CN-122019614-A - Knowledge-graph-based education resource semantic retrieval method

CN122019614ACN 122019614 ACN122019614 ACN 122019614ACN-122019614-A

Abstract

The invention relates to the technical field of computer information processing, and provides a knowledge-graph-based education resource semantic retrieval method. The method comprises the steps of carrying out structural processing on course teaching materials, courseware, multimedia resources and experimental videos to construct a science education knowledge graph, introducing a sequential hypothesis test and e-process evidence accumulation mechanism in the knowledge graph construction process, carrying out statistical significance control on entity identification results, improving stability and reliability of knowledge modeling, constructing a relational strength perception sub-graph neural network model, obtaining resource structure semantic representation conforming to teaching logic, fusing the resource structure semantic representation with text semantic representation, and realizing semantic retrieval and sequencing of knowledge graph enhancement. The method can output the search results of teaching logic coherence and adaptation of different teaching scenes, and improves the accuracy of the education resource search and the teaching auxiliary effect.

Inventors

LI YAMEI
LU JIJIAN
LIU KAI

Assignees

云南师范大学

Dates

Publication Date: 20260512
Application Date: 20260331

Claims (7)

1. The knowledge graph-based education resource semantic retrieval method is characterized by comprising the following steps of: Step S1, acquiring course education resources, and performing word segmentation, part-of-speech tagging and named entity recognition on titles, labels, text descriptions and subtitle information of the course education resources to obtain structured text data; S2, adopting BiLSTM-CRF model to make named entity identification on the structured text data, introducing sequential hypothesis test and e-process mechanism in the entity identification process, making sequential statistical test to define entity identification result; Step 3, adopting a text coding model to code the title, text description and caption text of a target educational resource in the course educational resource into text semantic vectors, and adopting a relationship strength perception subgraph GNN model to carry out graph embedding calculation based on knowledge point entities, experimental entities and chapter entities associated with the target educational resource in a theoretical education knowledge graph to obtain graph structure embedded vectors; Step S4, the teacher terminal and the student terminal submit a search request through a search interface to generate a semantic vector representation of a search sentence; Step S5, calculating a comprehensive relevance score based on the semantic vector representation of the search statement and the semantic search representation of the educational resource; And S6, selecting educational resources from high to low according to the comprehensive relevance score, and displaying the resources marked as dynamic courseware, dynamic demonstration video and experimental demonstration video in a search result interface of the teacher terminal and the student terminal preferentially.
2. The knowledge graph-based educational resource semantic retrieval method according to claim 1, wherein the relationship strength perception subgraph GNN model is constructed by constructing the GNN model, introducing a structured weight computing mechanism of relationship strength perception to improve the neighborhood information propagation and aggregation process of the GNN model, and constructing the relationship strength perception subgraph GNN model.
3. The educational resource semantic retrieval method based on the knowledge graph according to claim 1, wherein the step S2 specifically comprises the following steps: s21, performing unified coding and splicing on structured fields of structured text data to establish a sequence sample set; S22, inputting each sequence sample in the sequence sample set into a BiLSTM-CRF model to obtain a label score matrix and a decoded label sequence, analyzing fragments formed by continuous labels of the same type to generate candidate entity fragments, and constructing an entity candidate stream; Step S23, constructing a sequential hypothesis test pair for each candidate entity segment in the entity candidate stream, and carrying out statistics and judgment on the entity effectiveness of the candidate entity segment; Step S24, constructing an e-process statistical evidence accumulation process for each candidate entity fragment in the entity candidate stream, and initializing the e-process statistical evidence accumulation process, updating the e-process statistical evidence accumulation process according to a sequential recurrence rule meeting the statistical constraint under the zero assumption condition based on single-step statistical evidence increment, and continuously acquiring an e-process statistical evidence accumulation value corresponding to the candidate entity fragment; step S25, setting a statistical significance level, constructing a sequential rejection boundary based on the significance level, continuously monitoring e-process statistical evidence accumulation values for each candidate entity segment in the entity candidate stream, and executing dynamic stop judgment according to a first boundary crossing rule to acquire a statistical confirmation state of the candidate entity segment and a corresponding evidence index; Step S26, when the candidate entity fragments in the entity candidate stream collide on the entity boundary based on the statistical confirmation state of the candidate entity fragments and the corresponding evidence indexes, a collision resolution mechanism based on the sequential statistical characteristics is introduced to perform unified resolution, entity confirmation and freezing processing are performed on the candidate entity fragments selected by the collision resolution mechanism, evidence suppression processing is performed on the unselected candidate entity fragments after the collision resolution is completed, and an entity recognition result is formed through the parallel sequential verification and collision resolution mechanism, so that the entity in the science field is automatically extracted.
4. The educational resource semantic retrieval method based on knowledge graph according to claim 3, wherein the sequential hypothesis test pair comprises 1. Zero hypothesis that the candidate entity segment does not satisfy the judging condition of the target entity, 2. Alternative hypothesis that the candidate entity segment satisfies the judging condition of the target entity.
5. The method for semantic retrieval of educational resources based on knowledge graph according to claim 3, wherein the e-process statistical evidence accumulation process satisfies the constraint of martingale under the condition that zero assumption is satisfied, in particular, the condition expectation of the statistical evidence accumulation process under the condition of zero assumption does not rise with the increase of observation rounds.
6. The educational resource semantic retrieval method based on knowledge graph according to claim 3, wherein the conflict resolution mechanism comprises the following judging rules of 1. Out-of-range moment priority rule and 2. Evidence intensity priority rule.
7. The knowledge-graph-based educational resource semantic retrieval method according to claim 2, wherein the process of obtaining the graph structure embedded vector by performing graph embedding calculation by adopting a relationship strength perception subgraph GNN model comprises the following steps: s31, constructing a resource association subgraph based on a knowledge point entity, an experimental entity and a chapter entity which have association relation with a target education resource in a science education knowledge map; Step S32, calculating a relation strength weight value of each relation edge through a relation strength aware structured weight calculation mechanism based on semantic type identifications corresponding to different types of relation edges in a resource association subgraph, co-occurrence frequency of association entities in teaching resources, hierarchical distance of the association entities in a course chapter structure and corresponding compactness information between an experimental entity and a knowledge point entity, and taking the relation strength weight value as a structured modulation parameter in a neighborhood information aggregation process; Step S33, on the resource association subgraph, executing multi-round neighborhood information propagation and aggregation for each node based on a node representation updating mechanism of the GNN model; and step S34, generating a graph structure embedded vector after completing resource-related subgraph embedding propagation calculation.

Description

Knowledge-graph-based education resource semantic retrieval method Technical Field The invention relates to the technical field of computer information processing, in particular to a knowledge-graph-based education resource semantic retrieval method. Background With the application of artificial intelligence, natural language processing and deep learning technologies in the field of education informatization, the conventional educational resource retrieval system gradually introduces semantic vector representation, a pre-training language model and an intelligent recommendation algorithm so as to improve the automatic retrieval and recommendation capability of educational resources and solve the problem of insufficient semantic understanding capability of the traditional keyword matching method to a certain extent. However, most of the existing intelligent retrieval technologies still use text semantic similarity as a core, and it is difficult to effectively express the objective first-repair dependence, theoretical derivation and structural relationship between knowledge points and experimental processes in a science knowledge system, so that the retrieval results have defects in teaching logic consistency and knowledge evolution integrity. Meanwhile, an educational system with a knowledge graph introduced is usually subjected to entity identification and knowledge modeling by depending on static rules or fixed confidence thresholds, lacks of statistical control on model prediction uncertainty and sequence context, is easily influenced by noise or local fluctuation, influences the structural stability of the knowledge graph, and further weakens semantic retrieval and resource sequencing effects. In addition, the existing system lacks effective modeling for importance differences of different teaching relations in the resource sequencing and result display stage, and is difficult to consider comprehensive requirements for dynamic demonstration, experimental process and knowledge expansion in classroom teaching and personalized learning scenes, and the whole teaching auxiliary effect still has a lifting space. Disclosure of Invention Aiming at the problems that the existing intelligent educational resource retrieval technology has insufficient knowledge structure modeling capability and poor entity identification stability in a science teaching scene and the retrieval result is difficult to embody teaching logic, the invention provides an educational resource semantic retrieval method based on a knowledge graph, which realizes high-quality semantic retrieval oriented to teaching logic by fusing natural language semantic modeling with structural relation modeling of a science knowledge system; the method is characterized by comprising the steps of introducing a sequential hypothesis test and e-process evidence accumulation mechanism with statistical significance assurance in a theoretical science education knowledge graph construction process, carrying out dynamic statistical discrimination on entity prediction confidence, improving the stability of entity identification and knowledge modeling on the premise of not depending on a fixed threshold value or sample length hypothesis, simultaneously constructing a sub-graph neural network model with relationship strength perception aiming at importance differences of teaching relationships such as first-repair dependence, theoretical derivation and experimental correspondence, so that graph embedding can highlight knowledge relationships with higher contribution to teaching semantics, and on the basis, fusing text semantic representation and graph structure embedding, and combining a knowledge graph enhancement retrieval and sequencing mechanism, so that a retrieval result is driven to be converted from single semantic similarity to a semantic retrieval result reflecting the knowledge structure and teaching logic at the same time, thereby providing technical support for dynamic visual auxiliary teaching and personalized learning. The invention provides a knowledge graph-based education resource semantic retrieval method, which is applied to a teaching auxiliary learning scene, wherein the scene comprises a semantic retrieval server, a knowledge graph database server and an education resource storage server which are deployed in an education cloud data center, and teacher terminals and student terminals which are arranged in a science laboratory, a multimedia classroom and a student self-learning area, wherein the teacher terminals and the student terminals are in communication connection with the semantic retrieval server through a campus network, and the method is executed by the semantic retrieval server and comprises the following steps: step S1, acquiring course education resources from an education resource storage server, and performing word segmentation, part-of-speech tagging and named entity recognition on titles, labels, text descriptions and subtitle in