A digital-based construction method of a knowledge graph of a thought political course

Abstract

The application relates to the technical field of knowledge graph construction, and discloses a digital-based ideological and political course knowledge graph construction method. The method acquires ideological and political course voice, course structure, main knowledge graph and homologous evidence materials, encapsulates candidate semantic objects obtained by voice transcription into voice source objects, determines original role and generates structure-restricted alternative roles, respectively inserts the same voice source objects into a shadow graph which does not participate in formal service, extracts difference path segments between two types of placeholder objects trigger paths, determines structure propagation states according to the first appearance position of the difference path segments, continuous expansion range and consistent or conflict relationship with the source fragment course structure position, writes into the main knowledge graph, writes into a to-be-proved node cache or performs degradation processing, and releases, maintains or degrades the to-be-proved node based on the homologous evidence materials. The application can inhibit structural error propagation before the voice recognition result is written into the main knowledge graph.

Description

Technical Field

[0001] This invention relates to the field of knowledge graph construction technology, and more specifically, this application relates to a method for constructing a digital knowledge graph for ideological and political education courses. Background Technology

[0002] With the development of digital course construction, the chapter system, theoretical concepts, teaching resources, and retrieval entry points in ideological and political courses often need to be organized using knowledge graphs. This allows the course platform to provide retrieval, display, and access services according to knowledge point hierarchy, conceptual relationships, and resource associations. Since classroom lectures contain a large number of theoretical concepts and colloquial explanations, the actual construction process often involves first recognizing the classroom audio to obtain transcribed text, then extracting keywords, conceptual terms, or relational expressions from the transcribed text, and generating nodes, relationships, and index information in the course knowledge graph accordingly.

[0003] In this approach, speech recognition results are typically used as the foundational input for subsequent knowledge extraction and image processing. To mitigate the impact of recognition errors, existing solutions often focus on the accuracy of the transcribed text itself. For example, they may filter out low-confidence words based on recognition confidence levels, or verify the extracted terms using materials such as courseware, textbooks, or teacher-proofread texts. These methods are effective in general text processing scenarios because most term errors only cause localized textual deviations and do not necessarily affect the overall organization of the subsequent data structure.

[0004] However, in the context of constructing a course knowledge graph, the structural impact of different terms entering the graph varies. If ordinary background words or illustrative terms are misidentified, it typically only affects local segments; but if core theoretical terms, chapter-specific expressions, or relational predicates are misidentified and further generated into higher-level nodes, key connections, or a broader course index, the error will continue to amplify along the graph hierarchy, concept connections, and resource retrieval paths. Therefore, existing technologies struggle to address the problem of classroom speech recognition errors evolving from local textual errors into structural graph errors during the construction of ideological and political education course knowledge graphs. Summary of the Invention

[0005] To address the aforementioned technical issues, this paper provides a method for constructing a knowledge graph for ideological and political education courses based on digitalization. This technical solution resolves the problems mentioned in the background section.

[0006] To achieve the above objectives, the technical solution of the present invention is as follows:

[0007] This application provides a method for constructing a knowledge graph for ideological and political education courses based on digitalization, including:

[0008] Acquire the audio, course structure, main knowledge graph and corresponding evidence materials of ideological and political courses, transcribe the audio of ideological and political courses and extract candidate semantic objects, and encapsulate the candidate semantic objects with source segments, time positions, recognition confidence and course structure positions into audio source objects;

[0009] Based on the matching relationship between the voice source object and the existing nodes, relation edges and course affiliation index in the main knowledge graph, the original proposed role is determined, and a structurally restricted alternative role is generated based on at least one of the following constraints: local terms in the source segment, aliases of existing nodes or course indexes in the source segment.

[0010] A shadow graph that does not participate in the formal service is generated from the main knowledge graph. Objects from the same voice source are placed into the original role and the structurally restricted alternative role respectively, forming the original placeholder object and the alternative placeholder object.

[0011] Track the parent attachment paths, relationship connection paths, and course affiliation index paths triggered by the original placeholder object and the replacement placeholder object, and align and extract the difference path segments according to path level, node, relationship edge, and index unit.

[0012] The structure propagation status is determined based on the first occurrence position of the difference path segment, the continuous expansion range, and the consistency or conflict between the difference path segment and the existing superior node, relation endpoint, relation type, relation direction, and course affiliation index range corresponding to the course structure position of the source segment.

[0013] Based on the structural propagation state, the voice source object is written into the main knowledge graph, written into the cache of the node to be verified, or downgraded to a local candidate object. For the node to be verified written into the cache of the node to be verified, the term content of the node to be verified and the superior node, relation endpoint, relation type, relation direction or course affiliation index range corresponding to the original role are verified based on the corresponding evidence materials. The node to be verified is released, maintained or downgraded according to the verification results.

[0014] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0015] This application generates original roles and structurally restricted alternative roles for the same voice source object, and inserts the dual roles in the shadow graph that does not participate in the formal service. The difference path segment is used to locate the additional path unit generated by the original role relative to the minimum structural change. The difference path segment is not directly used as a risk conclusion, but is compared with the allowed superior node, relation endpoint, relation type, relation direction and course index range of the corresponding course structure position in the main knowledge graph, so as to identify the incremental structural changes that need to be reviewed.

[0016] This application determines the structure propagation state based on the first occurrence position of the difference path segment, the continuous expansion range, and the consistency or conflict relationship between the segment and the source fragment course structure position. It then diverts the voice source object to the main knowledge graph, the cache of the node to be verified, or the local candidate object, thereby blocking the mis-construction of high-level nodes, the mis-connection of relationship chains, and the diffusion of course indexes before writing. Attached Figure Description

[0017] The disclosure of this invention is illustrated with reference to the accompanying drawings. It should be understood that the drawings are for illustrative purposes only and are not intended to limit the scope of protection of this invention. Wherein:

[0018] Figure 1 This is a flowchart of a method for constructing a digital knowledge graph for ideological and political education courses according to the present invention.

[0019] Figure 2 This is a data flow diagram of a digital-based knowledge graph construction method for ideological and political education courses, as described in this invention. Detailed Implementation

[0020] It is readily understood that, based on the technical solution of this invention, those skilled in the art can propose various interchangeable structural methods and implementations without altering the essential spirit of the invention. Therefore, the following detailed embodiments and accompanying drawings are merely illustrative examples of the technical solution of this invention and should not be considered as the entirety of the invention or as limitations or restrictions on the technical solution of this invention.

[0021] In the existing construction of digital ideological and political education courses, course knowledge graphs are typically used to organize course chapters, theoretical concepts, teaching resources, knowledge point relationships, and course retrieval paths. The quality of its construction depends not only on the ability to extract correct terms from classroom audio, courseware text, or textbook text, but also on whether the extracted semantic content can be placed in a position that matches the source fragment, course chapter, and graph structure. In other words, constructing an ideological and political education course knowledge graph is not simply about writing the transcribed text into the graph; it requires ensuring a consistent relationship between the source location, semantic role, and graph connection path of candidate semantic objects to provide a reliable foundation for subsequent course retrieval, resource access, and knowledge point association.

[0022] However, existing technologies typically use classroom speech recognition results as the definitive input for keyword extraction, semantic recognition, and knowledge graph construction, focusing primarily on the accuracy of the transcribed text itself. They lack pre-writing verification to account for the structural propagation impact of speech recognition results after they enter the knowledge graph. Errors in recognizing ordinary background words or local descriptive words usually only affect the local text; however, once core theoretical terms, chapter-specific expressions, or relational predicates are mistakenly generated as higher-level nodes, key relational edges, or a large-scale course index, the error can continue to spread along node levels, concept connection paths, and course affiliation indexes. Furthermore, even if a candidate term can be found with the same expression in courseware, textbooks, or teacher-proofreading text, it does not necessarily mean that the term should assume the same graph role in the current classroom segment. Without distinguishing the term's location, adjacent semantic objects, and the original intended graph role, the system may still incorrectly release content that should be an existing node alias, local annotation, or classroom segment index as a new node or formal relational edge.

[0023] For example, in the course "Basic Principles of Marxism," when a teacher explains the knowledge point of "social existence and social consciousness," the speech recognition system might extract "social existence determines social consciousness" as a complete theoretical statement and mistakenly classify it as an independent first-level chapter node. If the system only processes the term based on completeness or recognition confidence, it might directly attach it to the course root node, causing content that originally belonged to a second-level knowledge point to be incorrectly promoted to a higher-level node. Similarly, when a teacher expresses "it cannot be understood as consciousness determining social existence," if the system only recognizes the term "consciousness determines social existence" without considering negation semantics and existing relational directions, it might mistakenly generate a relational edge of "social consciousness determines social existence." Once these errors enter the main knowledge graph, they will change the reachability paths of related concepts, the scope of the course index, and the subsequent retrieval affiliation.

[0024] If the aforementioned problems are not addressed, the digital ideological and political education course knowledge graph construction system will struggle to identify and suppress the structural spread of speech recognition errors. Incorrectly constructed high-level nodes can lead to erroneous expansions of the course chapter system; misidentification of relational predicates can result in incorrect connections between theoretical concepts; and incorrect expansion of the course index can cause classroom segments, courseware pages, and textbook paragraphs to be categorized as incorrect knowledge points. Consequently, localized speech recognition errors can be amplified into systemic shifts in the graph hierarchy, relational chains, and index ranges, thereby affecting the reliability of the ideological and political education course knowledge graph construction and the accuracy of digital course resource retrieval.

[0025] In this application, "main knowledge graph" refers to a knowledge graph used for formal course retrieval, learning path recommendation, teaching resource access, or course knowledge service; "shadow graph" refers to a temporary graph copy or isolated graph environment generated by the main knowledge graph that does not participate in formal service, used to simulate the structural propagation effect of candidate objects before they are formally written.

[0026] "Speech source object" refers to a data object formed by binding candidate semantic objects extracted from classroom speech-to-text with their source segments, time locations, recognition confidence levels, and course structure locations. Speech source objects are not simply word terms, but candidate graph writing objects with constraints on source, location, and course structure.

[0027] "Adjacent semantic objects" refer to semantic objects located within the same source segment as the candidate semantic object, and forming a sequential, modifying, parallel, causal, inclusive, or subordinate relationship with the candidate semantic object in terms of transcribing text order, semantic dependency, or curriculum structure attribution. For example, in source segment T1, the adjacent semantic objects of the candidate semantic object "social existence determines social consciousness" include "historical materialism" which is sequential in transcribing text order, and "social consciousness" which forms an inclusive relationship in terms of semantic dependency.

[0028] "Proposed role" refers to the new node role, new relationship edge role, or new course index role that the voice source object is to assume in the main knowledge graph after it has been matched with the existing nodes, relationship edges, and course affiliation indexes in the main knowledge graph.

[0029] "Structure-constrained alternative roles" refer to alternative roles generated for the same speech source object under constraints such as source segments, existing node aliases, nodes in the same chapter, or course indexes within segments, without expanding the graph hierarchy and course index range. These alternative roles can be existing node aliases, local annotations, candidate objects in the same chapter, or index objects within segments.

[0030] Local terms within the source fragment, existing node aliases, and course indexes within the source fragment are three types of constraint information used when generating structurally constrained substitute roles, and they are not graph roles at the same level. Structurally constrained substitute roles formed based on local terms are local annotations or candidates within the same chapter; structurally constrained substitute roles formed based on existing node aliases are existing node aliases; and structurally constrained substitute roles formed based on course indexes within the source fragment are index objects within the fragment.

[0031] Structurally constrained alternative roles are used only to represent the minimum structural changes without adding new parent nodes, changing existing relation endpoints and directions, or expanding the course index range. These roles are not presupposed to be correct. Difference path segments only represent additional path units generated by the original role relative to this minimum structural change. The existence of difference path segments is not itself equivalent to structural risk; whether a difference path segment constitutes structural risk is determined by whether it violates the existing parent nodes, relation endpoints, relation types, relation directions, and course affiliation index ranges in the main knowledge graph corresponding to the course structure position of the source fragment.

[0032] "Difference path segment" refers to the graph path portion triggered additionally by the original placeholder object relative to the replacement placeholder object. The path portion may include additional parent nodes in the parent attachment path, additional relation edges or reachable nodes in the relation connection path, and additional index units in the course affiliation index path.

[0033] "Structural propagation state" refers to the state result determined based on the first occurrence position of the differential path segment in the graph, the continuous expansion range, and the consistency or conflict relationship between it and the position of the source segment course structure. It includes at least the writable state, the unproven state, and the degraded state.

[0034] "Corresponding evidence object" refers to an evidence object generated from evidence fragments in courseware text, textbook text, or teacher proofreading text that are associated with the course structure position of the node to be proved. Its fields may include evidence terms, evidence course structure position, adjacent semantic objects, and existing graph nodes that can be connected.

[0035] To illustrate the correspondence between the main knowledge graph, course structure positions, relation edges, and course affiliation index in this application, in one implementation, the main knowledge graph for ideological and political education courses may include course root nodes, first-level chapter nodes, second-level knowledge point nodes, relation edges, and course affiliation indexes. This example is only used to illustrate the data structure and processing flow of this application and does not constitute a limitation on course names, chapter names, number of nodes, or relationship types.

[0036] For example, the main knowledge graph includes the following nodes:

[0037] Root node R1: Basic principles of Marxism;

[0038] Chapter 1 Node C1: The Materiality of the World and its Laws of Development;

[0039] Chapter 1 Node C2: Practice and Cognition and Their Development Laws;

[0040] Chapter 1 Node C3: Human Society and Its Development Laws;

[0041] Level 2 knowledge point K31: Social Existence and Social Consciousness;

[0042] Level 2 knowledge point K32: Basic social contradictions;

[0043] Level 2 knowledge point K33: The people are the creators of history;

[0044] Fragment index node I31: corresponds to a classroom lecture fragment on "Social Existence and Social Consciousness";

[0045] Resource node M31: corresponds to a courseware page, textbook paragraph, or teacher-proofread text segment.

[0046] For example, the main knowledge graph includes the following relation edges:

[0047] R1 connects C1, C2, and C3 through "containment" edges;

[0048] C3 connects K31, K32, and K33 through an edge that represents a "containment" relationship;

[0049] K31 connects the content expression of "social existence determines social consciousness" through the "interpretation" of relational edges;

[0050] K31 connects classroom cases or teaching resources related to the knowledge point through "related case" relationship edges;

[0051] I31 is connected to K31 through a "belonging" relationship edge;

[0052] M31 connects to K31 or I31 through the "source of evidence" relationship edge.

[0053] For example, a course attribution index can include the correspondence between course structure location, courseware page, teaching segment, and time location. For instance, course structure location L31 corresponds to secondary knowledge point node K31 under primary chapter node C3, and specifically to page 12 of the courseware, the first section of chapter 3 of the textbook, and the classroom audio time period from 00:18:20 to 00:18:35. This course attribution index is used to limit the source location and verifiable scope of the audio source object when it enters the main knowledge graph.

[0054] In the above example, if a certain voice source object originates from course structure position L31, then the original placeholder object and the alternative placeholder object formed by the voice source object in the shadow map can both correspond to L31, which is used to determine whether the triggered superior attachment path, relationship connection path or course affiliation index path exceeds the course structure position where the source segment is located.

[0055] like Figure 1-2 As shown, this embodiment provides a method for constructing a digital knowledge graph for ideological and political education courses, executed by a course knowledge graph construction system. The course knowledge graph construction system can be deployed on a server, a course resource platform, a local computing device, or a cloud computing environment. The method in this embodiment includes the following steps.

[0056] S1. Obtain the audio, course structure, main knowledge graph and corresponding evidence materials of the ideological and political course; transcribe the audio of the ideological and political course and extract candidate semantic objects; encapsulate the candidate semantic objects with the source segment, time position, recognition confidence and course structure position into audio source objects.

[0057] The process of obtaining audio recordings, course structure, main knowledge graphs, and equivalent evidence materials for ideological and political education courses includes:

[0058] The audio of ideological and political courses is read through the course recording interface, live audio interface, or teaching video audio track; the course structure is read through the course management database, courseware catalog, courseware page turning log, or teaching timeline; the main knowledge graph is read through the graph database interface to read the nodes, relation edges, and course affiliation index of the current effective version; the corresponding evidence materials are read through the courseware text library, textbook text library, or teacher proofreading text library, and the source identifier, version identifier, and corresponding course structure position are recorded for each evidence fragment.

[0059] The course structure is determined by aligning the time position of audio segments with the chapter timeline, courseware page-turning log, or teaching segment identifiers. During alignment, the corresponding courseware page or teaching segment is first determined based on the start and end times of the audio segment, and then the corresponding chapter node or knowledge point node is determined based on the course index identifier of the courseware page or teaching segment.

[0060] The source identifiers are recorded for materials used to determine the position of the course structure and for corresponding evidence materials used to verify the original role. When a piece of material has already been used to determine the position of the course structure, it can continue to be used to prove the position of the evidence fragment, but it cannot be used alone as sufficient evidence to support the original role; the original role should also be supported at the role level by at least one of the following sources: textbook text, courseware text, or teacher-proofread text.

[0061] The audio for ideological and political education courses can be classroom recordings, live course audio, pre-recorded course audio, or audio streams from teaching videos. The course structure can include one or more of the following: course name, chapter hierarchy, courseware pages, teaching paragraphs, course index units, or teaching timeline. The main knowledge graph can include existing course nodes, chapter nodes, theoretical concept nodes, relationship edges, node aliases, and course affiliation indexes. Supporting evidence can include at least one of the following: courseware text, textbook text, or teacher-proofread text.

[0062] Speech recognition is performed on the audio of ideological and political education courses to obtain transcribed text with time location and recognition confidence. Candidate semantic objects are extracted from the transcribed text. These can be theoretical concepts, chapter names, knowledge point descriptions, relational predicates, case names, or course index terms. The candidate semantic objects are then bound to their source segments, time locations, recognition confidence, and course structure positions to obtain the audio source objects.

[0063] In one implementation, the course structure location can be determined by the correspondence between the time position of an audio segment and the courseware page, chapter timeline, or teaching paragraph. For example, if an audio segment corresponds to the third section of the second chapter of a course, then all candidate semantic objects extracted within that segment are first bound to that course structure location. This binding does not mean that the candidate semantic objects can only be written to that location, but rather provides a data basis for subsequently determining whether their intended roles exceed the scope of the source structure.

[0064] In one implementation, the course structure location is determined by at least one of the following: course chapter, courseware page, teaching segment, or course index unit corresponding to the time location. The course structure location is used to limit the attachment and indexing range when the voice source object enters the main knowledge graph. That is, the parent node to which the voice source object is to be attached should be located within the set of allowed parent nodes corresponding to the course structure location, and the course index unit to which it is to belong should be located within the set of allowed indexes corresponding to the course structure location.

[0065] The speech source object obtained in this step includes at least the following fields: candidate semantic object identifier, word content, source segment, time location, recognition confidence, and course structure location. Optionally, it may also include preceding words, following words, adjacent semantic objects, and the page or teaching segment in the courseware.

[0066] For example, in the classroom audio of the course "Basic Principles of Marxism", the system obtained the following speech-to-text fragment:

[0067] Segment Number Time and location Transcription of text fragments Candidate semantic objects ASR recognition confidence level Course Structure Location T1 00：18：20-00：18：25 "Social existence determines social consciousness" Social existence determines social consciousness 0.62 L31 T2 00：18：25-00：18：30 This is a fundamental viewpoint in historical materialism. historical materialism 0.86 L31 T3 00：18：30-00：18：35 It should not be interpreted as consciousness determining social existence. Consciousness determines social existence 0.58 L31

[0068] Here, L31 represents the position in the course structure, which corresponds to the second-level knowledge point node K31 "Social Existence and Social Consciousness" under the first-level chapter node C3 "Human Society and its Development Laws". The system binds the candidate semantic object "Social Existence Determines Social Consciousness" in the segment T1 with the source segment T1, the time position 00:18:20-00:18:25, the recognition confidence level of 0.62, and the course structure position L31 to form the speech source object VO1.

[0069] In this embodiment, the recognition confidence is only one of the source fields of the speech source object, used to record the text reliability of the speech recognition result; the structure propagation state is not determined solely by whether the recognition confidence exceeds a certain fixed threshold, but by the consistency or conflict relationship between the subsequent difference path segment and the source segment course structure position.

[0070] For example, the voice source object VO1 can be represented as:

[0071] VO1={candidate semantic object: "social existence determines social consciousness"; source fragment: T1; time position: 00:18:20-00:18:25; recognition confidence: 0.62; course structure position: L31; adjacent semantic objects: "historical materialism" "social consciousness"}.

[0072] This step binds candidate semantic objects to source segments, temporal locations, recognition confidence levels, and course structure locations, making them traceable speech source objects constrained by course location. This prevents speech recognition results from being mistakenly written into the main knowledge graph as independent nodes, relational edges, or course indexes after being separated from the original classroom segments.

[0073] S2. Based on the matching relationship between the voice source object and the existing nodes, relation edges and course affiliation index in the main knowledge graph, determine the original role, and generate a structurally restricted alternative role based on at least one of the following constraints: local terms in the source segment, aliases of existing nodes or course indexes in the source segment.

[0074] The process of matching the voice source object with existing nodes, relation edges, and course affiliation indexes in the main knowledge graph to determine its original role includes:

[0075] Specifically, the word content, adjacent semantic objects, and course structure positions of the speech source object are mapped to the existing node names, node aliases, relation edge types, and course affiliation indexes in the main knowledge graph.

[0076] The correspondence is determined according to the following fields: when the term content of the candidate semantic object has the same normalized term identifier as the existing node name or node alias, a node match is formed; when the adjacent semantic objects of the candidate semantic object correspond to existing nodes in the main knowledge graph, and the candidate semantic object can determine the relationship type and relationship direction, a relationship edge match is formed, where the two existing nodes serve as the relationship start node and relationship end node, respectively; when the course structure position bound to the candidate semantic object is consistent with the course index identifier corresponding to the course chapter, courseware page, teaching paragraph, or time interval, a course affiliation index match is formed.

[0077] When a candidate semantic object does not form an existing node match, but its direct superior node and target node level can be determined by the course structure position, it is recorded as a candidate for the new node role; when a relationship edge match is formed and there is no relationship edge in the main knowledge graph with the same start node, end node, relationship type, and relationship direction, it is recorded as a candidate for the new relationship edge role; when a course belonging index match is formed and there is no course index record in the main knowledge graph with the same target index unit and index range, it is recorded as a candidate for the new course index role.

[0078] When only one role candidate is valid, that role candidate is determined as the original role; when two or more role candidates are valid at the same time, the role candidates that do not match are first excluded according to the target object type and target level corresponding to the position in the course structure; if the original role still cannot be uniquely determined, the voice source object is recorded as insufficient role evidence and processed as pending proof.

[0079] When the matching result points to a newly added concept, the corresponding newly added node role is determined as the original role; when the matching result points to a relationship connection between two existing nodes, the corresponding newly added relationship edge role is determined as the original role; when the matching result points to the scope of a course chapter, courseware page, or teaching segment, the corresponding newly added course index role is determined as the original role.

[0080] For example, if the term content of a speech source object does not completely correspond to an existing node name, but its adjacent semantic objects correspond to two existing theoretical concept nodes, the system can designate the speech source object as a relationship edge role between the two nodes. Similarly, if a speech source object is presented as a chapter-based description in the transcribed text, the system can designate it as a newly added parent node role. These designations are not directly written into the main knowledge graph but serve as input for subsequent shadow graph verification.

[0081] This step clarifies the graph role an object representing a speech source might assume when directly written by matching it with existing nodes, relationship edges, and course affiliation indexes in the main knowledge graph. This allows for the identification of the potential impact of the object on node, relationship, or index structure before the actual writing process, providing a basis for comparison in subsequent shadow graph verification.

[0082] The process of generating a structurally constrained alternative role based on the source segment of the audio source object, existing node aliases, or course index within the segment includes:

[0083] Specifically, local terms and adjacent semantic objects adjacent to candidate semantic objects are extracted from the source segment and mapped to existing node aliases, same-chapter nodes, or segment-level course indexes in the main knowledge graph. When a corresponding object exists, the speech source object is limited to existing node aliases, local annotations, same-chapter candidate objects, or segment-level index objects to form a structurally restricted substitute role. When no corresponding object exists, a source segment-restricted substitute role is generated based on the source segment, time position, and course structure position bound to the speech source object. This ensures that the speech source object only forms a segment placeholder object in the shadow graph bound to the corresponding course structure position of the source segment, without establishing new parent nodes or formal relationship edges.

[0084] The role of structurally restricted substitute roles is to provide source-specific correspondences for objects from the same speech source. These substitute roles do not aim to expand the map hierarchy or the scope of the course index, but rather to limit themselves to the source fragments, adjacent semantic objects, and the range supported by the existing map structure. Therefore, we can subsequently compare the path differences between the original role and the structurally restricted substitute roles after they enter the shadow map.

[0085] In one implementation, if a candidate semantic object is similar to an existing node alias, the structurally restricted substitute role can be the existing node alias; if a candidate semantic object corresponds to a local description in the current course paragraph, the structurally restricted substitute role can be a local annotation; if a candidate semantic object can only support the current segment index, the structurally restricted substitute role can be an in-segment index object.

[0086] When there are two or more structurally restricted substitute roles for the same speech source object, determine the number of new or changed hierarchical levels, relation edges, and course index units for each structurally restricted substitute role relative to the main knowledge graph. Compare these numbers sequentially, selecting the structurally restricted substitute role with the smallest structural change. If the comparison results are still the same, prioritize the structurally restricted substitute role that can be directly associated with an existing node or index unit corresponding to the course structure position of the speech source object. If a unique role still cannot be determined, use a placeholder object for a segment that is only bound to the course structure position of the source segment as the structurally restricted substitute role.

[0087] The process of determining structurally restricted alternative roles must not add any higher-level nodes outside the course structure position, must not change the relation endpoints, relation types, or relation directions of existing relation edges, and must not extend the course affiliation index beyond the range corresponding to the course structure position of the source fragment.

[0088] This step utilizes local terms within the source segment, existing node aliases, or segment-based course indexes to generate alternative roles for the same speech source object, constrained by the source location and existing graph structure. This allows for judgment based on the structural differences between the original proposed role and the constrained role, avoiding the erroneous expansion of speech results that can be categorized into existing node aliases, local annotations, or segment-based indexes into high-level graph objects.

[0089] S3. Generate a shadow graph from the main knowledge graph that does not participate in the formal service, and place the same voice source object into the original role and the structurally restricted replacement role respectively to form the original placeholder object and the replacement placeholder object;

[0090] Shadow graphs are generated from the main knowledge graph. A shadow graph can be a temporary copy of the main knowledge graph, a local subgraph copy, or a temporary graph space with isolation markers. Shadow graphs do not participate in formal course retrieval, learning path recommendation, teaching resource access, or formal graph updates.

[0091] Place objects from the same speech source into the shadow map according to their original intended roles to form the original intended placeholder objects; and place objects from the same speech source into the shadow map according to their structurally constrained alternative roles to form alternative placeholder objects.

[0092] In one implementation, the original placeholder object can be used to create a new node, a new relation edge, or a new course index in the shadow graph; the alternative placeholder object can be used to create an alias for an existing node, a local annotation, a candidate object for the same chapter, or an index object within a segment in the shadow graph. Since the two types of placeholder objects originate from the same speech source object, the differences in subsequent paths can reflect the differences in the impact of different roles of the same object on the graph structure.

[0093] This step generates a shadow graph that does not participate in the formal service, and places objects from the same speech source into both their original intended roles and structurally constrained alternative roles, exposing potential changes to the graph structure caused by candidate objects in an isolated environment. This allows for the simulation of the impact on graph hierarchy, relational connections, and course indexes in advance without polluting the main knowledge graph.

[0094] S4. Track the parent attachment path, relationship connection path and course affiliation index path triggered by the original placeholder object and the replacement placeholder object, align them by path level, node, relationship edge and index unit and extract the difference path segments.

[0095] Starting from the original placeholder object and the replacement placeholder object respectively, trace the parent attachment path, relationship connection path and course affiliation index path triggered by the two.

[0096] A parent-child attachment path can represent the attachment relationship between a placeholder object and its parent chapter node, parent theoretical concept node, or root node. A relationship connection path can represent the connection relationship between a placeholder object and adjacent nodes, reachable nodes, or relationship edges. A course affiliation index path can represent the affiliation relationship between a placeholder object and course chapters, courseware pages, teaching paragraphs, or resource index units.

[0097] Align the paths triggered by the original placeholder object with those triggered by the alternative placeholder object according to path level, nodes, relation edges, and index units, and identify the path segments triggered only by the original placeholder object and not by the alternative placeholder object as differential path segments.

[0098] For example, the path triggered by the intended placeholder object can be represented as: ;

[0099] in, This indicates the parent attachment path triggered by the originally intended placeholder object. This indicates the relationship connection path triggered by the originally intended placeholder object. This indicates the course affiliation index path triggered by the originally intended placeholder object.

[0100] The path triggered by the replacement placeholder object can be represented as follows: ;

[0101] in, , , These represent the parent attachment path, relationship connection path, and course affiliation index path triggered by the replacement placeholder object, respectively.

[0102] In one implementation, the difference path segment can be represented as: ;

[0103] in, This indicates the order based on path hierarchy, nodes, relation edges, and index units. The aligned set of paths; "\" represents the set difference operation. This represents the additional path segment triggered by the original placeholder object relative to the replacement placeholder object. The above formula is used to illustrate an optional calculation method for the difference path segment and does not constitute a limitation on the scope of protection of this application.

[0104] Align the original placeholder object path and the replacement placeholder object path into a path unit sequence according to the triggering order. The first occurrence position is the path unit position where the first node identity, relation endpoint, relation type, relation direction, or index unit in the two sequences are different. Starting from the first occurrence position, consecutive path units that differ until the same path unit appears again or the path end point is reached, forming a continuous difference path segment.

[0105] The continuously expanded range is recorded as a triple E(ΔP)=(nh,nr,ni), where nh is the number of newly added or changed higher-level attachments in the difference path segment, nr is the number of newly added or changed relation edges, and ni is the number of newly added or changed course index units. These quantities are used to record the scope of the difference's impact and are not used as a separate risk threshold.

[0106] A closed differential path segment within the source segment's course structure location means that the terminating node, relation endpoint, and index unit of the differential path segment all belong to the allowed set corresponding to the source segment's course structure location, and there are no differential path units that continue to point outside the allowed set.

[0107] This step involves tracing the parent link paths, relationship connection paths, and course affiliation index paths triggered by two types of placeholder objects, and aligning them by path level, nodes, relationship edges, and index units to extract the differential path segments triggered only by the original intended role. This transforms the risk of graph structure propagation into traceable and comparable path differences, providing an endogenous basis for subsequent state judgment.

[0108] S5. Determine the structure propagation status based on the first occurrence position and continuous expansion range of the difference path segment, as well as the consistency or conflict between the difference path segment and the existing superior node, relation endpoint, relation type, relation direction and course affiliation index range corresponding to the source segment course structure position.

[0109] Based on the first occurrence location, continuous expansion range, and consistency or conflict relationship between the differential path segment and the course structure location of the source segment, the structural propagation state of the speech source object is determined.

[0110] In this application, the difference path segment is used to locate the incremental structural changes generated by the original role relative to the structurally constrained substitute role. The structural propagation state is determined by the relationship between the incremental structural changes and the existing structural constraints of the main knowledge graph, rather than by the existence of the difference path segment.

[0111] For the source fragment course structure position Ls, the existing nodes, relation edges, and course affiliation index corresponding to Ls are read from the main knowledge graph to form the allowed set of parent nodes U(Ls), the allowed set of relations R(Ls), and the allowed set of indices I(Ls). Each relation record in the allowed set of relations R(Ls) includes at least a start node, an end node, a relation type, and a relation direction. Therefore, even if the differing path segment does not exceed the same chapter, if the relation endpoint, relation type, or relation direction is inconsistent with R(Ls), it is still determined as a relation connection conflict.

[0112] For path unit e in the difference path segment, it is considered consistent when e belongs to the corresponding U(Ls), R(Ls), or I(Ls); it is considered conflicting when e conflicts with the existing superior node level, relation endpoint, relation type, relation direction, or course affiliation index range in the corresponding set. Conflicts include e pointing to a superior node outside the allowed superior node set, the relation endpoint or relation direction of e being opposite or inconsistent with the same type of relation record in the allowed relation set, and e exceeding the index range limited by the allowed index set; when the main knowledge graph and source fragment information are insufficient to determine whether e belongs to the corresponding set, it is considered insufficient evidence.

[0113] In this application, "the original proposed role cannot be supported by the allowed set of parent nodes, allowed set of relations, or allowed set of indexes" means that each path unit in the differential path segment does not conflict with the corresponding allowed set, but the allowed set does not contain a corresponding record that supports the original proposed role as a new node role, a new relation edge role, or a new course index role. For example, the existing record in the allowed set corresponding to the candidate semantic object can only support merging it into an existing node alias, local annotation, or fragment index object. Thus, conflicting path units trigger a pending verification state, no conflict but lack of role support triggers a downgraded state, and no conflict and the existence of a corresponding supporting record in the allowed set triggers a writable state. These three states are mutually distinguishable.

[0114] The structural propagation state is determined in the order of pending verification, downgraded state, and writable state. A state is determined to be pending verification if there is at least one conflicting path unit in the differing path segment. If no conflicting path unit exists, and the original proposed role cannot be supported by the allowed set, but the candidate semantic object can be treated as an existing node alias, local annotation, or intra-segment index, it is determined to be downgraded. If no conflicting path unit exists and the downgraded state is not satisfied, and all path units corresponding to the original proposed role belong to the allowed set, it is determined to be writable. A state is determined to be pending verification if there is insufficient evidence in the path unit.

[0115] The above set affiliation rules are based on whether existing node levels, relationship endpoints, relationship types, relationship directions, and index ranges are violated. There is no need to manually specify "how many levels of extension" constitute a conflict. Continuous extension ranges are only used to record the degree of conflict impact and for subsequent review and location.

[0116] S6. Based on the structural propagation state, write the voice source object into the main knowledge graph, write it into the cache of the node to be verified, or downgrade it to a local candidate object. For the node to be verified written into the cache of the node to be verified, verify the term content of the node to be verified and the superior node, relation endpoint, relation type, relation direction or course affiliation index range corresponding to the original proposed role based on the corresponding evidence materials, and release, maintain or downgrade the node to be verified according to the verification results.

[0117] The process of writing off-flow based on the structure propagation state includes:

[0118] If the voice source object is in a writable state, then its corresponding node, relation edge, or course affiliation index is written to the main knowledge graph.

[0119] If the voice source object is in a pending verification state, it is written to the pending verification node cache. When writing to the pending verification node cache, the source segment, time position, course structure position, original role, structurally restricted substitute role, difference path segment, and conflict source that triggered the pending verification state are all written together. Conflict sources include at least one of the following: superior attachment extension, relational connection extension, or course affiliation index conflict corresponding to the difference path segment.

[0120] If the source object of the audio is in a downgraded state, it will be downgraded to a local candidate object, an existing node alias, or a local annotation. The downgraded object will not be included in the main knowledge graph as a new high-level node or key relationship edge, but it can be retained as a classroom segment description, node alias, or supplementary information for retrieval.

[0121] The above-mentioned write-and-distribute approach ensures that each state has a clear destination, preventing candidate objects from having no follow-up processing flow after being pending verification, frozen, or downgraded.

[0122] Based on the structure propagation status, this step writes the voice source object into the main knowledge graph, into the cache of the node to be verified, or downgrades it to a local candidate object, ensuring that different risky objects have a clear destination. This allows for the timely writing of trusted objects, the interception of objects with structure propagation risks, and the downgrading of objects that cannot stably support their original roles, reducing node-level pollution, relationship chain pollution, and index range pollution.

[0123] The process of generating corresponding evidence objects and releasing, retaining, or downgrading the node to be proven includes:

[0124] For a node to be proven in the cache, extract evidence fragments related to the course structure position of the node to be proven from the corresponding evidence materials to generate a corresponding evidence object. The corresponding evidence object includes at least evidence terms, evidence course structure position, adjacent semantic objects, and connectable existing graph nodes.

[0125] For example, for the node to be verified, DN1, its source speech segment is T1, the candidate semantic object is "social existence determines social consciousness", the course structure position is L31, the original role is the newly added first-level node role, and the structurally limited alternative role is a partial expression or alias of the existing second-level knowledge point node K31 "social existence and social consciousness".

[0126] The system extracted the following evidence fragments from the corresponding evidence materials:

[0127] Source of evidence Evidence location Evidence fragments Evidence terms Adjacent semantic objects Can connect to existing graph nodes Courseware Text Page 12, Title Section "Social Existence and Social Consciousness" Social Existence and Social Consciousness Historical materialism and social consciousness K31 Courseware Text Page 12, main text area "Social existence determines social consciousness" Social existence determines social consciousness Social existence, social consciousness K31 Textbook Chapter Three, Section One "Social existence determines social consciousness, and social consciousness, in turn, influences social existence." Social existence determines social consciousness Social existence, social consciousness K31 Teacher proofreading text 00:18:20-00:18:35 Corresponding proofreading section The content taught here belongs to the basic viewpoints under the knowledge point of 'social existence and social consciousness'. Social Existence and Social Consciousness Historical materialism and its basic viewpoints K31

[0128] Based on the aforementioned evidence fragments, the system generates corresponding evidence objects EV1, EV2, and EV3. Each corresponding evidence object includes an evidence term, its position within the evidence curriculum structure, adjacent semantic objects, and connectable existing graph nodes. For example, EV2 can be represented as:

[0129] EV2={Evidence term: "Social existence determines social consciousness"; Evidence course structure position: L31; Adjacent semantic objects: "Social existence" "Social consciousness"; Can connect to existing graph nodes: K31}.

[0130] When the system compares EV2 with the node to be proven DN1, if EV2 supports the existence of the term "social existence determines social consciousness" and its evidence course structure position is still L31, and it can connect to the existing graph node K31, then it means that the evidence supports DN1 as a local expression or alias under the existing secondary knowledge point node K31; if EV2 does not support DN1 as a new primary chapter node, then the system will not release DN1 as a new primary node, but will downgrade DN1 to an alias, local annotation or classroom segment index of K31.

[0131] Therefore, the evidence objects are not simply used to prove the appearance of terms, but to verify whether a consistent relationship is formed between terms, course structure positions, adjacent semantic objects, and graph roles.

[0132] The corresponding evidence object is compared with the node to be verified in terms of its position in the course structure, adjacent semantic objects, and original role. If the corresponding evidence object and the node to be verified are consistent in the above aspects and do not trigger the conflict source recorded by the node to be verified again, the node to be verified is released and written into the main knowledge graph. If the corresponding evidence object only covers the term content of the node to be verified, but does not cover the path corresponding to the original role, the node to be verified is downgraded to an existing node alias or a local annotation. If the corresponding evidence object and the node to be verified conflict in terms of its position in the course structure, adjacent semantic objects, or original role, the node remains in the state of being verified and the corresponding conflict source is recorded.

[0133] In this step, the co-occurrence evidence object does not simply prove the appearance of a term, but rather verifies whether the term holds true under the same or corresponding course structure position, adjacent semantic objects, and graph roles. Therefore, this release process can avoid erroneous writing caused by identical terms but different graph roles.

[0134] This step extracts corresponding evidence objects from the courseware text, textbook text, or teacher-proofread text that are associated with the course structure position of the node to be proven, and compares their consistency or conflict with the node to be proven in terms of course structure position, adjacent semantic objects, and the original proposed role. This allows for the differentiation between cases where "the term exists but does not support the original proposed role" and cases where "the term exists and supports the original proposed role," reducing erroneous releases caused by homonyms, near-synonyms, or cross-chapter references.

[0135] In addition to evidence terms, evidence course structure positions, and adjacent semantic objects, the corresponding evidence object also includes evidence source identifiers and role support fields corresponding to the original proposed role type. When the original proposed role is a new node role, the role support fields include the proposed new node name, allowed parent nodes, and node level; when the original proposed role is a new relation edge role, the role support fields include the relation start node, relation end node, relation type, and relation direction; when the original proposed role is a new course index role, the role support fields include the target course index unit and index range.

[0136] "Evidence supporting the original role" means that the evidence not only contains candidate terms, but also all role-supporting fields corresponding to the original role type. Merely proving that candidate terms appear in the same chapter, page, or teaching paragraph is not considered as supporting the original role.

[0137] When the material used to determine the position of the course structure shares the same source of evidence as the corresponding evidence object, the corresponding evidence object can only be used for position verification and cannot trigger the release of the node to be verified on its own. The original role of the node to be verified should also be supported by another source of evidence or by the teacher's proofreading text. Thus, the basis for determining the position of the course structure is separated from the basis for verifying the original role, avoiding the use of the same material to self-validate the same conclusion.

[0138] This application further proposes a process for secondary release of nodes that remain in a pending certification state, including:

[0139] For nodes that remain pending verification after processing, the system performs a secondary release process on the node after updating the corresponding evidence materials. Updating the corresponding evidence materials includes at least one of the following: supplementing courseware text, updating textbook text, or supplementing teacher-proofread text.

[0140] Specifically, the system reads nodes in the pending-verification state from the pending-verification node cache and obtains the original proposed role, structurally restricted alternative role, difference path segment, and conflict source corresponding to the node. Subsequently, the system extracts newly added co-existing evidence objects from the course structure location corresponding to the pending-verification node, the teaching paragraph corresponding to the source segment, or adjacent semantic objects within the same source segment as the candidate semantic object. Newly added co-existing evidence objects include newly added evidence terms, newly added evidence course structure locations, newly added adjacent semantic objects, and newly added connectable existing graph nodes.

[0141] The system will compare the newly added peer evidence object with the original proposed role, structurally restricted alternative role, and difference path segment of the node to be verified. If the newly added peer evidence object only covers the term content of the node to be verified, but does not cover the superior attachment path, relationship connection path, or course affiliation index path corresponding to the original proposed role, it means that the newly added evidence can only support the existence of the term, but cannot support the node to be verified being written into the main knowledge graph according to the original proposed role. The system will then downgrade the node to be verified to an existing node alias, local annotation, or classroom segment index.

[0142] If the newly added corresponding evidence object can cover the path corresponding to the original proposed role, and the path does not trigger the conflict source recorded in the node to be verified cache again, the system releases the node to be verified from the node to be verified cache and writes it into the main knowledge graph according to the original proposed role. If the newly added corresponding evidence object still conflicts with the course structure position, adjacent semantic objects, or connectable existing graph nodes of the node to be verified, the system maintains the pending verification state of the node to be verified and appends the newly added conflict content to the node to be verified cache.

[0143] Through the above processing, nodes awaiting verification that remain in a pending verification state have a clear destination of release, downgrading, or continued pending verification after the evidence materials are updated, avoiding long-term suspension of nodes awaiting verification and also avoiding incorrect writing into the main knowledge graph simply because the same terms appear in the new evidence.

[0144] This step, after updating the corresponding evidence materials, compares the newly added corresponding evidence objects with the original proposed role, structurally restricted alternative role, and differing path segments of the node to be proven, rather than simply releasing it based on term matching. This distinguishes between cases where "new evidence only proves the existence of a term" and cases where "new evidence can support the path corresponding to the original proposed role," creating a closed loop between releasing, downgrading, and maintaining the node to be proven, reducing the risk of long-term suspension of the node to be proven and erroneous writing into the main knowledge graph.

[0145] The cache for nodes awaiting verification records the creation time, the most recent evidence update time, the number of reviews, and the current status for each node. After each update of the corresponding evidence materials, the system re-executes role-level verification for nodes that remain in the pending verification state.

[0146] In one implementation, if 30 days have passed since the node to be verified was written into the cache, or if sufficient evidence to support the original role cannot be obtained after two consecutive updates of the same evidence materials, the node to be verified is transferred to the manual review queue. If the manual review confirms that the original role is established, it is written into the main knowledge graph according to the original role. If the manual review does not confirm the original role, it is downgraded to a local candidate object, and only the audit record is retained.

[0147] Local candidate objects undergo role-level verification again when the corresponding evidence materials are updated; those that gain role support can be upgraded to existing node aliases, local annotations, fragment index objects, or nodes to be verified; if no corresponding evidence is found after the review period expires, they are removed from the course retrieval service index, retaining only the source fragment and processing record. The aforementioned 30 days and two updates are merely optional implementation parameters and do not constitute a limitation on the scope of protection.

[0148] This application also proposes risk propagation review processing based on the source of conflict. When a subsequent voice source object and the node to be verified in the node cache are in the same course structure position, and their adjacent semantic objects or original proposed roles have a corresponding relationship, the parent attachment path, relationship connection path, or course affiliation index path corresponding to the source of conflict of the node to be verified is marked as a path to be reviewed. The path to be reviewed only indicates the path that needs to be reviewed more strongly, and is not directly used as a permanently prohibited path.

[0149] When a subsequent voice source object and the node to be verified are located in the same course structure, and there is a corresponding relationship between their adjacent semantic objects or original roles, the system still performs dual-role shadow graph insertion of the original role and structurally restricted substitute role for the subsequent voice source object. Only when the subsequent voice source object triggers a higher-level attachment conflict, relation endpoint / type / direction conflict, or course affiliation index conflict of the same type as the existing conflict source along the path to be verified, the system prevents the object from being directly written to the main knowledge graph and writes it to the cache of the node to be verified.

[0150] When subsequent audio source objects do not trigger the same type of conflict again, or when newly added co-existing evidence objects can provide corresponding role support fields, the system allows them to be written according to the status judgment result and cancels the corresponding pending review mark. Therefore, conflict sources only increase the review intensity of the same path, without directly converting unconfirmed doubts into prohibitive rules.

[0151] This step uses conflicts in the upper-level attachment extensions, relationship connection extensions, or course affiliation indexes recorded in the cache of the node to be verified as limiting paths, and conversely constrains the speech source objects corresponding to the same course structure positions or adjacent semantic objects. Thus, the exposed conflict sources can be used to prevent the repeated propagation of similar errors along the same graph path, reducing the risk of subsequent speech recognition errors causing mis-construction of high-level nodes, mis-connection of relationship edges, or diffusion of course indexes.

[0152] Scenario-based implementation examples:

[0153] The following example uses the classroom speech transcription result of the "Social Existence and Social Consciousness" knowledge point in the "Basic Principles of Marxism" course to illustrate the implementation method of this application. This embodiment is only used to illustrate the feasible process of this application and does not constitute a limitation on the course name, node name, or relationship type.

[0154] In this embodiment, the main knowledge graph includes the root node R1 "Basic Principles of Marxism", the first-level chapter node C3 "Human Society and its Development Laws", and the second-level knowledge point node K31 "Social Existence and Social Consciousness". The course structure position L31 corresponds to node K31, and corresponds to page 12 of the courseware, the first section of Chapter 3 of the textbook, and the classroom audio time period from 00:18:20 to 00:18:35.

[0155] The system transcribes the classroom audio within this time period to obtain audio segment T1: "Social existence determines social consciousness," and extracts the candidate semantic object "Social existence determines social consciousness." The system then binds this candidate semantic object with the source segment T1, time position, recognition confidence level, and course structure position L31 to form the audio source object VO1.

[0156] Since VO1 has a complete theoretical expression, the system may determine its original role as a new first-level node "Social Existence Determines Social Consciousness" and attach it to the root node R1. At the same time, based on the source fragment T1, the course structure position L31, and the existing second-level knowledge point node K31, the system generates a structurally limited alternative role, that is, VO1 as a local expression, node alias, or classroom fragment annotation under K31.

[0157] The system places VO1 into the shadow graph according to its original proposed role and structurally restricted alternative role. After path alignment, the system finds that the original proposed role additionally triggers a first-level node path attached to the root node R1. The set of allowed parent nodes corresponding to the course structure position L31 of the source fragment does not include the root node R1 as the direct parent node of this candidate object. Therefore, this difference path unit is determined to be a parent attachment conflict. The system determines VO1 to be in a pending verification state and writes it into the pending verification node cache, recording its original proposed role, structurally restricted alternative role, difference path segment, and conflict source. The conflict sources include parent attachment expansion and course affiliation index conflict.

[0158] Subsequently, the system extracts corresponding evidence objects from the initial courseware text and textbook text. The initial courseware material has already been used to determine the course structure position L31, so it is only used to verify the position; the textbook text and the teacher's proofreading text respectively provide K31 as a role support field for the higher-level knowledge point and the correct relational direction. The above evidence can support VO1 as a local expression under K31, but cannot support it as a newly added first-level node. Based on this, the system downgrades VO1 to a node alias, local annotation, or classroom segment index of K31.

[0159] Furthermore, if another speech segment T2 is extracted as a candidate semantic object "social consciousness reacts to social existence," but the initial appositional evidence can only prove the appearance of the two terms "social consciousness" and "social existence," and cannot confirm the direction of the relationship between them, then the system writes the candidate object into the cache of the node to be verified, forming the node to be verified DN2, and keeps it in the state of being to be verified. Even if "social consciousness" and "social existence" are both located in the same course structure position L31, when the original intended relationship direction is "social consciousness - decision - social existence" while the allowed relationship set record direction is "social existence - decision - social consciousness," the path unit is still judged as having a conflicting relationship direction, and is not determined to be writable simply because it does not cross chapters.

[0160] After the teacher proofreads and updates the text, the system obtains a newly added proofreading fragment: "Social existence determines social consciousness, and social consciousness has a reaction effect on social existence." Based on this newly added proofreading fragment, the system generates a new corresponding evidence object EV2. EV2 and DN2 have the same curriculum structure position L31, and their adjacent semantic objects include "social existence" and "social consciousness," and support the relational connection path of "social consciousness—reaction effect on—social existence."

[0161] The system compares the original proposed roles, structurally constrained alternative roles, and differing path segments of EV2 with those of DN2. If the original proposed role of DN2 is indeed the relationship edge establishing "social consciousness—reaction to—social existence," and EV2 does not trigger relationship connection conflicts or course index conflicts again, the system releases DN2 from the node cache to be proven and writes the relationship edge into the main knowledge graph. If the original proposed role of DN2 is the relationship edge establishing "social consciousness—determine—social existence," then EV2 can only prove the existence of relevant terms and the correct relationship direction, but cannot cover the path corresponding to the incorrect original proposed role. The system downgrades DN2 to a negative classroom annotation or a local candidate object, and does not write it into the main knowledge graph as a formal relationship edge.

[0162] Through the above process, this application implements three types of processing within the same course segment: for objects that can be supported by path-consistent relationships, they are written into or merged into the main knowledge graph; for objects that may cause mis-construction of high-level nodes, they are written into the node to be verified cache and downgraded based on corresponding evidence; for nodes to be verified that lack sufficient initial evidence, they are released or downgraded a second time after the corresponding evidence materials are updated. Thus, the system can suppress mis-construction of high-level nodes, incorrect connection of relation edges, and diffusion of course indexes before the speech recognition results officially enter the main knowledge graph, and ensure that nodes to be verified have a clear destination for subsequent processing.

[0163] The technical scope of this invention is not limited to the content described above. Those skilled in the art can make various modifications and variations to the above embodiments without departing from the technical concept of this invention, and all such modifications and variations should fall within the protection scope of this invention.

Claims

1. A method for constructing a knowledge graph for ideological and political education courses based on digitalization, characterized in that, include: Acquire the audio, course structure, main knowledge graph and corresponding evidence materials of ideological and political courses, transcribe the audio of ideological and political courses and extract candidate semantic objects, and encapsulate the candidate semantic objects with source segments, time positions, recognition confidence and course structure positions into audio source objects; Based on the matching relationship between the voice source object and the existing nodes, relation edges and course affiliation index in the main knowledge graph, the original proposed role is determined, and a structurally restricted alternative role is generated based on at least one of the following constraints: local terms in the source segment, aliases of existing nodes or course indexes in the source segment. A shadow graph that does not participate in the formal service is generated from the main knowledge graph. Objects from the same voice source are placed into the original role and the structurally restricted alternative role respectively, forming the original placeholder object and the alternative placeholder object. Track the parent attachment paths, relationship connection paths, and course affiliation index paths triggered by the original placeholder object and the replacement placeholder object, and align and extract the difference path segments according to path level, node, relationship edge, and index unit. The structure propagation status is determined based on the first occurrence position of the difference path segment, the continuous expansion range, and the consistency or conflict between the difference path segment and the existing superior node, relation endpoint, relation type, relation direction, and course affiliation index range corresponding to the course structure position of the source segment. Based on the structural propagation state, the voice source object is written into the main knowledge graph, written into the cache of the node to be verified, or downgraded to a local candidate object. For the node to be verified written into the cache of the node to be verified, the term content of the node to be verified and the superior node, relation endpoint, relation type, relation direction or course affiliation index range corresponding to the original role are verified based on the corresponding evidence materials. The node to be verified is released, maintained or downgraded according to the verification results.

2. The method for constructing a knowledge graph for ideological and political education courses based on digitalization according to claim 1, characterized in that, The course structure location is determined by at least one of the course chapters, courseware pages, teaching paragraphs, or course index units corresponding to the time location, and limits the attachment and indexing range when the voice source object enters the main knowledge graph; Adjacent semantic objects are those located in the same source segment as candidate semantic objects and that form a relationship of succession, modification, parallel explanation, causal explanation, inclusion explanation, or attribution explanation with candidate semantic objects in terms of transcribing text order, semantic dependency relationship, or course structure attribution relationship.

3. The method for constructing a knowledge graph for ideological and political education courses based on digitalization according to claim 1, characterized in that, The original proposed roles include: The word content of the speech source object and the adjacent semantic objects are matched with the existing node names, node aliases, relation edge types and course affiliation indexes in the main knowledge graph, respectively. When the matching result points to a newly added concept, relationship connection, or index affiliation, the corresponding newly added node role, newly added relationship edge role, or newly added course index role will be determined as the original proposed role.

4. The method for constructing a knowledge graph for ideological and political education courses based on digitalization according to claim 1, characterized in that, Generate structurally constrained alternative roles include: Extract local terms and adjacent semantic objects that are adjacent to the candidate semantic objects within the source fragment, and match them with existing node aliases, nodes in the same chapter, or course indexes within the source fragment in the main knowledge graph; When a corresponding object exists, the voice source object is limited to an existing node alias, local annotation, candidate object in the same chapter, or index object within a segment, in order to form a structurally restricted substitute role.

5. The method for constructing a knowledge graph for ideological and political education courses based on digitalization according to claim 1, characterized in that, The shadow graph is copied from the main knowledge graph and does not participate in course retrieval, learning path recommendation, teaching resource access, or official graph updates; The original placeholder objects are assigned corresponding nodes, relationship edges, or course affiliation indexes in the shadow graph according to their original roles. The replacement placeholder objects are assigned corresponding aliases, local annotations, candidate objects in the same chapter, or index objects within fragments in the shadow graph according to their structurally restricted replacement roles.

6. The method for constructing a knowledge graph for ideological and political education courses based on digitalization according to claim 1, characterized in that, The extracted difference path segments include: Starting from the original intended placeholder and the replacement placeholder, trace to their parent node, adjacent edges, reachable nodes and course index units respectively; Align the two types of tracking results according to path hierarchy, node identity, relation edge type, and index unit; The portion of the path that is triggered only by the original placeholder object and not by the replacement placeholder object is identified as a differential path segment.

7. The method for constructing a knowledge graph for ideological and political education courses based on digitalization according to claim 1, characterized in that, Determining the propagation state of the structure includes: Based on the existing nodes, relation edges, and course affiliation indexes corresponding to the course structure position of the source fragment in the main knowledge graph, the allowed set of parent nodes, allowed set of relations, and allowed set of indexes are determined respectively. The allowed set of relations includes relation endpoints, relation types, and relation directions. The judgment is made in the order of pending verification status, downgraded status, and writable status; When any path unit in the difference path segment does not belong to the corresponding set of allowed superior nodes, set of allowed relations, or set of allowed indexes, the speech source object is determined to be in a state to be verified. In cases where the original proposed role cannot be supported by the allowed set of parent nodes, allowed set of relations, or allowed set of indexes, but the candidate semantic object can correspond to an existing node alias, local annotation, or intra-segment index, the speech source object will be determined to be in a downgraded state. In the absence of a pending certification state or a downgraded state, when the path units corresponding to the original proposed role all belong to the allowed set of superior nodes, the allowed set of relations, and the allowed set of indexes, the voice source object is determined to be writable. When the correspondence between the path unit in the difference path segment and the allowed set of superior nodes, allowed set of relations, or allowed set of indexes is insufficient for determination, the speech source object is determined to be in a state to be verified.

8. The method for constructing a knowledge graph of ideological and political education courses based on digitalization according to claim 7, characterized in that, When writing a voice source object in the pending verification state into the pending verification node cache, its source segment, time position, course structure position, original role, structurally limited substitute role, difference path segment, and conflict source that triggered the pending verification state are all written into the pending verification node cache. The sources of conflict include at least one of the following: superior attachment extension, relational connection extension, or course affiliation index conflict corresponding to the difference path segment.

9. The method for constructing a knowledge graph of ideological and political education courses based on digitalization according to claim 8, characterized in that, Based on the source of the conflict, a risk propagation review process is conducted on subsequent audio source objects, including: When the subsequent voice source object and the node to be verified in the cache of the node to be verified are in the same course structure position, and there is a corresponding relationship between their adjacent semantic objects or original roles, the superior attachment path, relationship connection path or course affiliation index path corresponding to the conflict source is marked as a path to be reviewed. After the original role is generated for the subsequent voice source object, the subsequent voice source object is still placed into the shadow graph according to the original role and the structurally restricted replacement role; only when the original placeholder object triggers a conflict of the same type as the conflict source again along the path to be reviewed, the subsequent voice source object is written into the node to be verified cache and prevented from being directly written into the main knowledge graph along the path to be reviewed. If subsequent voice source objects do not trigger the same type of conflict again, or if the corresponding evidence supports their original role, the writing of subsequent voice source objects is not restricted, and the corresponding pending review mark is canceled.

10. The method for constructing a knowledge graph of ideological and political education courses based on digitalization according to claim 8, characterized in that, A secondary release process is performed on nodes that remain in the pending verification state, including: After the peer evidence materials are updated, new peer evidence objects with the same course structure position or adjacent semantic objects as the node to be proven are extracted. The newly added objects of equivalent evidence will be compared with the original proposed roles, structurally restricted alternative roles, and different path segments of the node to be proven. When a newly added co-existing evidence object only contains the term content of the node to be proven, but does not simultaneously contain the superior node and level corresponding to the new node role, the relation endpoint corresponding to the relation edge role, the relation type and relation direction, or the index unit and index range corresponding to the course index role, the node to be proven will be downgraded to an existing node alias, local annotation, or fragment index object. When a newly added co-existing evidence object contains a superior node and level, relation endpoint, relation type and relation direction, or index unit and index range corresponding to the original intended role type, and no conflict source is triggered again, the node to be proven is released and written into the main knowledge graph according to the original intended role. When newly added co-defendant evidence still conflicts with or is insufficient to support the original role, the case remains pending verification and the source of the new conflict or insufficient evidence is recorded.

Images