhancing the model's ability to perceive teaching logic. Based on the updated parameters, it performs state evolution, ensuring that boundary determination originates from the continuous state evolution process rather than a text similarity threshold, improving the interpretability of boundary determination. Finally, it determines the continuity of adjacent paragraphs based on the evolution results, segmenting the paragraph sequence so that the output fully meets the actual needs of continuous organization of knowledge point units and original text localization in the teaching resource database, improving the reliability of the teaching resources. Attached Figure Description

[0017] Figure 1This is a schematic diagram of the teaching resource knowledge point boundary segmentation method in one embodiment of the present invention; Figure 2 This is a schematic diagram illustrating the process of aligning paragraph unit sequences with original text position markers in one embodiment of the present invention; Figure 3 This is a schematic diagram of the MetaKoopman knowledge point attribution state evolution process in one embodiment of the present invention; Figure 4 This is a flowchart illustrating the knowledge point attribution status fusion and boundary output process in one embodiment of the present invention; Figure 5 This is a schematic diagram of the teaching resource knowledge point boundary segmentation system in one embodiment of the present invention. Detailed Implementation

[0018] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. The purpose of providing these embodiments is to make the disclosure of the present invention more thorough and comprehensive. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0019] In the description of this application, the terms "first," "second," "third," etc., are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Therefore, a feature defined with "first," "second," "third," etc., may explicitly or implicitly include one or more of that feature. In the description of this application, unless otherwise stated, "a plurality of" means two or more.

[0020] In the description of this application, it should be noted that, unless otherwise expressly specified and limited, the terms "installation," "connection," and "linking" should be interpreted broadly. For example, they can refer to fixed connections, detachable connections, or integral connections; they can refer to mechanical connections or electrical connections; they can refer to direct connections or indirect connections through an intermediate medium; and they can refer to the internal communication between two components. The terms "vertical," "horizontal," "left," "right," "upper," "lower," and similar expressions used herein are for illustrative purposes only and do not indicate or imply that the device or component referred to must have a specific orientation, or be constructed and operated in a specific orientation, and therefore should not be construed as limiting the invention. The term "and / or" as used herein includes any and all combinations of one or more of the associated listed items. Those skilled in the art can understand the specific meaning of the above terms in this application based on the specific circumstances.

[0021] In the description of the present application, it should be noted that, unless otherwise defined, all technical and scientific terms used in the present invention have the same meanings as commonly understood by those skilled in the art. The terms used in the specification of the present invention are only for the purpose of describing specific embodiments, and are not intended to limit the present invention. For those of ordinary skill in the art, the specific meanings of the above terms in the present application can be understood according to specific circumstances.

[0022] An embodiment of the present invention provides a method for splitting knowledge point boundaries of teaching resources. For details, please refer to Figure 1 , Figure 1 is a schematic flow chart of the method for splitting knowledge point boundaries of teaching resources in one embodiment of the present invention, which includes the following steps: S1, acquiring the original text of teaching resources to be stored, so as to construct a paragraph unit sequence carrying original text position marks.

[0023] In this embodiment, the acquired original text of the teaching resources to be stored is denoted as , wherein represents characters, line breaks, page breaks, blank line marks, indentation marks, serial number marks or separator marks arranged according to the display order of the original text of the teaching resources to be stored, represents a global character index, represents the total number of characters and layout marks contained in the original text of the teaching resources to be stored.

[0024] When acquiring the original text of the teaching resources to be stored, the sequence of page numbers, chapter positions, original paragraph numbers, character indexes, line breaks, blank line marks, page breaks, indentation marks, serial number marks and separator marks in is retained. The page numbers are from the page numbers of the original text of the teaching resources to be stored. When the original text of the teaching resources to be stored does not provide page numbers, the full-text page number is set as 1. The chapter positions are from the existing chapter, section, and subsection title records in the original text of the teaching resources to be stored. When the original text of the teaching resources to be stored does not provide structured chapter records, text lines that start with "第" and end with "章" or "节", text lines that start with Arabic numeral hierarchical numbers, and text lines that are on a separate line with blank lines before and after are identified as chapter position records.

[0025] Next, paragraph boundary recognition is performed on the original text of the teaching resources to be stored. For the specific process, please refer to Figure 2 shown. Specifically, during recognition, according to the global character index, scanning is performed from to , the identified initial teaching paragraph unit is denoted as , and each unit is recorded in the character start index and character end index , wherein , represents the number of initial teaching paragraph units.

[0026] Paragraph boundary recognition includes natural paragraph boundary recognition and boundary recognition of four types: question paragraph, analysis paragraph, example paragraph and exercise paragraph. Wherein, the boundary of a natural paragraph is jointly determined by paragraph tags, line breaks, blank line markers and paragraph termination punctuation of the source file. When there is a blank line marker between consecutive text lines, the last non-empty character before the blank line marker is taken as the termination position of the current natural paragraph, and the first non-empty character after the blank line marker is taken as the starting position of the next natural paragraph; the boundary of a question paragraph is determined by question number markers and question type names, the question number markers include Arabic numerals plus periods, Arabic numerals in parentheses, Chinese numerals plus顿号, and "第" plus numerals plus "题", the question type names include "multiple-choice question", "fill-in-the-blank question", "essay question" and "true-false question"; the boundary of an analysis paragraph is determined by the paragraph-initial words "analysis", "answer", "solution" and "comment"; the boundary of an example paragraph is determined by the paragraph-initial words "example" and "illustrative example"; the boundary of an exercise paragraph is determined by the paragraph-initial words "exercise", "consolidation training" and "synchronous training".

[0027] Exemplarily, during scanning, when the paragraph-initial content of the current text line meets the identification condition of any of the above teaching functional paragraph boundaries, the first character of the current text line is taken as the starting position of a new initial teaching paragraph unit; if the current initial teaching paragraph unit already has text content, the last non-empty character before the first character of the current text line is taken as the termination position of the previous initial teaching paragraph unit.

[0028] For a matched question paragraph, analysis paragraph, example paragraph or exercise paragraph, if the next text line does not meet the identification condition for a new teaching functional paragraph boundary and does not form a new natural paragraph boundary, the next text line is merged into the current initial teaching paragraph unit.

[0029] When a page break is located inside an initial teaching paragraph unit, it is only retained as a layout position record, and does not split the initial teaching paragraph unit separately; if there is a natural paragraph termination position before the page break, and the first non-empty text line after the page break meets the paragraph-initial identification condition, the non-empty text line is taken as the starting position of a new initial teaching paragraph unit; fragments that only contain blank characters, page breaks, indentation markers or separator line markers do not form initial teaching paragraph units.

[0030] Then, according to the index and , extract the original text position marker corresponding to the initial teaching paragraph unit from the original text of the teaching resource to be stored , wherein represents the starting page number of the initial teaching paragraph unit, represents the ending page number of the initial teaching paragraph unit, Indicates the initial offset within the page of the initial teaching paragraph unit. This indicates the end offset within the initial teaching paragraph unit page. Indicates the chapter number. Indicates the starting character index of the chapter. Indicates the index of the chapter terminator character. Indicates the sequence number within the chapter. Indicates the starting number of the original paragraph. This indicates the end number of the original paragraph. Indicates the starting offset within the paragraph. Indicates the end offset within the paragraph. Indicates the global character starting index. Indicates the global character terminating index. Indicates the starting index of the characters within the page. This indicates the terminating index of a character within the page.

[0031] The encoding rules are as follows: page numbers, chapter numbers, original paragraph numbers, and chapter sequence numbers use integer encoding starting from 1, while character indices and offsets use integer encoding starting from 0; if the chapter position of the original text of the teaching resource to be added to the database cannot be determined, the chapter number, chapter start character index, chapter end character index, and chapter sequence number are written as 0.

[0032] At this point, the original text location markers are aligned with the initial teaching paragraph units, thus forming the required candidate teaching paragraph units. This unit contains original text location markers. In this embodiment, during alignment, it is made... and Consistency, making and Consistent when candidate teaching paragraph units span multiple page numbers Write the starting page number. When writing the termination page number, if the candidate teaching paragraph unit spans multiple original paragraphs, Enter the original paragraph starting number. Enter the original paragraph end number, and after alignment, follow... Arrange all of them in ascending order. This forms the final sequence of paragraph units: .

[0033] Understandably, in MetaKoopman's fixed-sample-dimension configuration, the paragraph unit sequence... The original text segments will be entered in sequential order, organized by a continuous index into 128 teaching paragraph units at a time. ,in, This indicates the starting index of the input segment in the sequence of teaching paragraph units, each Includes candidate teaching paragraph units and 16-dimensional original text position markers 16-dimensional original text location markers The original text position marker field, written into the 160-dimensional input features of each teaching paragraph unit, is used to indicate the positional meaning of page number, chapter position, paragraph start and end position, and character start and end position in the original text of the teaching resource to be entered into the database.

[0034] S2. Extract teaching scenario information from the paragraph unit sequence to mark the relationship between adjacent paragraphs in the paragraph unit sequence, and obtain the knowledge point belonging status of each paragraph.

[0035] To better serve the subsequent dynamic evolution of the MetaKoopman framework, this step further obtains the knowledge point attribution status of each segment in the sequence.

[0036] Specifically, this involves extracting the heading level, question type hints, terminology context, and layout separation information for each paragraph from the paragraph unit sequence. For example: for each... Read the first non-empty text line and denote the first 32 characters of the first non-empty text line as the paragraph beginning content. If the number of characters is less than 32, retain the actual number of characters and read them in ascending order of the paragraph unit index. and combined The sorting order extracts heading levels, question type hints, terminology context, and layout separation information.

[0037] For heading levels, this embodiment preferably uses a four-level heading structure for identification: Level 1 headings are determined by paragraph beginnings containing "Chapter" and "Part," or by chapter titles on a separate line with no more than 20 characters; Level 2 headings are determined by sections containing "Section" and "Section," or by numbers in the form of AB; Level 3 headings are determined by numbers in the form of ABC; Level 4 headings are determined by formats such as "(I)" and "(II)" or short headings followed by a colon. A heading stack is then established. Record the sequential numbers of the first to fourth level headings closest to the current unit, i.e. to When a title row is scanned, the corresponding level number is updated and the lower level is set to 0. Non-title rows continue to use the title stack of the previous unit. Finally, a 16-dimensional title level vector hi is formed, which includes the title hit marker of each level, the title stack number, whether it is a title row, the title level change marker, the sequence number within the level, the distance from the nearest title row, whether it contains a numeric number, and the level missing marker.

[0038] For question type prompts, in this embodiment, they are extracted based on teaching prompt words. Question prompt words include "question," "multiple choice," "fill in the blank," "problem-solving," and "true / false"; example prompt words include "example" and "example question"; analysis prompt words include "analysis," "answer," "solution," and "commentary"; practice prompt words include "practice," "consolidation training," and "synchronous training"; definition prompt words include "definition," "concept," "called," and "refers to"; and explanation prompt words include "explanation," "note," and "method." Finally, a 16-dimensional question type prompt vector pi is formed, which includes various function hit markers, prompt word position information, option / sub-question number markers, comparison markers with the previous unit's functions, and the main teaching function code (determined according to the priority order of question, example, analysis, practice, definition, and explanation; set to 0 if no match is found).

[0039] Regarding the terminology context, in this embodiment, according to The document extracts the semantic content of knowledge point terms and adjacent teaching paragraphs / units, while the knowledge point terminology table T is provided by the subject knowledge directory, containing standard terms and synonyms. The longest-first matching is performed. When multiple terms are matched at the same character position, the one with the longest length is retained. If the lengths are the same, the one ranked earlier in the table is retained, thus forming a standard term identifier list. Select up to 64 terms from the entire list, based on their first occurrence, to form a local term slot. Empty spaces are left blank, generating a 64-dimensional term occurrence marker bi. Further, a 16-dimensional semantic content vector gi is constructed, containing definition hints, property hints, formula symbols, condition hints, conclusion hints, method hints, example / exercise / answer analysis references, causal / transition / enumeration structures, theorem hints, diagram hints, summary hints, and continuation hints. Each dimension is determined by the corresponding keyword hit. Finally, a term context is formed. It has a total of 96 dimensions. When the beginning or end of the sequence is missing, it is filled with a zero vector. Local term slots maintain the same slot order among all units.

[0040] For layout separation information, in this embodiment, layout separation information is extracted based on the paragraph spacing between adjacent paragraphs in the sequence and layout separation marks. Layout separation marks include blank lines, page breaks, indentations, serial numbers, and separator lines (three or more consecutive horizontal lines, underscores, or long horizontal lines). Finally, a layout separation vector is formed. First 8 dimensions record and Separation information between elements (number of blank lines, page break markers, indentation variation, sequence number markers, separator line markers, number of line breaks, original paragraph number intervals, same-page markers), and the last 8 dimensions of the record. and The corresponding information between them, and the missing positions at the beginning or end of the sequence are set to 0.

[0041] Based on this, the heading hierarchy Question type prompts Terminology context and layout separation information Indexed in the same order Align with paragraph unit sequence to form teaching scenario information In a fixed example dimension configuration, the 16-dimensional original text position markers ri and mi are concatenated to form a 160-dimensional input feature. Input the data into MetaKoopman in batches of 128 units, following a sequential order.

[0042] After obtaining the scene information, the corresponding attribution relationships are generated according to the paragraph order in the sequence, and the attribution status of knowledge points is generated based on this.

[0043] Specifically, based on the connection relationships between adjacent paragraphs in the paragraph unit sequence, the paragraph order relationship is constructed. For example: for each ,when At that time, When it is determined to be the previous teaching segment unit, When, the previous teaching segment unit is written into the empty space marker, when At that time, When it is determined to be the next teaching segment unit, When the time comes, write the next teaching paragraph unit into the empty space marker. Write the previous teaching paragraph unit index, the current teaching paragraph unit index, the next teaching paragraph unit index, the start and end positions of characters in the original text position marker, and the original text paragraph number interval into the paragraph order relationship. It is used to limit the source order of the attribution status of adjacent knowledge points.

[0044] Then, the previously generated scene information is used for annotation. Specifically, using the heading hierarchy, the paragraph order is annotated with chapter affixation. The heading hit marker, the first to fourth level heading sequence numbers, the heading stack change marker, and the sequence number within the hierarchy are read from hi. The current teaching paragraph unit is mapped to the chapter level defined by the nearest heading level. The heading line is written to the starting point of the corresponding chapter level, and the non-heading line is written to the chapter level defined by the nearest heading stack, forming the heading hierarchy affiliation relationship αi.

[0045] Using question type prompts, the hierarchical relationship of headings is labeled with teaching functions, and the information is retrieved. The system uses various function hit markers and main teaching function codes to write the corresponding paragraph function (question / example / analysis / exercise / definition / explanation / general explanation) into the current teaching paragraph unit, while retaining the differences and similarities between the teaching functions and the previous unit. This allows changes in teaching functions to be reflected in the status object, forming a question type hint attribution relationship. .

[0046] By utilizing terminology context, conceptual attribution is assigned to question type hints. The 64-dimensional term occurrence markers bi and adjacent semantic content vectors are read from wi. Within the same local term slot, if the term appears in both the current and previous units, a forward conceptual continuation is marked; if the term appears in both the current and next units, a backward conceptual continuation is marked; only if the term appears in the current unit, the current conceptual attribution is marked. Definition hints, method hints, example references, answer analysis hints, and continuation hints from adjacent semantic content vectors are retained, thus forming terminology context attribution relationships. .

[0047] Using layout separation information, paragraph boundaries are marked for the contextual attribution of terms. Eight-dimensional fields (number of blank lines, page break mark, indentation change, sequence number mark, separator mark, number of line breaks, original paragraph number interval, and same-page mark) are read from the `di` module and written between the current and adjacent units. When the number of blank lines is greater than 0, page break mark is 1, separator mark is 1, and original paragraph number interval is greater than 1, a paragraph separation enhancement mark is written; when the same-page mark is 1 and the paragraph number interval is 1, a paragraph continuation mark is written, thus forming a layout separation attribution relationship. .

[0048] These attribution relationships are integrated to generate the attribution status of knowledge points. In this process, to ensure that the attribution relationships of question type prompts and terminology context participate in state transitions under the constraints of heading-level attribution relationships and layout separation attribution relationships, in one implementation of this embodiment, adjacent paragraphs are... and The state transition constraint value ηi is calculated as follows: In the formula, Indicates adjacent teaching paragraphs / units and State transition limits between them Indicates the sequential index within a paragraph unit sequence. This represents the dimension indices from the 5th to the 8th dimension in the heading level, where the 5th to 8th dimensions are sequentially numbered from the first-level heading to the fourth-level heading, respectively. This indicates that the values ​​within a specified index range are summed. This is a flag function that assigns a value of 1 if the condition is true and a value of 0 if the condition is false. This indicates that both conditions are met simultaneously. Indicates the first The heading level of each teaching segment unit Dimensional value, Indicates the first The heading level of each teaching segment unit Dimensional value, This indicates the local term slot index in the term occurrence marker. Indicates the first The teaching paragraph unit in the first The term in each local term slot is marked with a value of 0 or 1. Indicates the first The teaching paragraph unit in the first The term in each local term slot is marked with a value of 0 or 1. This indicates that the smaller value among the input values ​​is selected. Indicates the first Each teaching segment unit is marked with a different teaching function relative to the previous teaching segment unit, with a value of 0 or 1. express and The number of blank lines between them express and Page breaks between express and The amount of indentation change between them This represents the absolute value of the change in indentation. express and The sequence number markings between them express and The separator line marking between them express and The number of line breaks between them express and The interval between the original paragraph numbers, express and The same page mark between them.

[0049] The state transition constraint ηi is written in the paragraph order relationship. The adjacent state transition field in the Bayesian posterior update is used in subsequent Bayesian posterior updates.

[0050] Finally, the paragraph order relationship Write the heading hierarchy relationship between paragraphs. Question type hints: attribution relationship Terminology context attribution Layout separation and ownership relationship Write the corresponding paragraphs, and concatenate the original text location markers, heading levels, question type hints, terminology context, and layout separation information in the same order to form a 160-dimensional input feature. In this embodiment, the question type hint attribution relationship and the term context attribution relationship do not enter the state transition independently. Instead, they participate in the adjacent state transition organization under the joint constraints of the title level attribution relationship, the layout separation attribution relationship, and the state transition constraint value ηi, keeping the 160-dimensional input feature dimension unchanged.

[0051] In this process, the state input mapping in the MetaKoopman main skeleton is used. Generating knowledge point attribution status: 160-dimensional input features are processed through a 256-dimensional fully connected layer (including ReLU) to generate a 256-dimensional intermediate representation, which is then processed through a 128-dimensional fully connected layer (including ReLU) to generate a 128-dimensional knowledge point attribution status zi. This is done each time 128 teaching segment units are processed, following the sequential order of the segment unit sequence. , This represents the starting order index of the 128 input paragraph segments. `zi` is defined by the paragraph order and various hierarchical relationships, serving as a 128-dimensional state object for MetaKoopman's adjacent state transition processing.

[0052] S3. Generate a state transition description based on the differences in the ownership status of adjacent knowledge points, and integrate teaching scenario information to obtain the teaching function transition adjustment amount.

[0053] Specifically, firstly, based on the attribution status of knowledge points, the direction of status change between adjacent paragraphs is calculated. When the difference between z(i+1) and zi is greater than the threshold (Preferred value) When marked as the reinforcement direction, less than The direction is weakened when the indicator changes, and maintained otherwise. This is combined with markers indicating changes in title level, question type, contextual terms, and layout separation to generate a state transition description. .

[0054] Next, based on the heading level and question type prompts, the teaching function transfer direction is marked in the state transition description: The main teaching function codes of adjacent paragraphs are read (1-6 for question, example, analysis, exercise, definition, and explanation, 0 for no match) and the heading level sequence number are used to determine which paragraph the teaching function is transitioning to and whether it crosses heading levels. The following are written into the state transition description: teaching function retention, definition to example, example to analysis, analysis to exercise, explanation to example, question to analysis, and cross-heading level conversion. Thus forming a description of the transfer of teaching functions. .

[0055] Then, based on the terminology context and layout separation information, the description of teaching function transfer is validated for concept orientation and paragraph separation: the first 64 dimensions of adjacent paragraphs are read to find the markers bi and b(i+1). When both local term slots are hit, a concept continuation marker is written; when only one side is hit, a concept switching marker is written. The 9th to 16th dimensions of di are read (number of blank lines, page break marker, indentation change, sequence number marker, separator line marker, number of line breaks, original paragraph number interval, same page marker). When page breaks / separator lines / blank lines ≥ 2 / paragraph interval > 1, a paragraph separation enhancement marker is written; when the same page has an interval of 1, a paragraph continuity marker is written, thus forming the meaning of teaching function transfer. Understandably, when adjacent paragraphs are at the same chapter level, have concept continuation markers, and are not marked with paragraph separation enhancement markers, and the direction of the teaching function transfer is from definition to example, example to analysis, analysis to practice, explanation to example, or question to analysis, they are marked as internal expansion within the same knowledge point; when there are changes in first- or second-level headings and paragraph separation enhancement, they are marked as cross-knowledge point switching; when there are the same terms but the heading level changes, they are marked as heading change constraints; when there are changes in teaching function but the heading level remains the same, they are marked as teaching function change constraints. The verified description and meaning are then written together into the adjustment quantity generation record.

[0056] Based on the meaning of teaching function transfer, a set of teaching function transfer codes that allows for the representation of expansion within the same knowledge point is defined. The corresponding functions of instructional function transfer are: definition to example, example to analysis, analysis to practice, explanation to example, and problem to analysis. It is expressed as follows: In the formula, Indicates the first The main teaching function code for each teaching segment unit. Indicates the first The main teaching function code for each teaching segment unit.

[0057] S4. Introduce the teaching function transfer adjustment into the Koopman operator Bayesian posterior update loop of MetaKoopman, and output the state transition sample accuracy to update the Koopman operator posterior parameters.

[0058] This step aims to adjust the amount of teaching function transfer. This paper introduces a state transition sample accuracy calculation layer in the Bayesian posterior update stage using the MetaKoopman operator. Specifically, it will... , and Concatenate into a 257-dimensional precision input vector After state transition sample precision calculation layer The state transition sample accuracy λi is calculated. Understandably, since τi and its neighboring states participate in the computation along the same computational path, λi carries the meaning of the teaching function transition and serves as the input parameter for the Koopman operator Bayesian posterior update stage.

[0059] It should be understood that the posterior parameters of the Koopman operator A posterior mean matrix containing 128x128 Koopman operators And the posterior precision matrix of the 128x128 Koopman operator , Indicates the first The input state dimension to the first The state evolution connection weights of each output state dimension This represents the posterior precision of the connection weights corresponding to the state evolution. If MetaKoopman has already stored the posterior parameters of the Koopman operator, it can be read directly. and Otherwise, Set the main diagonal elements to 1 and the off-diagonal elements to 0, and then... Set all elements to 1.

[0060] Based on the above-initialized posterior parameters of the Koopman operator, the participation degree of state transition samples in the formation of adjacent knowledge point ownership states is first calculated according to the precision of the state transition samples. For example: this can be based on the state transition samples... The dimensional changes of the 128 state dimensions in the middle, for By performing a bounded participation transition, we obtain the participation level of the state transition sample. It is used to determine state transition samples. The participation strength in the Koopman operator Bayesian posterior update phase ( At that time, state transition sample Do not write to the Koopman operator posterior parameter update; At that time, state transition sample according to (Write to the posterior parameter update of the Koopman operator), degree of participation of state transition samples Specifically, it is expressed as follows: In the formula, Indicates the first In the knowledge point attribution status of each teaching segment / unit, the first The values ​​of each state dimension, Indicates the first In the knowledge point attribution status of each teaching segment / unit, the first The values ​​of each state dimension, The threshold value represents the state's participation in stabilization; the preferred value is... , This represents the threshold for cutting off state changes, with a preferred value of 1.

[0061] Then based on this The state transition samples in the Koopman operator Bayesian posterior update process are allocated as follows: As a state transition sample The previous state will As a state transition sample The target state will As a state transition sample The level of participation is marked to form a posteriori update input with implications for the transfer of teaching functions. The teaching function transfer meaning of λi is achieved by ωi participating in the posterior parameter update of the Koopman operator.

[0062] Specifically, the Koopman operator posterior parameter update layer operates within a 128-dimensional state space. The update layer contains 128 x 128 fully connected state evolution connections, each corresponding to... and For each group of adjacent knowledge points, the attribution status of the previous knowledge point is determined by... The status of the next knowledge point and the degree of participation of state transition samples Write the Bayesian posterior update process corresponding to the state evolution connection, for the th Each state evolution connection, first... Increase and The product of the squared values ​​yields the updated posterior precision of the Koopman operator. Then update the previous version and The product of and and Add the products together and divide the sum by . The updated posterior mean of the Koopman operator is obtained. Complete the first After updating the attribution status of adjacent knowledge points in a group, and Composition of the updated Koopman operator posterior parameters and will As the object for reading the posterior parameters of the Koopman operator after the adjacent order index is increased.

[0063] S5. Based on the updated posterior parameters of the Koopman operator, perform state evolution on the ownership state of adjacent knowledge points to obtain the knowledge point ownership state evolution result.

[0064] This step involves using the updated Koopman operator posterior parameters. This describes the process of calculating the state evolution of adjacent knowledge points. For details, please refer to [link / reference needed]. Figure 3 As shown.

[0065] Specifically, the updated Koopman operator posterior mean matrix As a 128x128 fully connected state evolution connection, the first... Attribution status of each knowledge point The input is a 128x128 fully connected state evolution connection, and the 128 output neurons each correspond to one of the state dimensions in the state evolution result. For the i-th... Each output neuron will The 128 state dimensions are respectively with to Multiply them, and sum the 128 products to form the state evolution result at the 1st digit. The numerical values ​​of each state dimension ,right to The same process is performed sequentially to produce a 128-dimensional state evolution result. .

[0066] based on ,calculate and The state evolution deviation is obtained by taking the average of the dimension-wise absolute deviations. , which represents the bias generated when the updated Koopman operator posterior parameters interpret the continuous evolution of the attribution status of adjacent knowledge points.

[0067] Then, based on the state evolution deviation and the updated Koopman operator posterior parameters Generate the evolution results of the knowledge point attribution status. It is used to indicate whether adjacent states can be explained by the continuous evolution relationship of the same knowledge point state.

[0068] S6. Determine adjacent paragraphs based on the evolution of knowledge point ownership status, and segment the paragraph unit sequence according to the determination results.

[0069] This step is the process of sequence segmentation.

[0070] The previously calculated state evolution deviation Compared with preset condition threshold (Preferred value) The system compares the data and marks adjacent paragraphs based on the comparison results, generating a list of marked paragraphs. When the state evolution deviation exceeds a preset threshold, i.e., when... When the condition is met, adjacent paragraphs are marked as the same knowledge point unit; when the state evolution deviation does not exceed the preset threshold, the condition is not met and adjacent paragraphs are marked as knowledge point boundaries.

[0071] Extract knowledge point boundaries from the list of marked paragraphs, and then... In the paragraph unit sequence, extract the left-side teaching paragraph unit index i corresponding to each boundary to form the knowledge point boundary index sequence B. (M is the number of boundaries), sorted in ascending order of value, with only one duplicate index retained.

[0072] Next, the segment unit sequence is segmented using the boundary index sequence B. When At that time, a knowledge point unit scope is formed. ;when and At that time, a knowledge point unit scope is formed. ;when and When writing the scope of the first knowledge point unit, it should be as follows: Indexing the boundary between two adjacent knowledge points and Write the scope of intermediate knowledge points as Write the scope of the last knowledge point unit as .

[0073] In some implementations of this embodiment, the range of the knowledge point units formed after segmentation is further denoted as... ( The sequential numbering indicates the range of knowledge points within a given unit. The unit index indicates the starting teaching paragraph within the scope of a knowledge point unit. This indicates the end of the teaching paragraph unit index within the knowledge point unit range, and matches it with the original text position marker. Specifically, it reads the starting unit. and End Unit The original text location markers will include the starting page number, chapter position, paragraph start position, and character start position, as well as the ending page number, chapter position, paragraph end position, and character end position, in the knowledge point unit position relationship. .

[0074] Furthermore, based on the boundaries of knowledge points Match the original text location markers to determine the adjacent original text locations corresponding to the knowledge point boundaries, thus forming the knowledge point boundary locations. Specifically, for the boundaries of each knowledge point Read the left cell Original text location markers (end page number, chapter position, paragraph end position, character end position) and right-side cells The original text location markers (starting page number, chapter position, paragraph start position, character start position) together constitute the boundary position of the knowledge point. .

[0075] Finally, based on the positional relationship of the knowledge points units and knowledge point unit scope Extract the content of the corresponding paragraph units to form the smallest unit of knowledge. Among them, the scope of knowledge point units. ,according to to Read candidate teaching paragraph unit content in sequence and according to the teaching paragraph unit sequence The original text is arranged in the order of the knowledge points and written into the smallest unit. At the same time, when At that time, Corresponding knowledge point boundary position Write as the smallest unit of knowledge points The left-hand boundary position of the knowledge point, when At that time, Corresponding knowledge point boundary position Write as the smallest unit of knowledge points The right-hand boundary of the knowledge point, the smallest unit of the knowledge point. The knowledge point boundaries are output in the order of the teaching paragraph / unit sequence. For details on the knowledge point attribution status fusion and boundary output process, please refer to [reference needed]. Figure 4 As shown.

[0076] In summary, this embodiment constructs a paragraph sequence carrying location markers; extracts teaching scenario information to label adjacent paragraphs, obtaining the knowledge point attribution status of each paragraph; generates teaching function transfer adjustment quantities by fusing scenario information based on the differences between adjacent states, introduces the MetaKoopman operator Bayesian posterior update step, and outputs sample accuracy to update posterior parameters; evolves adjacent states based on the updated parameters, determines paragraph continuity based on the evolution results, and segments the paragraph sequence accordingly.

[0077] One embodiment of the present invention provides a knowledge point boundary segmentation system for teaching resources. For details, please refer to [link / reference]. Figure 5 , Figure 5The diagram shown is a schematic representation of a teaching resource knowledge point boundary segmentation system according to one embodiment of the present invention, comprising: The sequence construction module M1 is used to obtain the original text of the teaching resources to be added to the database, so as to construct a sequence of paragraph units carrying the original text position markers; The paragraph association annotation module M2 is used to extract teaching scenario information from the paragraph unit sequence to annotate the association relationship between adjacent paragraphs in the paragraph unit sequence, and obtain the knowledge point attribution status of each paragraph; The teaching function transfer analysis module M3 is used to generate a state transfer description based on the difference in the belonging status of adjacent knowledge points, and to obtain the teaching function transfer adjustment amount by integrating the teaching scenario information. The posterior update module M4 is used to introduce the teaching function transfer adjustment into the MetaKoopman operator Bayes posterior update loop, and output the state transition sample accuracy to update the posterior parameters of the Koopman operator. The state evolution module M5 is used to perform state evolution on the adjacent knowledge point ownership states based on the updated Koopman operator posterior parameters, and obtain the knowledge point ownership state evolution results. The sequence segmentation module M6 is used to determine adjacent paragraphs based on the evolution result of the knowledge point attribution status, and to segment the paragraph unit sequence according to the determination result.

[0078] In one embodiment of this application, a computer device is provided, which includes a memory and a processor. The memory stores machine-executable instructions that can be executed by the processor. When the processor executes a computer program, it implements the above steps. The computer device provided in this embodiment has a similar implementation principle and technical effect to the above method embodiments, and will not be described again here.

[0079] In one embodiment of this application, a computer-readable storage medium is provided, on which a computer program is stored, and when the computer program is executed by a processor, it performs the above steps; the implementation principle and technical effects of the computer-readable storage medium provided in this embodiment are similar to those of the above method embodiments, and will not be repeated here.

[0080] The technical features and effects of the teaching resource knowledge point boundary segmentation system proposed in this embodiment are the same as those of the teaching resource knowledge point boundary segmentation method proposed in this embodiment, and will not be repeated here.

[0081] The embodiments described above are merely illustrative of several implementations of the present invention, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of the present invention. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the present invention, and these modifications and improvements all fall within the scope of protection of the present invention. Therefore, the scope of protection of this patent should be determined by the appended claims.

Claims

1. A method for dividing the boundaries of knowledge points in teaching resources, characterized in that, include: Obtain the original text of the teaching resources to be added to the database, and construct a sequence of paragraph units carrying the original text location markers; Teaching scenario information is extracted from the paragraph unit sequence to mark the relationship between adjacent paragraphs in the paragraph unit sequence, thereby obtaining the knowledge point affiliation status of each paragraph; A state transition description is generated based on the difference in the attribution status of adjacent knowledge points, and the teaching scenario information is integrated to obtain the teaching function transition adjustment amount; The teaching function transfer adjustment is introduced into the Koopman operator Bayes posterior update loop of MetaKoopman, and the state transition sample accuracy is output to update the Koopman operator posterior parameters. Based on the updated posterior parameters of the Koopman operator, the state evolution of the knowledge point attribution state is performed to obtain the knowledge point attribution state evolution result. The adjacent paragraphs are determined based on the evolution of the knowledge point's attribution status, and the paragraph unit sequence is segmented according to the determination result.

2. The method for dividing the boundaries of knowledge points in teaching resources as described in claim 1, characterized in that, The process of obtaining the original text of the teaching resources to be added to the database, in order to construct a sequence of paragraph units carrying the original text's location markers, includes: Boundary identification is performed on various paragraphs in the original text of the teaching resources to be added to the database to form initial teaching paragraph units; Extract the original text location markers from the original text of the teaching resources to be added to the database, and align the original text location markers with the initial teaching paragraph units to form candidate teaching paragraph units; The candidate teaching paragraph units are arranged in order to obtain the paragraph unit sequence.

3. The method for dividing the boundaries of knowledge points in teaching resources as described in claim 1, characterized in that, The step of extracting teaching scenario information from the paragraph unit sequence to annotate the relationship between adjacent paragraphs in the paragraph unit sequence and obtain the knowledge point attribution status of each paragraph includes: Based on the connection relationships between adjacent paragraphs in the paragraph unit sequence, a paragraph order relationship is constructed; Extract the heading level, question type hints, terminology context, and layout separation information of each paragraph from the paragraph unit sequence; Using the title hierarchy, the paragraph order is labeled with chapter affixation to form a title hierarchy affiliation relationship; Using the question type hints, the hierarchical relationship of the titles is marked with teaching functions to form the question type hint hierarchy; Using the terminology context, the attribution relationship of the question type prompts is labeled with conceptual orientation, thus forming a terminology context attribution relationship; Using the layout separation information, paragraph boundaries are marked to define the contextual attribution of the terms, thus forming a layout separation attribution relationship. The relationship between paragraph order, heading level, question type, terminology context, and layout separation is integrated to generate the attribute status of the knowledge points.

4. The method for dividing the boundaries of knowledge points in teaching resources as described in claim 1, characterized in that, The step of generating a state transition description based on the differences in the attribution status of adjacent knowledge points and integrating the teaching scenario information to obtain a teaching function transition adjustment amount includes: Based on the attribution status of the knowledge points, the direction of state change between adjacent paragraphs is calculated to obtain the state transition description. Based on the title level and the question type prompts, the teaching function transfer direction is marked on the state transition description to form a teaching function transfer description; Based on the terminology context and the layout separation information, the description of teaching function transfer is checked for conceptual orientation and paragraph separation to form the meaning of teaching function transfer. Based on the meaning of the teaching function transfer, the adjustment amount of the teaching function transfer is calculated.

5. The method for dividing the boundaries of knowledge points in teaching resources as described in claim 1, characterized in that, The method of updating the posterior parameters of the Koopman operator includes: Based on the state transition sample accuracy, the degree of participation of the state transition sample formed by the adjacent knowledge point attribution states is calculated; Based on the participation level of the state transition samples, the state transition samples in the Koopman operator Bayes posterior update stage are allocated to obtain the posterior update input. The posterior parameters of the Koopman operator are updated based on the posterior update input.

6. The method for dividing the knowledge point boundaries of teaching resources as described in claim 1, characterized in that, The step of determining adjacent paragraphs based on the evolution result of the knowledge point's attribution status, and segmenting the paragraph unit sequence according to the determination result, includes: Based on the evolution results of the knowledge point attribution status, calculate the state evolution deviation between adjacent paragraphs; The state evolution deviation is compared with a preset condition threshold, and a list of marked paragraphs is generated based on the comparison result between adjacent paragraphs. Extract knowledge point boundaries from the list of marked paragraphs, and use the knowledge point boundaries to segment the paragraph unit sequence.

7. The method for dividing the boundaries of knowledge points in teaching resources as described in claim 6, characterized in that, The step of comparing the state evolution deviation with a preset condition threshold and marking adjacent paragraphs based on the comparison result includes: When the state evolution deviation exceeds a preset threshold, adjacent paragraphs are marked as the same knowledge point unit; When the state evolution deviation does not exceed the preset condition threshold, the adjacent paragraphs are marked as knowledge point boundaries.

8. A knowledge point boundary segmentation system for teaching resources, characterized in that, include: The sequence construction module is used to obtain the original text of the teaching resources to be added to the database, so as to construct a sequence of paragraph units carrying the original text position markers; The paragraph association annotation module is used to extract teaching scenario information from the paragraph unit sequence to annotate the association relationship between adjacent paragraphs in the paragraph unit sequence, and obtain the knowledge point attribution status of each paragraph; The teaching function transfer analysis module is used to generate a state transfer description based on the differences in the ownership status of adjacent knowledge points, and to obtain the teaching function transfer adjustment amount by integrating the teaching scenario information. The posterior update module is used to introduce the teaching function transfer adjustment into the MetaKoopman operator Bayes posterior update loop, and output the state transition sample accuracy to update the Koopman operator posterior parameters. The state evolution module is used to perform state evolution on the adjacent knowledge point ownership states based on the updated Koopman operator posterior parameters, and obtain the knowledge point ownership state evolution results. The sequence segmentation module is used to determine adjacent paragraphs based on the evolution result of the knowledge point attribution status, and to segment the paragraph unit sequence according to the determination result.

9. A computer device, characterized in that, It includes a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor, wherein the processor executes the computer program to implement the teaching resource knowledge point boundary segmentation method as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program, wherein when the device containing the computer-readable storage medium executes the computer program, it implements the teaching resource knowledge point boundary segmentation method as described in any one of claims 1 to 7.

Images