Education knowledge graph construction method, related device, equipment and storage medium

Abstract

The invention discloses an educational knowledge graph construction method, a related device, equipment and a storage medium, and the construction method comprises the steps: carrying out the feature extraction based on a first educational graph, obtaining the first feature representation of each node in the first educational graph, carrying out the feature extraction based on a teaching behavior sequence of a current time sequence, and obtaining the first feature representation of each node in the first educational graph; obtaining a second feature representation of each knowledge point involved in the teaching behavior sequence; selecting knowledge points which are involved in the teaching behavior sequence and are not involved in the first education map as target knowledge points; determining an updating mode of the target knowledge point based on the similarity between the second feature representation of the target knowledge point and each first feature representation; and updating the target knowledge point to the first education map according to the updating mode of the target knowledge point to obtain a second education map which is self-evolved in the teaching practice. According to the scheme, the multi-level requirements of the educational knowledge graph can be met, and the self-evolution of the educational knowledge graph is realized, so that the updating efficiency of the educational knowledge graph is improved.

Description

Technical Field

[0001] The present application relates to the field of smart education technology, and in particular to a method for constructing an educational knowledge graph and related devices, equipment and storage media. Background Art

[0002] With the continuous development of scientific education, teaching content has become interdisciplinary, multi-level and dynamic, and the intersection and connection of knowledge between different disciplines has become increasingly complex.

[0003] However, when constructing educational knowledge graphs, they are usually mainly oriented towards the subject knowledge level, which cannot meet the multi-level needs of educational knowledge graphs and may lead to a relatively slow iterative update of educational knowledge graphs. In view of this, how to meet the multi-level needs of educational knowledge graphs and realize the self-evolution of educational knowledge graphs to improve the update efficiency of educational knowledge graphs has become an urgent problem to be solved. Summary of the invention

[0004] The main technical problem solved by this application is to provide a method for constructing an educational knowledge graph and related devices, equipment and storage media, which can meet the multi-level needs of the educational knowledge graph and realize the self-evolution of the educational knowledge graph to improve the updating efficiency of the educational knowledge graph.

[0005] In order to solve the above-mentioned technical problems, the first aspect of the present application provides a method for constructing an educational knowledge graph, comprising: performing feature extraction based on a first educational graph to obtain a first feature representation of each node in the first educational graph, and performing feature extraction based on a current time series teaching behavior sequence to obtain a second feature representation of each knowledge point involved in the teaching behavior sequence; wherein the node type to which each node belongs includes: at least one of curriculum standard knowledge points, cutting-edge hot knowledge points, and teaching scenario knowledge points; selecting knowledge points involved in the teaching behavior sequence and not involved in the first educational graph as target knowledge points; determining an update method for the target knowledge points based on the similarity between the second feature representations of the target knowledge points and each first feature representation; updating the target knowledge points to the first educational graph according to the update method of the target knowledge points, and obtaining a second educational graph that self-evolves in teaching practice.

[0006] In order to solve the above-mentioned technical problems, the second aspect of the present application provides a device for constructing an educational knowledge graph, including: a feature extraction module, a knowledge point selection module, an update method determination module and a knowledge graph update module, the feature extraction module is used to perform feature extraction based on a first education graph to obtain a first feature representation of each node in the first education graph, and to perform feature extraction based on a current time series teaching behavior sequence to obtain a second feature representation of each knowledge point involved in the teaching behavior sequence; wherein the node type to which each node belongs includes: at least one of curriculum standard knowledge points, cutting-edge hot knowledge points, and teaching scenario knowledge points; the knowledge point selection module is used to select knowledge points involved in the teaching behavior sequence and not involved in the first education graph as target knowledge points; the update method determination module is used to determine the update method of the target knowledge point based on the similarity between the second feature representation of the target knowledge point and each first feature representation; the knowledge graph update module is used to update the target knowledge point to the first education graph according to the update method of the target knowledge point, and obtain a second education graph that self-evolves in teaching practice.

[0007] In order to solve the above-mentioned technical problems, the third aspect of the present application provides an electronic device, which at least includes a memory and a processor coupled to each other, wherein the memory at least stores program instructions, and the processor is used to execute the program instructions to implement the method for constructing an educational knowledge graph in the above-mentioned first aspect.

[0008] In order to solve the above-mentioned technical problems, the fourth aspect of the present application provides a computer-readable storage medium storing program instructions that can be executed by a processor, and the program instructions are used to implement the method for constructing an educational knowledge graph of the first aspect mentioned above.

[0009] The above scheme performs feature extraction based on the first education map to obtain the first feature representation of each node in the first education map, and performs feature extraction based on the current time series teaching behavior sequence to obtain the second feature representation of each knowledge point involved in the teaching behavior sequence, and the node type to which each node belongs includes: curriculum standard knowledge points, cutting-edge hot knowledge points, and teaching scenario knowledge points. Then, the knowledge points involved in the teaching behavior sequence and not involved in the first education map are selected as target knowledge points, so as to determine the update method of the target knowledge points based on the similarity between the second feature representation of the target knowledge points and the first feature representation, and then according to the update of the target knowledge points The target knowledge points are updated to the first education map in this way, and the second education map that self-evolves in teaching practice is obtained. Therefore, on the one hand, in addition to the curriculum standards, cutting-edge hot spots, teaching scenarios and other levels related to subject knowledge, the knowledge points involved in the teaching behavior level are further perceived, which can meet the multi-level needs of the education knowledge map. On the other hand, by extracting the feature representation of each knowledge point involved in the teaching behavior sequence and measuring its similarity with the feature representation of each node in the knowledge map, the relevant knowledge points involved in the teaching behavior sequence are updated to the knowledge map, which can realize the self-evolution of the education knowledge map in teaching practice, and help improve the updating efficiency of the education knowledge map. Therefore, it can meet the multi-level needs of the education knowledge map and realize the self-evolution of the education knowledge map to improve the updating efficiency of the education knowledge map. BRIEF DESCRIPTION OF THE DRAWINGS

[0010] Figure 1 It is a flowchart of an embodiment of a method for constructing an educational knowledge graph of the present application; Figure 2a It is a process diagram of an embodiment of a method for constructing an educational knowledge graph of the present application; Figure 2b It is a process diagram of another embodiment of the method for constructing an educational knowledge graph of the present application; Figure 3 It is a schematic diagram of the framework of an embodiment of a device for constructing an educational knowledge graph of the present application; Figure 4 It is a schematic diagram of the framework of an embodiment of the electronic device of the present application; Figure 5 It is a schematic diagram of a framework of an embodiment of a computer-readable storage medium of the present application. DETAILED DESCRIPTION

[0011] The scheme of the embodiment of the present application is described in detail below in conjunction with the drawings of the specification.

[0012] In the following description, for the purpose of explanation rather than limitation, specific details such as specific system structures, interfaces, and technologies are provided to facilitate a thorough understanding of the present application.

[0013] The terms "system" and "network" are often used interchangeably in this article. The term "and / or" in this article is only a description of the association relationship of associated objects, indicating that there can be three relationships. For example, A and / or B can mean: A exists alone, A and B exist at the same time, and B exists alone. In addition, the fragment " / " in this article generally indicates that the associated objects before and after are in an "or" relationship. In addition, "many" in this article means two or more than two.

[0014] See also Figure 1 , Figure 1 This is a flow chart of an embodiment of a method for constructing an educational knowledge graph of this application. Figure 2a and Figure 2b , Figure 2a It is a process diagram of an embodiment of a method for constructing an educational knowledge graph of the present application. Figure 2b This is a schematic diagram of another embodiment of the method for constructing an educational knowledge graph of the present application. The disclosed embodiment may specifically include the following steps: Step S11: Perform feature extraction based on the first education map to obtain a first feature representation of each node in the first education map, and perform feature extraction based on the current time series teaching behavior sequence to obtain a second feature representation of each knowledge point involved in the teaching behavior sequence.

[0015] In the disclosed embodiment, the node types to which each node belongs include: at least one of curriculum standard knowledge points, cutting-edge hot knowledge points, and teaching scenario knowledge points. It should be noted that curriculum standard knowledge points can come from curriculum standard documents (such as textbooks, etc.), cutting-edge hot knowledge points can come from literature databases (such as extended reading materials related to curriculum standards, etc.), and teaching scenario knowledge points can come from teaching classroom data (such as classroom lecture recordings, videos, etc.). In addition, in the actual application process, according to the actual application needs, the node types to which each node in the first education atlas belongs can specifically only include curriculum standard knowledge points, or can only include cutting-edge hot knowledge points, or can only include teaching scenario knowledge points, or can include at least two of the above at the same time, and no examples will be given one by one here.

[0016] In an implementation scenario, as a possible example in the actual application process, the node types to which each node belongs include: curriculum standard knowledge points, cutting-edge hot knowledge points and teaching scenario knowledge points. In order to construct the first education map, the curriculum standard knowledge points in the curriculum standard document can be specifically mapped to the existing education map to obtain the third education map, and then the cutting-edge hot knowledge points can be predicted and mapped to the third education map based on the literature corpus to obtain the fourth education map, and the teaching scenario knowledge points can be modeled and mapped to the fourth education map based on classroom teaching data to obtain the first education map. The above method can gradually enrich the existing education map by sequentially mapping the curriculum standard knowledge points in the curriculum standard document, predicting and mapping the cutting-edge hot knowledge points based on the literature corpus, and modeling and mapping the teaching scenario knowledge points based on classroom teaching data to the existing education map, which helps to improve the richness of the first education map finally constructed.

[0017] In a specific implementation scenario, in order to map the curriculum standard knowledge points in the curriculum standard document to the existing education map, it is possible to perform feature extraction based on each curriculum standard text in the curriculum standard document, obtain the third feature representation of each curriculum standard text, and then map the curriculum standard knowledge points representing the curriculum standard text to the existing education map based on the similarity between the third feature representation of the curriculum standard text and the node feature representation of each node in the existing education map to obtain the third education map. For example, the curriculum standard document can be pre-annotated, such as marking the curriculum standard text of 353 knowledge points in the compulsory stage, the curriculum standard text of 1201 knowledge points in the high school stage, etc. Of course, the above example is only a possible example of the curriculum standard text, and the method of obtaining the curriculum standard text and the specific number are not limited here. Exemplarily, for any curriculum standard text, if the similarity between its third feature representation and the node feature representation of a node in the existing education map is higher than a preset threshold, it can be considered that the curriculum standard text has a strong semantic association with the knowledge point represented by this node, and the curriculum standard text can be merged into this node, that is, at this time, the new attribute information of this node not only contains the original attribute information, but also contains the curriculum standard text. On the contrary, if the similarity between its third feature representation and the node feature representation of any node in the existing education map is not higher than the preset threshold, it can be considered that the curriculum standard text has no strong semantic association with the knowledge point represented by any node in the existing education map, and the new node represented by the curriculum standard text can be added to the existing education map. Of course, in order to further improve the accuracy of mapping the curriculum standard knowledge points in the curriculum standard document to the existing education map, after performing the above operation, the above mapping operation can also be verified in combination with manual intervention and correction mechanism, and can be corrected when necessary, thereby realizing human-computer collaborative knowledge extraction. For ease of description, the curriculum standard text can be denoted as T, and its third feature representation can be embedded through a pre-trained language model such as BERT (Bidirectional Encoder Representation from Transformers) to convert the text into a high-dimensional semantic vector as the third feature representation V: …… (1) In addition, the node feature representation of each node in the existing knowledge graph can be obtained by extracting features from the existing knowledge graph through methods such as GNN (Graph Neural Network). The specific process can be found in the technical details of GNN, which will not be repeated here. In the above method, feature extraction is performed on each curriculum standard text in the curriculum standard document to obtain the third feature representation of each curriculum standard text, and then based on the similarity between the third feature representation of the curriculum standard text and the node feature representation of each node in the existing education graph, the curriculum standard knowledge points representing the curriculum standard text are mapped to the existing education graph to obtain the third education graph, which can map the curriculum standard knowledge points representing the curriculum standard text to the existing education graph at the feature level, which helps to improve the mapping accuracy of the curriculum standard text.

[0018] In a specific implementation scenario, after obtaining the third education map, in order to predict and map the cutting-edge hot knowledge points to the third education map based on the literature corpus, it is possible to obtain the predicted word frequency of each knowledge topic in the literature corpus based on the historical word frequency of the knowledge topic at each historical moment, and then select the knowledge topic as a candidate topic based on the predicted word frequency of the knowledge topic, so as to perform topic modeling based on the candidate topic, select the candidate topic as the hot topic, and finally map the cutting-edge hot knowledge points representing the hot topic to the third education map based on the similarity between the fourth feature representation of the hot topic and the node feature representation of each node in the third education map to obtain the fourth education map. As mentioned above, the literature corpus may include but is not limited to extended reading materials related to the curriculum standards. For example, newspapers, magazines, online abstracts, etc., and the specific content of the literature corpus is not limited here. For any knowledge topic, since it has different historical word frequencies at different historical moments, the historical word frequencies at historical moments can be used for prediction, such as the predicted word frequency f(t) of the knowledge topic can be obtained through time series modeling: …… (2) In the above formula (2), a i represents the weight of the ith historical moment, f(ti) represents the historical word frequency of the ith historical moment, Represents a random error term. As a possible example, after obtaining the predicted word frequency of each knowledge topic, the predicted word frequency of the knowledge topic can be regarded as the evolution trend of the knowledge topic, that is, if the predicted word frequency of the knowledge topic is high, it can be considered that the knowledge topic has a strong evolution trend, and vice versa, if the predicted word frequency of the knowledge topic is low, it can be considered that the knowledge topic has a weak evolution trend. Therefore, for any knowledge topic, if the predicted word frequency of the knowledge topic is higher than the word frequency threshold, the knowledge topic can be selected as a candidate topic, and vice versa, if the predicted word frequency of the knowledge topic is not higher than the word frequency threshold, the knowledge topic can be not selected as a candidate topic. After that, topic modeling can be performed on each candidate topic through a model such as Dirichlet Allocation (LDA) to extract potential hot topics. For the specific process, please refer to the technical details of topic modeling techniques such as Dirichlet Allocation, which will not be repeated here. On this basis, we can refer to the relevant description of the aforementioned mapping curriculum standard text, and map the cutting-edge hot knowledge points representing the hot topics to the third education map based on the similarity between the fourth feature representation of the hot topic and the node feature representation of each node in the third education map. That is to say, as a possible example, for any hot topic, if the similarity between its fourth feature representation and the node feature representation of a node in the third education map is higher than a preset threshold, then the hot topic can be merged into this node, that is, at this time, the new attribute information of this node not only contains the original attribute information, but also includes the hot topic. Conversely, if the similarity between its fourth feature representation and the node feature representation of any node in the third education map is not higher than the preset threshold, it can be considered that there is no strong semantic association between the hot topic and the knowledge points represented by any node in the third education map, and the new node represented by the hot topic can be added to the third education map. Exemplarily, the similarity can be represented by cosine similarity as : …… (3) In the above formula (3), V1 represents the fourth feature representation of the hot topic, and V2 represents the node feature representation of a node in the third education map. In the above method, for each knowledge topic in the literature corpus, the predicted word frequency of the knowledge topic is obtained based on the historical word frequency of the knowledge topic at each historical moment, and then the knowledge topic is selected as a candidate topic based on the predicted word frequency of the knowledge topic, so as to perform topic modeling based on the candidate topic, select the candidate topic as the hot topic, and then map the frontier hot knowledge points representing the hot topic to the third education map based on the similarity between the fourth feature representation of the hot topic and the node feature representation of each node in the third education map to obtain the fourth education map, which can extract potential hot topics through time series modeling, and provide support for the expansion of the curriculum standard knowledge system through semantic matching in the feature dimension.

[0019] In a specific implementation scenario, after obtaining the fourth education map, in order to model and map the teaching scene knowledge points to the fourth education map based on the classroom teaching data, a syntactic dependency analysis can be performed based on the classroom teaching data to generate a dependency tree, and the dependency tree contains the various knowledge points involved in the classroom teaching data and their associated logic, and based on the dependency tree, the potential knowledge points in the teaching scene are extracted, and then based on the fifth feature representation of the potential knowledge points and the similarity between the node feature representations of each node in the fourth education map, the teaching scene knowledge points representing the potential knowledge points are mapped to the fourth education map to obtain the first education map. As mentioned above, classroom teaching data can specifically include but is not limited to classroom recording data, classroom video data and other teaching scene related data obtained by speech recognition and other technologies. Based on this, the core semantics can be identified through syntactic analysis and sentiment mining technology. For example, the dependency tree generated by syntactic dependency analysis can be used to model the knowledge points and their logical relationships involved in the teacher's teaching language. As a possible example, it can be defined as: …… (4) In the above formula (4), W represents the word set in the teaching text, and R is the dependency set (for example, a certain knowledge point belongs to another knowledge point, etc.). By analyzing D, the explicit or implicit knowledge point logic in the teaching scene can be extracted, and then based on the similarity between the fifth feature representation of the potential knowledge point and the node feature representation of each node in the fourth education map, the teaching scene knowledge point representing the potential knowledge point can be mapped to the fourth education map to obtain the first education map. The specific process can refer to the relevant description in the aforementioned mapping curriculum standard text or literature database, and will not be repeated here. Through the aforementioned process steps, the first education map containing three different node types of curriculum standard knowledge points, cutting-edge hot knowledge points and teaching scene knowledge points can be modeled. Exemplarily, a triple structure can be used to represent each node in the first education map and the semantic relationship between them. For ease of description, the triple structure It can be expressed as: …… (5) In the above formula (5), h is the head entity (i.e., a node representing a certain knowledge point in the first education graph), t is the tail entity (i.e., another node representing a certain knowledge point in the first education graph), and r is the semantic relationship between the head entity and the tail entity. The above method performs syntactic dependency analysis based on classroom teaching data to generate a dependency tree, and the dependency tree contains each knowledge point involved in the classroom teaching data and its associated logic, and based on the dependency tree, extracts the potential knowledge points in the teaching scene, and then maps the teaching scene knowledge points representing the potential knowledge points to the fourth education graph based on the similarity between the fifth feature representation of the potential knowledge point and the node feature representation of each node in the fourth education graph, so as to obtain the first education graph. It can realize modeling based on classroom teaching data and mapping the teaching scene knowledge points to the fourth education graph through a series of operations such as syntactic dependency and semantic association, which helps to further expand the richness of the first education graph.

[0020] In an implementation scenario, as a possible example, after obtaining the first education map containing three types of nodes, namely, curriculum standard knowledge points, frontier hot knowledge points and teaching scenario knowledge points, entity disambiguation and semantic alignment can be performed on the first education map to achieve accurate mapping of interdisciplinary knowledge points, and association strength analysis can be performed to build association relationships. Specifically, the feature similarity between different nodes in the first education map can be obtained, and based on the feature similarity between different nodes, the corresponding nodes can be selected to merge or retain to perform entity disambiguation and semantic alignment on the first education map. For each node in the updated first education map, the first word frequency, the second word frequency of co-occurrence and the similarity between the context information of the knowledge point represented by the node in any two different subject documents are obtained, and based on the first word frequency, the second word frequency and the similarity, the semantic association strength between the context information of the knowledge point represented by the node in the corresponding two different subject documents is determined. The above method, through entity disambiguation, semantic alignment and construction of association relationships, can realize semantic-rich deep association mining, which is conducive to deep semantic-rich association of interdisciplinary knowledge and improves the accurate mapping and deep semantic association of interdisciplinary knowledge points in scientific education.

[0021] In a specific implementation scenario, due to the differences in knowledge expression and concept definition between different disciplines, the same knowledge point may have multiple semantic representations in different disciplines. For this, an embedding representation method can be used to map cross-disciplinary knowledge points into a unified semantic space. For example, given a set of knowledge points (i.e., the knowledge points represented by each node in the first education graph), they can be mapped into high-dimensional vectors using pre-trained language models such as BERT and RoBERTa: …… (6) In the above formula (6), Represents the feature vector of the i-th knowledge point (i.e., the i-th node in the first education graph). On this basis, the feature similarity of different nodes in the first education graph can be measured by, for example, cosine similarity: …… (7) In the above formula (7), Respectively represent the feature vectors of the i-th node and the j-th node in the first education graph, Indicates the feature similarity between the i-th node and the j-th node in the first education graph. In addition, in order to improve the accuracy of vector mapping, the aforementioned pre-trained language model can also be trained by comparative learning through the triplet loss function (Triplet Loss) to minimize the intra-class spacing (i.e., minimize the feature similarity between the same knowledge points) and maximize the inter-class spacing (i.e., maximize the feature similarity between different knowledge points). Exemplarily, the triplet loss function L can be expressed as: …… (8) In the above formula (8), represents the i-th positive sample knowledge point, represents the i-th negative sample knowledge point, Represents the jth negative sample knowledge point. For details, please refer to the technical details of the triplet loss function, which will not be repeated here. By optimizing the above triplet loss function L, the semantic distinction ability of interdisciplinary knowledge points can be effectively enhanced, which is helpful to achieve high-precision entity disambiguation. As a possible example, after obtaining the feature similarity, if the feature similarity between two nodes in the first education graph is large (e.g., higher than the set threshold), the two nodes can be merged. Conversely, if the feature similarity between two nodes in the first education graph is small (e.g., not higher than the set threshold), the two nodes can be retained to achieve entity disambiguation, coreference resolution and semantic alignment.

[0022] In a specific implementation scenario, for each node in the first education graph after the above process is updated, the context information of the represented knowledge point in different subject documents can be obtained. ,in, represents the context information of the knowledge point represented by the node in the i-th subject document. Then the semantic association strength between the context information in any two different subject documents can be expressed as: …… (9) In the above formula (9), Indicates the first word frequency of the context information of the knowledge point represented by the node in the i-th subject literature, Indicates the first word frequency of the context information of the knowledge point represented by the node in the jth subject literature, The second word frequency of the knowledge point represented by the node in the context of the i-th subject document and the context of the j-th subject document, It represents the similarity between the context information of the knowledge point represented by the node in the i-th subject document and the context information in the j-th subject document. It can be seen that for the semantic association strength between the context information in any two different subject documents, it is negatively correlated with the first word frequency of each, and positively correlated with the co-occurring second word frequency, and positively correlated with the similarity between them. Of course, the above example is only one possible calculation method for the semantic association strength, and it does not limit the possible calculation methods. Other possible calculation methods for the semantic association strength will not be given examples one by one here.

[0023] In an implementation scenario, as a possible example, after obtaining the first education map containing three types of nodes, namely, curriculum standard knowledge points, cutting-edge hotspot knowledge points, and teaching scenario knowledge points, or after the aforementioned entity disambiguation, coreference resolution, and semantic alignment, the node feature representation of each node in the first education map can be obtained first, and then in response to the update of a new node in the first education map, the node originally connected to the new node in the first education map is selected as the node to be updated, and the node feature representation of the new node is obtained, and the node feature representation of the node to be updated and the node feature representation of the new node are weighted to obtain the updated node feature representation of the node to be updated. The above method can realize the graph update mechanism of the time series by fusing the feature representation of the original node in the education map with the feature representation of the new node, so that the knowledge structure can continuously integrate cutting-edge hotspot knowledge, realize the dynamic evolution of the scientific education knowledge network, and provide timely knowledge support for scientific education.

[0024] In a specific implementation scenario, in order to further realize the dynamic update and deep integration of interdisciplinary knowledge points, we can use graph neural networks to extract features from each node in the first education graph. For ease of description, the first education graph can be recorded as , where V represents the set of interdisciplinary knowledge points (i.e., the set of nodes in the first education graph), and E represents the semantic association between knowledge points (i.e., the set of node relationships between nodes in the first education graph). The graph neural network can be used to update the features of node v in the first education graph, so that the embedded representation of each knowledge point can be integrated into the neighborhood information. The update formula is as follows: …… (10) In the above formula (10), is the updated feature of node v (i.e., node feature representation), is the initial feature of the node, Used to extract feature information of neighbor node u, is the feature fusion function. Through this process, the representation of the knowledge point v can dynamically fuse the information and semantic relationship of its associated nodes to obtain the node feature representation of each node in the first education graph.

[0025] In a specific implementation scenario, for newly added knowledge points and associated edges, the knowledge graph can be dynamically expanded. , and learn representations based on a time-aware node embedding update formula: …… (11) In the above formula (11), represents the feature representation of node v at time t (i.e., the node feature representation after the node to be updated is updated), represents the feature representation of node v at the previous time step (i.e., t-1) (i.e., the node feature representation of the node to be updated), The node feature representation of the newly added node, represents the smoothing parameter. In this example, and That is, they are the weight factors of the nodes to be updated and the weight factors of the newly added nodes respectively.

[0026] In one implementation scenario, after obtaining the latest first education graph, feature extraction can be performed on it to obtain the first feature representation of each node in the first education graph. Specifically, each node can be represented as a high-dimensional vector through embedding technology: …… (12) In the above formula (12), represents the feature dimension of the embedding vector, Represents the first education map, represents the total number of nodes in the first education graph, Represents the first feature representation of each node in the first education graph. It should be noted that the first feature representation of each node in the first education graph can be obtained by extracting features from the first education graph, such as through a graph neural network, and the specific method of extracting the first feature representation is not limited here.

[0027] In an implementation scenario, the teaching behavior sequence may include several teaching behaviors, and the teaching behavior may define the knowledge points involved in the teaching behavior and the operation type of the teaching behavior. Based on this, a time series analysis may be performed based on the definition information of each teaching behavior in the teaching behavior sequence, a dynamic relationship diagram between each knowledge point involved in the teaching behavior sequence may be constructed, and then feature extraction may be performed based on the dynamic relationship diagram to obtain a second feature representation of each knowledge point involved in the teaching behavior sequence. The above method, by performing a series of operations such as time series analysis, relationship diagram construction, and feature extraction on the teaching behavior sequence to extract the second feature representation of each knowledge point involved in the teaching behavior sequence, can capture the dynamic relationship pattern between the knowledge point and the teaching behavior, and help improve the accuracy of the second feature representation.

[0028] In a specific implementation scenario, for ease of description, the teaching behavior sequence can be recorded as , where each teaching behavior It can be expressed as a set of knowledge points involved and their operational relationships: …… (13) In the above formula (13), Indicates teaching behavior The operation type (such as import, expand, associate, etc.), Indicates teaching behavior On this basis, a dynamic relationship diagram between knowledge points can be established through time series analysis of teaching behavior sequences. ,in, represents the set of knowledge points in time series t, is the set of relations in time series t. It should be noted that the dynamic relationship diagram For the establishment process, please refer to the technical details of time series analysis, which will not be repeated here.

[0029] In a specific implementation scenario, in order to perceive the dynamic evolution of the knowledge graph, the evolution of the dynamic relationship graph can be modeled through feature extraction models such as Temporal Graph Neural Networks (TGNN). For example, the adjacency matrix of the dynamic relationship graph can be used and the node embedding matrix , the update formula can be expressed as: …… (14) In the above formula (14), represents the embedding of the TGNN layer l timing graph at time t, is the weight parameter of the lth layer of TGNN, is a nonlinear activation function. In addition, when TGNN refers to the above formula (11) and repeats until it is processed by the last layer, the second feature representation of each node in the winter relationship graph can be output. In other words, by introducing time dependency, the above model can capture the dynamic change pattern between knowledge points and teaching behaviors and generate a knowledge graph embedding representation that evolves over time.

[0030] Step S12: Select knowledge points that are involved in the teaching behavior sequence and not involved in the first education map as target knowledge points.

[0031] Specifically, each knowledge point involved in the teaching behavior sequence can be detected separately to detect whether the first education map also involves the knowledge point. If the first education map does not involve it, it can be determined that the knowledge point is a newly added knowledge point, so it can be used as a target knowledge point. On the contrary, if the first education map involves it, it can be determined that the knowledge point is not a newly added knowledge point and cannot be used as a newly added knowledge point. In addition, as mentioned above, the second feature representation is extracted from the dynamic relationship graph between the various knowledge points involved in the teaching behavior sequence. As a possible example, in addition to detecting newly added knowledge points, it is also possible to detect whether there is a node relationship in the dynamic relationship graph that does not exist in the first education map. If it exists, it can be determined that the node relationship is a newly added relationship, so it can be used as a target relationship.

[0032] Step S13: Determine the update method of the target knowledge point based on the similarity between the second feature representation of the target knowledge point and each first feature representation.

[0033] Specifically, the correlation between the target knowledge point and the first education map can be obtained based on the similarity between the second feature representation of the target knowledge point and each first feature representation, and then the update method of the target knowledge point can be determined based on the size relationship between the correlation and the preset threshold. It should be noted that the similarity between the first feature representation and the second feature representation can be obtained by a measurement method such as cosine similarity, which is not limited here. In addition, the update method may include: adding a new node representing the target knowledge point to the node in the first education map, and adding a new node in the first education map to represent the target knowledge point. In other words, for any target knowledge point, its update method can be any of the above. The above method determines the correlation between the target knowledge point and the first education map through feature similarity, and then determines the update method of the target knowledge point based on this, which can improve the accuracy of the update method.

[0034] In one implementation scenario, as a possible example, for each target knowledge point, the average of the similarities between the second feature representation of the target knowledge point and each first feature representation can be taken as the correlation between the target knowledge point and the first education graph. : …… (15) In the above formula (15), represents the target knowledge point, Represents the target knowledge point The second feature representation of the node in the first education graph The first feature represents the similarity between Represents the total number of nodes in the first education graph.

[0035] In one implementation scenario, in response to the relevance If it is greater than the preset threshold θ, the target knowledge point can be determined The update method is to merge into the target node in the first education map. It should be noted that the target node is determined by the similarity between the first feature representation of each node in the first education map and the second feature representation of the target knowledge point. For example, the node corresponding to the highest similarity in the first education map can be selected as the target node.

[0036] In one implementation scenario, different from the above situation, in response to the correlation If it is not greater than the preset threshold θ, the target knowledge point can be determined The update method is to add it as a new node in the first education graph.

[0037] Step S14: updating the target knowledge point to the first education map according to the updating method of the target knowledge point, and obtaining the second education map that self-evolves in the teaching practice.

[0038] In an implementation scenario, as mentioned above, the updating method may include: updating the target knowledge point to a node in the first education map, and adding a new node representing the target knowledge point in the first education map. That is to say, for any target knowledge point, the updating method may be any of the above. On this basis, for any target knowledge point, it can be updated to the first education map according to the updating method determined by it, and a second education map that self-evolves in teaching practice is obtained. In addition, as mentioned above, the second feature representation is extracted from the dynamic relationship graph between the various knowledge points involved in the teaching behavior sequence, and the node relationship that exists in the dynamic relationship graph but does not exist in the first education map can be selected as the target relationship. In this case, the first education map can also be updated based on the target relationship to obtain a second education map that self-evolves in teaching practice. For ease of description, the second education map can be expressed as: …… (16) In the above formula (16), represents the first education graph at time t, represents the newly added nodes and edges of the teaching behavior sequence after time t, The educational knowledge graph at time t+1 is represented as the second educational graph. Through the above process steps, the self-evolution perception ability of the educational knowledge graph in teaching practice can be realized, and it can dynamically respond to scientific education needs and changes in teaching behavior, thereby improving the adaptability and support of the educational knowledge graph to actual teaching scenarios.

[0039] In an implementation scenario, after obtaining the second education graph, the second education graph can be further enhanced for credibility, so as to discover and correct related problems that may arise due to inaccurate reasoning semantics during the reconstruction and updating of the education knowledge graph. Specifically, each reasoning path in the second education graph can be obtained first, and then for each reasoning path, the reliability of the reasoning path, the causal strength of the target node pairs in the reasoning path, and the activation of the reasoning path can be obtained; and the reliability represents the credibility of the reasoning path, the node relationship between the target node pairs is a causal relationship, and the activation represents the possibility of the reasoning path being activated, and then based on the reliability, causal strength and activation, at least the weight parameters of the node relationship between adjacent nodes in the second education graph can be updated. For the sake of ease of description, any reasoning path in the second education graph can be recorded as ,in, Represents a node in the reasoning path. It should be noted that starting from any node in the second education graph and continuing to walk along the connecting edges between nodes until a node is no longer connected to the next node, a reasoning path can be obtained. The above method combines multiple dimensions such as reliability, causal strength, and activation to conduct a credible assessment of the graph path, so as to optimize and reconstruct it accordingly, which helps to improve the credibility and applicability of the educational knowledge graph in scientific education scenarios.

[0040] In a specific implementation scenario, in order to obtain the reliability of the reasoning path, the sub-reliability of the adjacent nodes can be obtained by multiplying the semantic similarity between adjacent nodes in the reasoning path and the confidence of the node relationship between adjacent nodes, and then the reliability of the reasoning path can be obtained by cumulatively multiplying the sub-reliability of each group of adjacent nodes in the reasoning path. It should be noted that the semantic similarity can refer to the relevant description of the aforementioned feature similarity (such as measured by cosine similarity), and the confidence can refer to the relevant description of the aforementioned semantic association strength, which will not be repeated here. For the convenience of description, the reliability of the reasoning path P can be expressed as: …… (17) In the above formula (17), Represents a set of adjacent nodes (i.e. nodes) in the reasoning path ), Represents adjacent nodes in the reasoning path Node relationship The confidence level, That is, it represents a set of adjacent nodes (i.e., nodes) in the reasoning path. The above method can comprehensively measure the reliability of each node and edge in the reasoning path, so as to quantify the reliability of the overall path.

[0041] In a specific implementation scenario, the target node pair may include a first node and a second node. For ease of description, the first node in the target node pair may be denoted as A and the second node as B. Then, in order to obtain the causal strength of the target node pair in the reasoning path, the first probability distribution P(B|do(A)) of the second node B under the intervention of the first node A may be obtained, and the second probability distribution P(B) of the second node may be obtained. Then, the difference between the first probability distribution P(B|do(A)) and the second probability distribution P(B) may be obtained as the causal strength CI(A,B) of the target node pair: …… (18) In the above formula (18), represents the probability distribution of B after intervention A, and Represents the probability distribution of B. The above method, by analyzing the causal strength between knowledge points, can help to optimize the selection of nodes and edges in the reasoning path in the future, and ensure that the reasoning path has a high causal correlation as much as possible.

[0042] In a specific implementation scenario, in order to obtain the activation degree of the reasoning path, the total number of occurrences can be obtained based on the sum of the number of occurrences of each reasoning path in the second education map, and then the activation degree A(P) of the corresponding reasoning path can be obtained based on the ratio of the number of occurrences of any reasoning path in the second education map to the total number of occurrences: …… (19) In the above formula (19), represents the number of occurrences of the reasoning path P in the second education graph, is the path set of each reasoning path in the second education atlas, It represents the sum of the occurrence times of each reasoning path in the second education map (i.e., the total number of occurrences). In other words, through the above operation, the occurrence times of each reasoning path can be normalized as the activation degree of the reasoning path, so as to avoid excessive bias of high-frequency paths on the reasoning results as much as possible, thereby improving the distribution balance of the reasoning process.

[0043] In a specific implementation scenario, after obtaining the reliability, causal strength and activation, at least the weight parameters of the node relationship between adjacent nodes in the second education graph can be updated based on the reliability, causal strength and activation. For example, the semantic association path of the second education graph can be reconstructed by combining the reliability, causal strength and activation. The path set after reconstruction is It can be expressed as: …… (20) In the above formula (20), , Respectively represent the screening thresholds of reliability and causal strength to ensure that the screened reasoning path meets the requirements in terms of reliability and causality, while This means ensuring that the activation degree (the probability of being activated during the reasoning process) of each of the selected reasoning paths is as balanced as possible. Based on the goal of formula (20), the second education graph can be optimized to achieve the reconstruction of the associated path, such as updating the weight parameters of the node relationship between adjacent nodes in the second education graph. For example, increase the weight parameter, or lower the weight parameter, or even cut off a certain node relationship, etc. The specific method of optimizing the reasoning path is not limited here. Through the above-mentioned causal-inspired knowledge graph trust enhancement, not only the reasoning deviation caused by semantic errors is effectively reduced, but also the adaptability of the knowledge graph in dynamic updates and complex reasoning scenarios is enhanced, thereby further improving its practical application value in the field of scientific education.

[0044] In an implementation scenario, as mentioned above, the node types to which each node belongs include: curriculum standard knowledge points, frontier hot knowledge points and teaching scenario knowledge points. After obtaining the second education map, it is also possible to construct a scientific knowledge layer based on each node in the second education map whose node types are curriculum standard knowledge points and frontier hot knowledge points, and to construct a teaching behavior layer based on each node in the second education map whose node types are teaching scenario knowledge points. On this basis, a dynamic association mapping between the teaching behavior layer and the scientific knowledge layer can be formed based on the mapping relationship between the nodes of the scientific knowledge layer and the teaching behavior layer, and the dynamic association mapping at least represents that each teaching behavior in the teaching behavior layer corresponds to the knowledge point node in the scientific knowledge layer. The above method, by constructing the scientific knowledge layer and the teaching behavior layer, can further make the multi-level three-dimensional of the education knowledge map under the premise of meeting multi-level needs, and establish a dynamic association mapping between the levels, so as to realize the interactive linkage between the levels.

[0045] In a specific implementation scenario, as mentioned above, the scientific knowledge layer is the core foundation of the three-dimensional multi-level knowledge graph, carrying the curriculum standard knowledge and cutting-edge hot knowledge in scientific education. The scientific knowledge layer is guided by the curriculum standards and constructs a structured knowledge network through the embedded representation and association relationship of knowledge points. Assume that the knowledge point set is , whose embedding representation can be a node in a high-dimensional vector space: ……(twenty one) In the above formula (21), d represents the vector dimension. On this basis, the association relationship between nodes can be represented by the edge weight matrix W, and the weight value can be calculated by the semantic similarity or co-occurrence relationship between knowledge points: ……(twenty two) In the above formula (22), That is, it represents the semantic similarity between the high-dimensional vectors of the i-th node and the j-th node, which can be calculated by cosine similarity. The specific process can be referred to the above description, which will not be repeated here. Based on this, a scientific knowledge layer can be constructed, and the hierarchical expression and reasoning support of scientific knowledge points can be realized with the same scientific knowledge layer.

[0046] In a specific implementation scenario, the teaching behavior layer can be constructed based on the behavior logic relationship network in scientific teaching practice, which mainly reflects the interaction mode between teachers and students and the dynamic association of knowledge points in teaching activities. Assume that the teaching behavior set is , each teaching behavior It can be expressed as a set of knowledge points involved and their teaching operations: ……(twenty three) In the above formula (23), The operation type that represents the teaching behavior (e.g., introducing knowledge, strengthening understanding, cross-disciplinary connection, etc.), Indicates teaching behavior The teaching behavior layer establishes a logical relationship network between different behaviors by analyzing typical teaching models and excellent teaching cases (both of which contain teaching thinking chains). Its topological structure can be represented by a directed graph. It indicates that, Represents a set of teaching behavior nodes, Represents the logical relationship between teaching behaviors.

[0047] In a specific implementation scenario, based on the construction of the scientific knowledge layer and the teaching behavior layer, a dynamic association mechanism between the two layers can be further established. By mapping the knowledge points of the teaching behavior layer to the scientific knowledge layer, a dynamic association mapping between teaching behavior and scientific knowledge can be formed. The mapping function can be defined as: ……(twenty four) In the above formula (24), Indicates a teaching behavior A certain knowledge point involved The set of knowledge points involved in this teaching behavior , and forms a dynamic association mapping with the scientific knowledge layer .

[0048] In an implementation scenario, after obtaining the second education atlas, it is also possible to obtain the element set of each teaching stage, and the element set of any teaching stage includes each knowledge point and teaching behavior involved in the teaching stage, and then based on the element set of each teaching stage, the correlation between the knowledge points involved in different teaching stages and the correlation between the teaching behaviors involved are dynamically adjusted in the second education atlas. It should be noted that each teaching stage may include but is not limited to: introduction stage, goal statement stage, knowledge teaching stage, ..., extended evaluation stage, etc., and the specific connotations of the teaching stages are no longer given one by one. Exemplarily, the "introduction stage" of the teaching stage may include but is not limited to teaching behaviors such as asking questions, the "goal statement stage" of the teaching stage may include but is not limited to teaching behaviors such as explanations, and the "extended evaluation stage" of the teaching stage may include but is not limited to teaching behaviors such as evaluation. The teaching behaviors involved in the teaching stages are not limited here. For the sake of ease of description, the knowledge points and teaching behaviors involved in the tth teaching stage can be expressed as : …… (25) In the above formula (25), T represents the total number of teaching stages, B t is the set of teaching behaviors in teaching stage t. In addition, For the specific meanings of parameters such as , please refer to the above-mentioned related descriptions, which will not be repeated here. It should be noted that through the staged set design, the logical relationship between knowledge points and behaviors in different teaching stages can be clarified and dynamically adjusted in the multi-level map. Finally, through the combination of the scientific knowledge layer and the teaching behavior layer, a three-dimensional multi-level scientific education knowledge map is constructed, which can not only capture the semantic association and hierarchical structure of scientific knowledge points, but also reflect the dynamic interaction between teaching behaviors and knowledge points. In addition, its dynamic association mechanism supports the optimization of knowledge point mapping in the teaching stage, and ensures that the educational knowledge map can adapt to the complex multi-level and multi-stage needs in scientific education practice as much as possible, and provide systematic intelligent support for scientific education.

[0049] The above scheme performs feature extraction based on the first education map to obtain the first feature representation of each node in the first education map, and performs feature extraction based on the current time series teaching behavior sequence to obtain the second feature representation of each knowledge point involved in the teaching behavior sequence, and the node type to which each node belongs includes: curriculum standard knowledge points, cutting-edge hot knowledge points, and teaching scenario knowledge points. Then, the knowledge points involved in the teaching behavior sequence and not involved in the first education map are selected as target knowledge points, so as to determine the update method of the target knowledge points based on the similarity between the second feature representation of the target knowledge points and the first feature representation, and then according to the update of the target knowledge points The target knowledge points are updated to the first education map in this way, and the second education map that self-evolves in teaching practice is obtained. Therefore, on the one hand, in addition to the curriculum standards, cutting-edge hot spots, teaching scenarios and other levels related to subject knowledge, the knowledge points involved in the teaching behavior level are further perceived, which can meet the multi-level needs of the education knowledge map. On the other hand, by extracting the feature representation of each knowledge point involved in the teaching behavior sequence and measuring its similarity with the feature representation of each node in the knowledge map, the relevant knowledge points involved in the teaching behavior sequence are updated to the knowledge map, which can realize the self-evolution of the education knowledge map in teaching practice, and help improve the updating efficiency of the education knowledge map. Therefore, it can meet the multi-level needs of the education knowledge map and realize the self-evolution of the education knowledge map to improve the updating efficiency of the education knowledge map.

[0050] See also Figure 3 , Figure 3 It is a schematic diagram of a framework of an embodiment of a construction device for an educational knowledge graph of the present application. The construction device 30 for an educational knowledge graph includes: a feature extraction module 31, a knowledge point selection module 32, an update mode determination module 33 and a knowledge graph update module 34. The feature extraction module 31 is used to perform feature extraction based on the first educational graph to obtain a first feature representation of each node in the first educational graph, and to perform feature extraction based on the current time sequence of teaching behavior sequences to obtain a second feature representation of each knowledge point involved in the teaching behavior sequence; wherein the node type to which each node belongs includes: at least one of the curriculum standard knowledge point, the frontier hotspot knowledge point, and the teaching scenario knowledge point; the knowledge point selection module 32 is used to select the knowledge point involved in the teaching behavior sequence and not involved in the first educational graph as the target knowledge point; the update mode determination module 33 is used to determine the update mode of the target knowledge point based on the similarity between the second feature representation of the target knowledge point and each first feature representation; the knowledge graph update module 34 is used to update the target knowledge point to the first educational graph according to the update mode of the target knowledge point, and obtain the second educational graph that self-evolves in the teaching practice.

[0051] In the above scheme, the educational knowledge graph construction device 30 performs feature extraction based on the first educational graph to obtain the first feature representation of each node in the first educational graph, and performs feature extraction based on the current time series teaching behavior sequence to obtain the second feature representation of each knowledge point involved in the teaching behavior sequence, and the node type to which each node belongs includes: curriculum standard knowledge points, cutting-edge hot knowledge points, and teaching scenario knowledge points. Then, the knowledge points involved in the teaching behavior sequence and not involved in the first educational graph are selected as target knowledge points, so as to determine the update method of the target knowledge points based on the similarity between the second feature representation of the target knowledge points and the first feature representation, and then according to the target knowledge points, the target knowledge points are updated. The update method of the target knowledge point updates the target knowledge point to the first education map, and obtains the second education map that self-evolves in teaching practice. Therefore, on the one hand, in addition to the curriculum standards, frontier hot spots, teaching scenarios and other levels related to subject knowledge, it further perceives the knowledge points involved in the teaching behavior level, which can meet the multi-level needs of the education knowledge map. On the other hand, by extracting the feature representation of each knowledge point involved in the teaching behavior sequence and measuring its similarity with the feature representation of each node in the knowledge map, the relevant knowledge points involved in the teaching behavior sequence are updated to the knowledge map, which can realize the self-evolution of the education knowledge map in teaching practice, which helps to improve the updating efficiency of the education knowledge map. Therefore, it can meet the multi-level needs of the education knowledge map and realize the self-evolution of the education knowledge map to improve the updating efficiency of the education knowledge map.

[0052] In some disclosed embodiments, a teaching behavior sequence includes several teaching behaviors, and the teaching behavior definition includes the knowledge points involved in the teaching behavior and the operation type of the teaching behavior. The feature extraction module 31 includes a timing analysis submodule, which is used to perform timing sequence analysis based on the definition information of each teaching behavior in the teaching behavior sequence, and construct a dynamic relationship diagram between each knowledge point involved in the teaching behavior sequence; the feature extraction module 31 includes a feature extraction submodule, which is used to perform feature extraction based on the dynamic relationship diagram to obtain a second feature representation of each knowledge point involved in the teaching behavior sequence.

[0053] In some disclosed embodiments, the update method determination module 33 includes an association calculation submodule, which is used to obtain the association between the target knowledge point and the first educational atlas based on the similarity between the second feature representation of the target knowledge point and each first feature representation; the update method determination module 33 includes a method determination submodule, which is used to determine the update method of the target knowledge point based on the size relationship between the association and a preset threshold.

[0054] In some disclosed embodiments, the method determination submodule includes a first determination unit for determining, in response to a correlation degree being greater than a preset threshold, that a method for updating the target knowledge point is to merge it into a target node in the first education map; wherein the target node is determined by selecting the similarity between the first feature representation of each node in the first education map and the second feature representation of the target knowledge point; the method determination submodule includes a second determination unit for determining, in response to a correlation degree not being greater than a preset threshold, that a method for updating the target knowledge point is to add it as a new node in the first education map.

[0055] In some disclosed embodiments, the second feature representation is extracted from a dynamic relationship graph between various knowledge points involved in the teaching behavior sequence. The educational knowledge graph construction device 30 also includes a node relationship selection module for selecting node relationships that exist in the dynamic relationship graph but do not exist in the first educational graph as target relationships; the knowledge graph update module 34 is also used to update the first educational graph based on the target relationship to obtain a second educational graph that self-evolves in teaching practice.

[0056] In some disclosed embodiments, the educational knowledge graph construction device 30 also includes an inference path acquisition module for acquiring each inference path in the second educational graph; the educational knowledge graph construction device 30 also includes an inference path measurement module for acquiring, for each inference path, the reliability of the inference path, the causal strength of the target node pairs in the inference path and the activation of the inference path; wherein the reliability represents the degree of credibility of the inference path, the node relationship between the target node pairs is a causal relationship, and the activation represents the possibility of the inference path being activated; the educational knowledge graph construction device 30 also includes an inference path optimization module for updating at least the weight parameters of the node relationship between adjacent nodes in the second educational graph based on the reliability, causal strength and activation.

[0057] In some disclosed embodiments, the reasoning path measurement module includes a sub-reliability measurement sub-module, which is used to obtain the sub-reliability of adjacent nodes based on multiplying the semantic similarity between adjacent nodes in the reasoning path and the confidence of the node relationship between adjacent nodes; the reasoning path measurement module includes a sub-reliability accumulation sub-module, which is used to accumulate the sub-reliability of each group of adjacent nodes in the reasoning path to obtain the reliability of the reasoning path.

[0058] In some disclosed embodiments, the reasoning path measurement module includes a probability distribution acquisition submodule, which is used to obtain a first probability distribution of a second node under the intervention of the first node, and to obtain a second probability distribution of the second node; the reasoning path measurement module includes a causal strength acquisition submodule, which is used to obtain the difference between the first probability distribution and the second probability distribution as the causal strength of the target node pair.

[0059] In some disclosed embodiments, the reasoning path measurement module includes an occurrence count submodule, which is used to obtain the total number of occurrences based on the sum of the number of occurrences of each reasoning path in the second education atlas; the reasoning path measurement module includes an activation degree measurement submodule, which is used to obtain the activation degree of the corresponding reasoning path based on the ratio of the number of occurrences of any reasoning path in the second education atlas to the total number of occurrences.

[0060] In some public embodiments, the node types to which each node belongs include: curriculum standard knowledge points, cutting-edge hot knowledge points and teaching scenario knowledge points. The educational knowledge graph construction device 30 also includes a first mapping module for mapping curriculum standard knowledge points in the curriculum standard document to the existing education graph to obtain a third education graph; the educational knowledge graph construction device 30 also includes a second mapping module for predicting and mapping cutting-edge hot knowledge points to the third education graph based on the literature corpus to obtain a fourth education graph; the educational knowledge graph construction device 30 also includes a third mapping module for modeling based on classroom teaching data and mapping teaching scenario knowledge points to the fourth education graph to obtain a first education graph.

[0061] In some public embodiments, the first mapping module includes a curriculum feature extraction submodule, which is used to perform feature extraction based on each curriculum standard text in the curriculum standard document to obtain a third feature representation of each curriculum standard text; the first mapping module includes a curriculum knowledge point mapping submodule, which is used to map the curriculum standard knowledge points representing the curriculum standard text to the existing education map based on the similarity between the third feature representation of the curriculum standard text and the node feature representation of each node in the existing education map to obtain a third education map.

[0062] In some disclosed embodiments, the second mapping module includes a word frequency prediction submodule for obtaining, for each knowledge topic in the literature corpus, a predicted word frequency of the knowledge topic based on the historical word frequency of the knowledge topic at each historical moment; the second mapping module includes a topic selection submodule for selecting a knowledge topic as a candidate topic based on the predicted word frequency of the knowledge topic; the second mapping module includes a hot spot determination submodule for performing topic modeling based on the candidate topics and selecting the candidate topics as hot topics; the second mapping module includes a hot spot mapping submodule for mapping cutting-edge hot knowledge points representing hot topics to the third education map based on the similarity between the fourth feature representation of the hot topic and the node feature representation of each node in the third education map, so as to obtain a fourth education map.

[0063] In some disclosed embodiments, the third mapping module includes a dependency analysis submodule, which is used to perform syntactic dependency analysis based on classroom teaching data to generate a dependency tree; wherein the dependency tree contains each knowledge point involved in the classroom teaching data and its associated logic; the third mapping module includes a potential extraction submodule, which is used to extract potential knowledge points in the teaching scene based on the dependency tree; the third mapping module includes a potential mapping submodule, which is used to map the teaching scene knowledge points representing the potential knowledge points to the fourth education map based on the similarity between the fifth feature representation of the potential knowledge points and the node feature representation of each node in the fourth education map, so as to obtain the first education map.

[0064] In some disclosed embodiments, the educational knowledge graph construction device 30 also includes a similarity measurement module for obtaining the feature similarity between different nodes in the first educational graph; the educational knowledge graph construction device 30 also includes an ambiguity alignment module for selecting to merge or retain corresponding nodes based on the feature similarity between different nodes, so as to perform entity disambiguation and semantic alignment on the first educational graph; the educational knowledge graph construction device 30 also includes a strength determination module for obtaining, for each node in the updated first educational graph, the first word frequency, the co-occurring second word frequency and the similarity between the context information of the knowledge point represented by the node in any two different subject documents, and determining the semantic association strength between the context information of the knowledge point represented by the node in the corresponding two different subject documents based on the first word frequency, the second word frequency and the similarity.

[0065] In some public embodiments, the educational knowledge graph construction device 30 also includes a feature representation acquisition module for acquiring node feature representations of each node in the first educational graph; the educational knowledge graph construction device 30 also includes a feature representation update module for selecting nodes originally connected to the newly added nodes in the first educational graph as nodes to be updated in response to the update of new nodes in the first educational graph, and acquiring node feature representations of the newly added nodes, as well as weighting the node feature representations of the nodes to be updated and the node feature representations of the newly added nodes to obtain updated node feature representations of the nodes to be updated.

[0066] In some public embodiments, the node types to which each node belongs include: curriculum standard knowledge points, cutting-edge hot knowledge points and teaching scenario knowledge points. The construction device 30 of the educational knowledge graph also includes a hierarchical construction module, which is used to construct a scientific knowledge layer based on each node in the second educational graph whose node type is curriculum standard knowledge points and cutting-edge hot knowledge points, and to construct a teaching behavior layer based on each node in the second educational graph whose node type is teaching scenario knowledge points; the construction device 30 of the educational knowledge graph also includes a hierarchical association module, which is used to form a dynamic association mapping between the teaching behavior layer and the scientific knowledge layer based on the mapping relationship between the nodes of the scientific knowledge layer and the teaching behavior layer; wherein the dynamic association mapping at least represents that each teaching behavior in the teaching behavior layer corresponds to a knowledge point node in the scientific knowledge layer.

[0067] In some public embodiments, the educational knowledge graph construction device 30 also includes a stage set acquisition module, which is used to obtain an element set for each teaching stage; wherein the element set of any teaching stage includes each knowledge point and teaching behavior involved in the teaching stage; the educational knowledge graph construction device 30 also includes a stage knowledge association module, which is used to dynamically adjust the association relationship between the knowledge points involved in different teaching stages and the association relationship between the teaching behaviors involved in different teaching stages in the second education graph based on the element set of each teaching stage.

[0068] See also Figure 4 , Figure 4 It is a schematic diagram of the framework of an embodiment of an electronic device of the present application. The electronic device 40 includes at least a memory 41 and a processor 42 coupled to each other, the memory 41 stores at least program instructions, and the processor 42 is used to execute the program instructions to implement the steps in any of the above-mentioned educational knowledge graph construction method embodiments. For details, please refer to the aforementioned disclosed embodiments, which will not be repeated here. As a possible example, the electronic device 40 may include but is not limited to an office notebook, an e-book reader, a smart phone, a tablet computer, a server, etc., and the specific type of the electronic device 40 is not limited here.

[0069] Specifically, the processor 42 is used to control itself and the memory 41 to implement the steps in any of the above-mentioned educational knowledge graph construction method embodiments. The processor 42 can also be called a CPU (Central Processing Unit). The processor 42 may be an integrated circuit chip with signal processing capabilities. The processor 42 can also be a general-purpose processor, a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA) or other programmable logic devices, discrete gates or transistor logic devices, discrete hardware components. The general-purpose processor can be a microprocessor or the processor can also be any conventional processor, etc. In addition, the processor 42 can be implemented by an integrated circuit chip.

[0070] In the above scheme, the electronic device 40 performs feature extraction based on the first education map to obtain the first feature representation of each node in the first education map, and performs feature extraction based on the current time sequence of teaching behavior sequences to obtain the second feature representation of each knowledge point involved in the teaching behavior sequence, and each node belongs to a node type including: curriculum standard knowledge points, cutting-edge hot knowledge points, and teaching scenario knowledge points. Then, the knowledge points involved in the teaching behavior sequence and not involved in the first education map are selected as target knowledge points, so as to determine the update method of the target knowledge points based on the similarity between the second feature representations of the target knowledge points and the first feature representations, and then according to the target knowledge points, the target knowledge points are updated. The target knowledge points are updated to the first education map in an updating manner, and the second education map that self-evolves in teaching practice is obtained. Therefore, on the one hand, in addition to the curriculum standards, cutting-edge hot spots, teaching scenarios and other levels related to subject knowledge, the knowledge points involved in the teaching behavior level are further perceived, which can meet the multi-level needs of the education knowledge map. On the other hand, by extracting the feature representation of each knowledge point involved in the teaching behavior sequence and measuring its similarity with the feature representation of each node in the knowledge map, the relevant knowledge points involved in the teaching behavior sequence are updated to the knowledge map, which can realize the self-evolution of the education knowledge map in teaching practice, and help improve the updating efficiency of the education knowledge map. Therefore, it can meet the multi-level needs of the education knowledge map and realize the self-evolution of the education knowledge map to improve the updating efficiency of the education knowledge map.

[0071] See also Figure 5 , Figure 5Schematic diagram of a framework of an embodiment of a computer-readable storage medium 50 of the present application. The computer-readable storage medium 50 stores program instructions 51 that can be executed by a processor, and the program instructions 51 are used to implement the steps in any of the above-mentioned educational knowledge graph construction method embodiments.

[0072] In the above scheme, the computer-readable storage medium 50 performs feature extraction based on the first education map to obtain the first feature representation of each node in the first education map, and performs feature extraction based on the current time series teaching behavior sequence to obtain the second feature representation of each knowledge point involved in the teaching behavior sequence, and the node type to which each node belongs includes: curriculum standard knowledge point, cutting-edge hot knowledge point, and teaching scenario knowledge point. Then, the knowledge point involved in the teaching behavior sequence and not involved in the first education map is selected as the target knowledge point, so as to determine the update method of the target knowledge point based on the similarity between the second feature representation of the target knowledge point and the first feature representation, and then according to the target knowledge point, the target knowledge point is updated. The updating method of knowledge points updates the target knowledge points to the first education map, and obtains the second education map that self-evolves in teaching practice. Therefore, on the one hand, in addition to the curriculum standards, frontier hot spots, teaching scenarios and other levels related to subject knowledge, it further perceives the knowledge points involved in the teaching behavior level, which can meet the multi-level needs of the education knowledge map. On the other hand, by extracting the feature representation of each knowledge point involved in the teaching behavior sequence and measuring its similarity with the feature representation of each node in the knowledge map, the relevant knowledge points involved in the teaching behavior sequence are updated to the knowledge map, which can realize the self-evolution of the education knowledge map in teaching practice, which helps to improve the updating efficiency of the education knowledge map. Therefore, it can meet the multi-level needs of the education knowledge map and realize the self-evolution of the education knowledge map to improve the updating efficiency of the education knowledge map.

[0073] In some embodiments, the functions or modules included in the device provided by the embodiments of the present disclosure can be used to execute the method described in the above method embodiments. The specific implementation can refer to the description of the above method embodiments, and for the sake of brevity, it will not be repeated here.

[0074] The above description of various embodiments tends to emphasize the differences between the various embodiments. The same or similar aspects can be referenced to each other, and for the sake of brevity, they will not be repeated herein.

[0075] In the several embodiments provided in the present application, it should be understood that the disclosed methods and devices can be implemented in other ways. For example, the device implementation described above is only schematic. For example, the division of modules or units is only a logical function division. There may be other division methods in actual implementation, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, and the indirect coupling or communication connection of devices or units can be electrical, mechanical or other forms.

[0076] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the present embodiment.

[0077] In addition, each functional unit in each embodiment of the present application may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware or in the form of software functional units.

[0078] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application is essentially or the part that contributes to the prior art or all or part of the technical solution can be embodied in the form of a software product. The computer software product is stored in a storage medium, including several instructions to enable a computer device (which can be a personal computer, server, or network device, etc.) or a processor to execute all or part of the steps of each implementation method of the present application. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk and other media that can store program code.

[0079] If the technical solution of this application involves personal information, the product using the technical solution of this application has clearly informed the personal information processing rules and obtained the individual's voluntary consent before processing the personal information. If the technical solution of this application involves sensitive personal information, the product using the technical solution of this application has obtained the individual's separate consent before processing the sensitive personal information, and at the same time meets the "explicit consent" requirement. For example, on personal information collection devices such as cameras, clear and prominent signs are set to inform that the personal information collection scope has been entered and personal information will be collected. If the individual voluntarily enters the collection scope, it is deemed that he or she agrees to the collection of his or her personal information; or on the device that processes personal information, the personal information processing rules are notified by obvious signs / information, and the individual's authorization is obtained through pop-up information or by asking the individual to upload his or her personal information; among them, the personal information processing rules may include information such as the personal information processor, the purpose of personal information processing, the processing method, and the type of personal information processed.

Claims

1. A method for constructing an educational knowledge graph, characterized in that: include: Based on the first education map, feature extraction is performed to obtain a first feature representation of each node in the first education map, and based on the current time series teaching behavior sequence, feature extraction is performed to obtain a second feature representation of each knowledge point involved in the teaching behavior sequence; wherein the node type to which each node belongs includes: at least one of a curriculum standard knowledge point, a frontier hotspot knowledge point, and a teaching scenario knowledge point; Selecting knowledge points involved in the teaching behavior sequence and not involved in the first education map as target knowledge points; Determining an update method for the target knowledge point based on similarities between the second feature representation of the target knowledge point and each of the first feature representations; The target knowledge points are updated to the first education map according to the updating method of the target knowledge points, so as to obtain a second education map that self-evolves in teaching practice.

2. The method according to claim 1, characterized in that The teaching behavior sequence includes a plurality of teaching behaviors, wherein the teaching behavior definition includes knowledge points involved in the teaching behavior and operation types of the teaching behavior, and the teaching behavior sequence based on the current time sequence is subjected to feature extraction to obtain a second feature representation of each knowledge point involved in the teaching behavior sequence, including: Performing time series analysis based on definition information of each teaching behavior in the teaching behavior sequence, and constructing a dynamic relationship diagram between each knowledge point involved in the teaching behavior sequence; Feature extraction is performed based on the dynamic relationship graph to obtain a second feature representation of each knowledge point involved in the teaching behavior sequence.

3. The method according to claim 1, characterized in that The determining the update method of the target knowledge point based on the similarity between the second feature representation of the target knowledge point and each of the first feature representations includes: Based on the similarity between the second feature representation of the target knowledge point and each of the first feature representations, obtaining the correlation between the target knowledge point and the first education graph; Based on the magnitude relationship between the degree of association and a preset threshold, an updating method of the target knowledge point is determined.

4. The method according to claim 3, characterized in that The determining of the update method of the target knowledge point based on the magnitude relationship between the correlation degree and a preset threshold value includes at least one of the following: In response to the correlation being greater than the preset threshold, determining that the target knowledge point is updated in a manner of merging it into a target node in the first education graph; wherein the target node is selected and determined by the similarity between the first feature representation of each of the nodes in the first education graph and the second feature representation of the target knowledge point; In response to the degree of association being not greater than the preset threshold, it is determined that the target knowledge point is updated by adding it as a new node of the first education graph.

5. The method according to claim 1, characterized in that The second feature representation is extracted from a dynamic relationship diagram between various knowledge points involved in the teaching behavior sequence, and the method further includes: Selecting a node relationship that exists in the dynamic relationship graph but does not exist in the first education graph as a target relationship; The first education map is updated based on the target relationship to obtain a second education map that self-evolves in teaching practice.

6. The method according to claim 1, characterized in that After updating the target knowledge point to the first education map according to the update method of the target knowledge point to obtain a second education map that is self-evolved in teaching practice, the method further includes: Obtaining each reasoning path in the second education graph; For each of the reasoning paths, the reliability of the reasoning path, the causal strength of the target node pair in the reasoning path, and the activation degree of the reasoning path are obtained; wherein the reliability represents the credibility of the reasoning path, the node relationship between the target node pair is a causal relationship, and the activation degree represents the possibility of the reasoning path being activated; Based on the reliability, the causal strength and the activation degree, at least the weight parameters of the node relationship between adjacent nodes in the second education graph are updated.

7. The method according to claim 6, characterized in that The step of obtaining the reliability includes: Based on the semantic similarity between the adjacent nodes in the reasoning path and the confidence of the node relationship between the adjacent nodes, the sub-reliability of the adjacent nodes is obtained by multiplying them; The reliability of the reasoning path is obtained by multiplying the sub-reliabilities of each group of adjacent nodes in the reasoning path.

8. The method according to claim 6, characterized in that The target node pair includes a first node and a second node, and the step of acquiring the causal strength includes: Acquire a first probability distribution of the second node under the intervention of the first node, and acquire a second probability distribution of the second node; A difference between the first probability distribution and the second probability distribution is obtained as the causal strength of the target node pair.

9. The method according to claim 6, characterized in that The step of acquiring the activation degree comprises: Obtaining a total number of occurrences based on the sum of the number of occurrences of each of the reasoning paths in the second education graph; Based on the ratio of the number of occurrences of any one of the reasoning paths in the second education map to the total number of occurrences, the activation degree of the corresponding reasoning path is obtained.

10. The method according to claim 1, characterized in that The node types to which each node belongs include: curriculum standard knowledge points, cutting-edge hot knowledge points and teaching scenario knowledge points. The first education map acquisition step includes: Mapping the curriculum standard knowledge points in the curriculum standard document to the existing education map to obtain the third education map; Predict and map cutting-edge hot knowledge points to the third education map based on the literature corpus to obtain a fourth education map; Based on classroom teaching data modeling and mapping teaching scenario knowledge points to the fourth education map, the first education map is obtained.

11. The method according to claim 10, characterized in that The mapping of the curriculum standard knowledge points in the curriculum standard document to the existing education map to obtain the third education map includes: Based on each course standard text in the course standard document, feature extraction is performed to obtain a third feature representation of each course standard text; Based on the similarity between the third feature representation of the curriculum standard text and the node feature representation of each node in the existing education map, the curriculum standard knowledge points representing the curriculum standard text are mapped to the existing education map to obtain the third education map.

12. The method according to claim 10, characterized in that The method of predicting and mapping cutting-edge hot knowledge points to the third education map based on the literature corpus to obtain the fourth education map includes: For each knowledge topic in the document corpus, based on the historical word frequency of the knowledge topic at each historical moment, a predicted word frequency of the knowledge topic is obtained; selecting the knowledge topic as a candidate topic based on the predicted word frequency of the knowledge topic; Perform topic modeling based on the candidate topics, and select the candidate topics as hot topics; Based on the similarity between the fourth feature representation of the hot topic and the node feature representation of each node in the third education map, the cutting-edge hot knowledge points representing the hot topic are mapped to the third education map to obtain the fourth education map.

13. The method according to claim 10, characterized in that The modeling based on classroom teaching data and mapping teaching scenario knowledge points to the fourth education map to obtain the first education map includes: Performing syntactic dependency analysis based on the classroom teaching data to generate a dependency tree; wherein the dependency tree includes various knowledge points involved in the classroom teaching data and their associated logic; Based on the dependency tree, potential knowledge points in the teaching scenario are extracted; Based on the similarity between the fifth feature representation of the potential knowledge point and the node feature representation of each node in the fourth education map, the teaching scenario knowledge point representing the potential knowledge point is mapped to the fourth education map to obtain the first education map.

14. The method according to claim 10, characterized in that After the modeling based on classroom teaching data and mapping the teaching scenario knowledge points to the fourth education map to obtain the first education map, the method further includes: Obtaining feature similarities between different nodes in the first education graph; Based on the feature similarity between the different nodes, select to merge or retain the corresponding nodes to perform entity disambiguation and semantic alignment on the first education graph; For each node in the first education map after update, obtain the first word frequency, the co-occurring second word frequency and the similarity between the context information of the knowledge point represented by the node in any two different subject documents, and based on the first word frequency, the second word frequency and the similarity, determine the semantic association strength between the context information of the knowledge point represented by the node in the corresponding two different subject documents.

15. The method according to claim 10, characterized in that After the modeling based on classroom teaching data and mapping the teaching scenario knowledge points to the fourth education map to obtain the first education map, the method further includes: Obtaining node feature representations of each node in the first education graph; In response to a new node being updated in the first education map, a node originally connected to the new node in the first education map is selected as a node to be updated, and a node feature representation of the new node is obtained, and the node feature representation of the node to be updated and the node feature representation of the new node are weighted to obtain an updated node feature representation of the node to be updated.

16. The method according to claim 1, characterized in that The node types to which each node belongs include: curriculum standard knowledge points, cutting-edge hot knowledge points and teaching scenario knowledge points. After updating the target knowledge points to the first education map in accordance with the update method of the target knowledge points to obtain a second education map that self-evolves in teaching practice, the method further includes: Based on each node in the second education graph whose node type is the curriculum standard knowledge point and the frontier hot knowledge point, construct a scientific knowledge layer, and based on each node in the second education graph whose node type is the teaching scenario knowledge point, construct a teaching behavior layer; Based on the mapping relationship between the nodes of the scientific knowledge layer and the teaching behavior layer, a dynamic association mapping between the teaching behavior layer and the scientific knowledge layer is formed; wherein the dynamic association mapping at least represents that each teaching behavior in the teaching behavior layer corresponds to a knowledge point node in the scientific knowledge layer.

17. The method according to claim 1, characterized in that After updating the target knowledge point to the first education map according to the update method of the target knowledge point to obtain a second education map that is self-evolved in teaching practice, the method further includes: Obtaining element sets of each teaching stage; wherein the element set of any teaching stage includes each knowledge point and teaching behavior involved in the teaching stage; Based on the element set of each teaching stage, the association relationship between the knowledge points involved in the different teaching stages and the association relationship between the teaching behaviors involved in the different teaching stages are dynamically adjusted in the second education map.

18. A device for constructing an educational knowledge graph, characterized in that: include: A feature extraction module is used to perform feature extraction based on the first education map to obtain a first feature representation of each node in the first education map, and to perform feature extraction based on the current time sequence of teaching behaviors to obtain a second feature representation of each knowledge point involved in the teaching behavior sequence; wherein the node types to which each node belongs include: at least one of a curriculum standard knowledge point, a frontier hotspot knowledge point, and a teaching scenario knowledge point; A knowledge point selection module, used for selecting knowledge points involved in the teaching behavior sequence and not involved in the first education map as target knowledge points; An update mode determination module, used to determine an update mode of the target knowledge point based on the similarity between the second feature representation of the target knowledge point and each of the first feature representations; The knowledge graph updating module is used to update the target knowledge point to the first education graph according to the updating method of the target knowledge point, so as to obtain a second education graph that self-evolves in teaching practice.

19. An electronic device, characterized in that: It at least includes a memory and a processor coupled to each other, wherein the memory at least stores program instructions, and the processor is used to execute the program instructions to implement the method for constructing an educational knowledge graph as described in any one of claims 1 to 17.

20. A computer-readable storage medium, characterized in that: Program instructions that can be executed by a processor are stored, and the program instructions are used to implement the method for constructing an educational knowledge graph as described in any one of claims 1 to 17.

Images