A knowledge point annotation method, device, equipment and storage medium

Abstract

The present application discloses a knowledge point annotation method, device, equipment and storage medium. The method includes: obtaining test question text data and a knowledge point structure tree; inputting the test question text data into a semantic recognition network for semantic recognition to obtain text feature information of the test question text data and word feature information of multiple word segmentation data of the test question text data; based on the feature recognition network, performing feature recognition on the knowledge point structure tree to obtain knowledge point feature information of the knowledge point structure tree; fusing the knowledge point feature information, word feature information and text feature information to obtain target fusion information; inputting the target fusion information into the knowledge point recognition network for knowledge point annotation to obtain the annotated knowledge point information corresponding to the test question text data. In the scenario of knowledge point annotation of test questions, the present application improves the accuracy of the representation of test question features, thereby improving the accuracy and effectiveness of knowledge point annotation of test questions.

Description

Technical Field

[0001] The present application relates to the field of artificial intelligence technology, and in particular to a knowledge point annotation method, apparatus, device, and storage medium. Background Art

[0002] Currently, knowledge point annotation for test questions is widely used in educational products such as test resource search, test question recommendation, question bank construction, and personalized learning. Existing knowledge point annotation for test questions is typically implemented using traditional machine learning methods, such as using convolutional neural networks to annotate test question text with knowledge points.

[0003] However, existing knowledge point annotation methods only use shallow semantic features of the test text to annotate knowledge points, ignoring the association features between knowledge points, resulting in low accuracy of knowledge point annotation. Therefore, an accurate and effective technical solution is needed. Summary of the Invention

[0004] In order to solve the problem of low accuracy when applying existing technologies to knowledge point annotation, this application provides a knowledge point annotation method, device, equipment and storage medium:

[0005] According to a first aspect of the present application, a knowledge point annotation method is provided, the method comprising:

[0006] Obtain test question text data and knowledge point structure tree;

[0007] Inputting the test question text data into a semantic recognition network for semantic recognition, and obtaining text feature information of the test question text data and word feature information of a plurality of word segmentation data of the test question text data;

[0008] Based on a feature recognition network, feature recognition is performed on the knowledge point structure tree to obtain knowledge point feature information of the knowledge point structure tree, wherein the knowledge point feature information represents a structural semantic feature of the knowledge point structure tree;

[0009] Fusing the knowledge point feature information, the word feature information, and the text feature information to obtain target fusion information;

[0010] The target fusion information is input into a knowledge point recognition network for knowledge point annotation to obtain the annotated knowledge point information corresponding to the test question text data.

[0011] According to a second aspect of the present application, a knowledge point annotation device is provided, the method comprising:

[0012] The test question text data acquisition module is used to obtain the test question text data and the knowledge point structure tree;

[0013] A semantic recognition module is used to input the test question text data into a semantic recognition network for semantic recognition, and obtain text feature information of the test question text data and word feature information of a plurality of word segmentation data of the test question text data;

[0014] A feature recognition module, configured to perform feature recognition on the knowledge point structure tree based on a feature recognition network to obtain knowledge point feature information of the knowledge point structure tree, wherein the knowledge point feature information represents a structural semantic feature of the knowledge point structure tree;

[0015] A fusion processing module, configured to fuse the knowledge point feature information, the word feature information, and the text feature information to obtain target fusion information;

[0016] The knowledge point annotation module is used to input the target fusion information into the knowledge point recognition network for knowledge point annotation, and obtain the annotated knowledge point information corresponding to the test question text data.

[0017] According to the third aspect of the present application, a knowledge point annotation device is provided, which includes a processor and a memory, wherein the memory stores at least one instruction or at least one program, and the at least one instruction or the at least one program is loaded and executed by the processor to implement the knowledge point annotation method as described in the first aspect.

[0018] According to the fourth aspect of the present application, a computer-readable storage medium is provided, in which at least one instruction or at least one program is stored. The at least one instruction or the at least one program is loaded and executed by a processor to implement the knowledge point annotation method as described in the first aspect.

[0019] According to a fifth aspect of the present application, a computer program product or computer program is provided, comprising computer instructions stored in a computer-readable storage medium. A processor of a computer device reads the computer instructions from the computer-readable storage medium and executes the computer instructions, causing the computer device to perform the knowledge point labeling method described in the first aspect.

[0020] The present application provides a knowledge point annotation method, apparatus, device, and storage medium, which have the following technical effects:

[0021] This application obtains test text data and a knowledge point structure tree; then inputs the test text data into a semantic recognition network for semantic recognition to obtain text feature information of the test text data and word feature information of multiple word segmentation data of the test text data; and based on the feature recognition network, performs feature recognition on the knowledge point structure tree to obtain knowledge point feature information of the knowledge point structure tree, where the knowledge point feature information represents the structured semantic features of the knowledge point structure tree; then fuses the knowledge point feature information, word feature information and text feature information to obtain target fusion information; finally, inputs the target fusion information into the knowledge point recognition network for knowledge point annotation to obtain annotated knowledge point information corresponding to the test text data. In the scenario of knowledge point annotation of test question text, this application fuses the knowledge point feature information of the structured semantic features of the knowledge point structure tree, the text feature information of the test question text data, and the word feature information of multiple word segmentation data of the test question text data to obtain target fusion information, and annotates the target fusion information with identification points. The deep semantic association between the test question text and the knowledge points in the target fusion information and the hierarchical dependency between the knowledge point structures can be utilized to improve the accuracy of the representation of the test question features, thereby greatly improving the accuracy and effectiveness of knowledge point annotation of the test questions. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] In order to more clearly illustrate the technical solutions and advantages of the embodiments of the present application or the prior art, the following is a brief introduction to the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without any creative work.

[0023] Figure 1 is a schematic diagram of an application environment provided by an embodiment of the present application;

[0024] Figure 2 This is a flow chart of a knowledge point annotation method provided in an embodiment of the present application;

[0025] Figure 3 This is a knowledge point structure tree composed of some knowledge points of mathematics provided in an embodiment of the present application;

[0026] Figure 4 This is a flow chart of an embodiment of the present application, which provides a method for inputting test text data into a semantic recognition network for semantic recognition, and obtaining text feature information of the test text data and word feature information of multiple word segmentation data of the test text data;

[0027] Figure 5This is a flow chart of a method for performing feature recognition on a knowledge point structure tree based on a feature recognition network to obtain knowledge point feature information of the knowledge point structure tree, provided by an embodiment of the present application;

[0028] Figure 6 This is a flow chart of a method for fusing knowledge point feature information, word feature information, and text feature information to obtain target fusion information, provided by an embodiment of the present application;

[0029] Figure 7 This is a flow chart of an embodiment of the present application, which provides a method for inputting knowledge point feature information and word feature information into a fusion network for fusion processing to obtain initial fusion information;

[0030] Figure 8 This is a flow chart of a network joint training method provided in an embodiment of the present application;

[0031] Figure 9 This is a schematic diagram of a knowledge point annotation network provided in an embodiment of the present application;

[0032] Figure 10 This is a block diagram of a knowledge point annotation device provided in an embodiment of the present application;

[0033] Figure 11 This is a structural diagram of a knowledge point annotation device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0034] The following will be combined with the accompanying drawings in the embodiments of this application to clearly and completely describe the technical solutions in the embodiments of this application. Obviously, the embodiments described are only part of the embodiments of this application, not all of the embodiments. Based on the embodiments of this application, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of this application.

[0035] It should be noted that the terms "including" and "having" and any variations thereof in the specification and claims of this application and the above-mentioned drawings are intended to cover non-exclusive inclusions. For example, a process, method, system, product or server that includes a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or are inherent to these processes, methods, products or devices.

[0036] It is understandable that in the specific implementation of this application, related data such as user information is involved. When the above embodiments of this application are applied to specific products or technologies, user permission or consent is required, and the collection, use and processing of relevant data must comply with relevant laws, regulations and standards of relevant countries and regions.

[0037] See also Figure 1 , Figure 1 It is a schematic diagram of an application environment provided by an embodiment of the present application, which may include a client 10 and a server 20, and the client 10 and the server 20 may be directly or indirectly connected via wired or wireless communication. The user may send a knowledge point annotation request for the test question text data to the server 20 via the client 10. The server 20 determines the corresponding test question text data to be annotated based on the knowledge point annotation request, and then inputs the test question text data to be annotated into the semantic recognition network for semantic recognition to obtain text feature information of the test question text data to be identified and word feature information of multiple word segmentation data of the test question text data to be identified, and then performs feature recognition on the knowledge point structure tree based on the feature recognition network to obtain knowledge point feature information of the knowledge point structure tree, and then fuses the knowledge point feature information, word feature information and text feature information to obtain target fusion information, and then inputs the target fusion information into the knowledge point recognition network for knowledge point annotation to obtain the annotated knowledge point information corresponding to the test question text data to be annotated, and returns the annotated knowledge point information to the client 10. It should be noted that Figure 1 Just an example.

[0038] The client can be a physical device such as a smartphone, computer (e.g., desktop, tablet, laptop), digital assistant, intelligent voice interaction device (e.g., smart speaker), or smart wearable device. It can also be software running on a physical device, such as a computer program. The client's operating system can be Android, iOS (a mobile operating system developed by Apple), Linux, or Microsoft Windows.

[0039] The server side can be an independent physical server, a server cluster or distributed system composed of multiple physical servers, or a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communications, middleware services, domain name services, security services, CDN (Content Delivery Network), and big data and artificial intelligence platforms. The server can include network communication units, processors, and memory, etc. The server side can provide backend services for the corresponding client.

[0040] The client 10 and server 20 described above can be used to construct a system for knowledge point annotation, which can be a distributed system. Taking a blockchain system as an example, a distributed system is composed of multiple nodes (any form of computing device connected to the network, such as a server or user terminal) and clients. The nodes form a peer-to-peer (P2P) network. The P2P protocol is an application layer protocol that runs on top of the Transmission Control Protocol (TCP). In a distributed system, any machine, such as a server or terminal, can join and become a node. A node includes a hardware layer, an intermediate layer, an operating system layer, and an application layer.

[0041] The functions of each node in the above blockchain system include:

[0042] 1) Routing: A basic function of a node, used to support communication between nodes.

[0043] In addition to the routing function, nodes can also have the following functions:

[0044] 2) Applications, deployed in the blockchain, implement specific services based on actual business needs, record data related to the implementation of functions to form record data, carry digital signatures in the record data to indicate the source of the task data, and send the record data to other nodes in the blockchain system for other nodes to add the record data to a temporary block when they successfully verify the source and integrity of the record data.

[0045] 3) Blockchain, including a series of blocks that are connected to each other in the order of their generation. Once a new block is added to the blockchain, it will not be removed. The block records the record data submitted by the nodes in the blockchain system.

[0046] The following describes a specific embodiment of a knowledge point annotation method provided by this application. Figure 2 It is a flow chart of a knowledge point annotation method provided by an embodiment of the present application. The present application provides method operation steps as described in the embodiment or flow chart, but may include more or fewer operation steps based on conventional or non-creative labor. The order of steps listed in the embodiment is only one way of executing the order of many steps and does not represent the only execution order. When the actual system or product is executed, it can be executed in sequence or in parallel according to the method shown in the embodiment or the accompanying drawings (for example, in a parallel processor or multi-threaded processing environment). Specifically, Figure 2 As shown, the method may include:

[0047] S201, obtaining the test question text data and the knowledge point structure tree.

[0048] In an embodiment of the present application, the test question text data may include test question text data of multiple subjects. Specifically, the test question text data may include but is not limited to the test question stem, test question answers, test question analysis and other parts. The data type of the test question text data may include but is not limited to text, formulas, charts, etc.

[0049] In the embodiment of the present application, the knowledge point structure tree may include knowledge point structure trees of multiple subjects, wherein the knowledge point structure tree of each subject may be a tree-like knowledge system obtained by hierarchically subdividing subject knowledge points in educational documents such as subject textbooks, subject curriculum standards, and subject examination guidelines according to subject knowledge logic. Figure 3 , Figure 3 It is a knowledge point structure tree composed of some knowledge points of the mathematics subject provided by an embodiment of the present application.

[0050] In a specific embodiment, the subject knowledge point framework or the subject knowledge point mind map may be converted to obtain a knowledge point structure tree of the current subject.

[0051] In practical applications, the subject information corresponding to the test question text data can be annotated in advance to obtain a knowledge point structure tree matching the subject information.

[0052] S202 , inputting the test question text data into a semantic recognition network for semantic recognition, and obtaining text feature information of the test question text data and word feature information of a plurality of word segmentation data of the test question text data.

[0053] In the embodiments of the present application, text feature information can represent the contextual semantic features of the test question text, and word feature information can represent the semantic features of the corresponding word segmentation data in the test question text. Specifically, after the test question text data is input into the semantic recognition network, the semantic recognition network can perform semantic recognition based on the context of the test question text data to obtain text feature information and word feature information.

[0054] In a specific embodiment, the text feature information may be expressed in the form of a text feature vector, and the word feature information may be expressed in the form of a word feature vector.

[0055] In a specific embodiment, the semantic recognition network can be obtained by performing semantic recognition training on a preset semantic recognition network, wherein the preset semantic recognition network can include a word segmentation layer, a semantic recognition layer and a semantic fusion layer. Specifically, Figure 4 As shown, the above-mentioned inputting the test question text data into the semantic recognition network for semantic recognition, obtaining the text feature information of the test question text data and the word feature information of the multiple word segmentation data of the test question text data may include:

[0056] S401: Input the test text data into the word segmentation layer for word segmentation preprocessing to obtain a plurality of word segmentation data.

[0057] In a specific embodiment, the test question stem, test question answer and test question analysis in the test question text data can be spliced ​​to obtain test question text splicing data, and the test question text splicing data can be input into the word segmentation layer for word segmentation preprocessing to obtain a target word segmentation sequence, which can include multiple word segmentation data.

[0058] In an optional embodiment, the plurality of word segmentation data may further include a preset separator, wherein the preset separator may include a classification separator.

[0059] S402: Input the multiple word segmentation data into the semantic recognition layer for semantic recognition, and obtain initial semantic information, position information and sentence information corresponding to the multiple word segmentation data.

[0060] Specifically, the initial semantic information can represent the semantic features of the corresponding word segmentation data itself, the sentence information can represent the identification features of the sentence where the corresponding word segmentation data is located in the test text data, and the position information can represent the position features of the corresponding word segmentation data in the sentence. Among them, the expression form of the initial semantic information can be an initial semantic vector, the expression form of the sentence information can be a sentence vector, and the expression form of the position information can be a position vector.

[0061] In a specific embodiment, the semantic recognition layer may include a first vectorization layer, a second vectorization layer and a third vectorization layer. Multiple word segmentation data are input into the first vectorization layer for semantic feature extraction to obtain an initial semantic vector. Multiple word segmentation data are input into the second vectorization layer for identification feature extraction to obtain a sentence vector. Multiple word segmentation data are input into the third vectorization layer for position feature extraction to obtain a position vector.

[0062] S403: Input the initial semantic information, position information and sentence information into the semantic fusion layer to perform contextual semantic fusion on the multiple word segmentation data to obtain text feature information and word feature information.

[0063] In a specific embodiment, the semantic fusion layer may include a bidirectional semantic encoder, which inputs the initial semantic information, position information and sentence information into the bidirectional semantic encoder, and fuses the semantic features of the context of each word segmentation data through the bidirectional semantic encoder to obtain the word feature information of each word segmentation data.

[0064] In an optional embodiment, since the above-mentioned multiple word segmentation data can also include classification separators, the initial semantic information, position information and sentence information corresponding to the above-mentioned multiple word segmentation data can include: the initial semantic information, position information and sentence information corresponding to the classification separator; accordingly, the above-mentioned inputting the initial semantic information, position information and sentence information into the semantic fusion layer to perform contextual semantic fusion on the multiple word segmentation data to obtain text feature information and word feature information can include: inputting the initial semantic information, position information and sentence information corresponding to the classification separator into the semantic fusion layer to perform contextual semantic fusion on the multiple word segmentation data to obtain the word feature information of the classification separator. Since in actual applications, the word feature information of the classification separator can represent the contextual semantic features of the test paper text data, the word feature information of the classification separator is used as the text feature information.

[0065] In another optional embodiment, semantic feature aggregation processing may be performed on the word feature information of each word segmentation data to obtain text feature information. Optionally, performing semantic feature aggregation processing on the word feature information of each word segmentation data to obtain text feature information may include performing mean processing on the word feature information of each word segmentation data to obtain text feature information.

[0066] In addition, it should be noted that the semantic recognition network described in the embodiment of the present application is not limited to the above-mentioned preset semantic recognition network. In practical applications, it can also include other machine learning networks, such as recurrent neural networks, ELMO models (embedded representation models based on language models), GPT models (generative pre-trained language models), etc. The embodiment of the present application is not limited to the above-mentioned machine learning networks.

[0067] It can be seen from the above embodiments that the contextual semantic recognition of the test text data is performed through a semantic recognition network including a word segmentation layer, a semantic recognition layer and a semantic fusion layer, so as to realize the fusion of the contextual semantic features of the test text data, thereby improving the accuracy of the representation of the semantic features of the test questions.

[0068] S203 , performing feature recognition on the knowledge point structure tree based on the feature recognition network to obtain knowledge point feature information of the knowledge point structure tree, where the knowledge point feature information represents the structured semantic features of the knowledge point structure tree.

[0069] In an embodiment of the present application, the knowledge point feature information can represent the structured semantic features of the knowledge point structure tree. The structured semantic features of the knowledge point structure tree may include the structural features and semantic features of the knowledge point structure tree. The structural features of the knowledge point structure tree may be: the structural hierarchical relationship of the knowledge points in the knowledge point structure tree and the dependency relationship between the knowledge points at each layer. The semantic features of the knowledge point structure tree may be the aggregated semantics from the end node to the root node.

[0070] In a specific embodiment, the feature recognition network can be obtained by performing feature recognition training on a preset feature recognition network, wherein the preset feature recognition network can include multiple feature aggregation layers corresponding to multiple nodes in a preset knowledge point structure tree, specifically, Figure 5 As shown, the feature recognition network is used to perform feature recognition on the knowledge point structure tree, and the knowledge point feature information obtained from the knowledge point structure tree may include:

[0071] S501 , performing semantic recognition on knowledge point data of each of the multiple nodes to obtain node semantic information corresponding to each node.

[0072] Specifically, the node semantic information can represent the semantic features of the knowledge point data of the corresponding node.

[0073] In a specific embodiment, the feature recognition network may further include a knowledge point semantic recognition layer, and the knowledge point data of each node is input into the knowledge point semantic recognition layer for semantic recognition to obtain node semantic information corresponding to each node.

[0074] S502: Input the node semantic information corresponding to the end node of the knowledge point structure tree into the feature aggregation layer corresponding to the end node to obtain the node aggregation feature information corresponding to the end node.

[0075] S503: Starting from the parent node of the last node, traverse multiple nodes.

[0076] S504, inputting the node aggregation feature information corresponding to the child nodes of the currently traversed node and the node semantic information corresponding to the currently traversed node into the feature aggregation layer corresponding to the currently traversed node to obtain the node aggregation feature information corresponding to the currently traversed node.

[0077] Specifically, the node aggregation feature information can represent the structural hierarchical relationship and aggregation semantics from the end node to the corresponding node.

[0078] S505: After traversing multiple nodes, aggregate feature information of the nodes corresponding to the root node in the knowledge point structure tree is used as knowledge point feature information.

[0079] In a specific embodiment, the node semantic information may be expressed in a node semantic vector, the node aggregation feature information may be expressed in a node aggregation feature vector, and correspondingly, the knowledge point feature information may be expressed in a knowledge point feature vector. The above-mentioned feature recognition based on the feature recognition network for the knowledge point structure tree to obtain the knowledge point feature information of the knowledge point structure tree may include the feature recognition based on the feature recognition network for the knowledge point structure tree to obtain the knowledge point feature vector of the knowledge point structure tree.

[0080] Correspondingly, the above-mentioned feature recognition network-based feature recognition of the knowledge point structure tree to obtain the knowledge point feature vector of the knowledge point structure tree may include: performing semantic recognition on the knowledge point data of each node in multiple nodes to obtain the node semantic vector corresponding to each node; inputting the node semantic vector corresponding to the end node of the knowledge point structure tree into the feature aggregation layer corresponding to the end node in the feature recognition network to obtain the node aggregation feature vector corresponding to the end node; starting from the parent node of the end node, traversing multiple nodes from bottom to top; inputting the node aggregation feature vector corresponding to the child node of the currently traversed node and the node semantic vector corresponding to the currently traversed node into the feature aggregation layer corresponding to the currently traversed node in the feature recognition network to obtain the node aggregation feature vector corresponding to the currently traversed node; after traversing multiple nodes, the node aggregation feature vector corresponding to the root node in the knowledge point structure tree is used as the knowledge point feature vector.

[0081] In addition, it should be noted that the feature recognition network described in the embodiment of the present application is not limited to the above-mentioned preset feature recognition network. In practical applications, it can also include other machine learning networks, such as GCN (graph convolutional neural network), GraphSAGE (graph sampling and aggregation network), GAN (graph attention network), Tree-LSTM (tree-structured long short-term memory network), etc. The embodiment of the present application is not limited to the above-mentioned machine learning network.

[0082] It can be seen from the above embodiments that features of the knowledge point structure tree are extracted through a feature extraction network including multiple feature aggregation layers corresponding one-to-one to multiple nodes in the knowledge point structure tree, thereby aggregating the structural hierarchical relationship and semantic features of the knowledge point structure tree from the end node to the root node, thereby improving the accuracy of the representation of the features of the knowledge point structure tree.

[0083] S204: Fusing the knowledge point feature information, word feature information, and text feature information to obtain target fusion information.

[0084] In an embodiment of the present application, the target fusion information can be the test question text expression obtained by fusing the semantic features of the test question text data with the structural features and semantic features of the knowledge point structure tree, which can effectively capture the correlation between the test question text data and the knowledge point structure tree features.

[0085] In a specific embodiment, the target fusion information may be expressed as a target fusion vector.

[0086] In a specific embodiment, Figure 6 As shown, the above fusion processing of knowledge point feature information, word feature information and text feature information can obtain target fusion information, which may include:

[0087] S601: Input the knowledge point feature information and word feature information into the fusion network for fusion processing to obtain initial fusion information.

[0088] Specifically, the initial fusion information may be a test question text representation of structured semantic features embedded in a knowledge point structure tree, wherein the initial fusion information may be expressed in an initial fusion vector.

[0089] In a specific embodiment, the fusion network can be obtained by fusion processing and training a preset fusion network. The preset fusion network may include an association analysis layer, a normalization layer, and a weighting layer. Specifically, Figure 7 As shown, the above-mentioned knowledge point feature information and word feature information are input into the fusion network for fusion processing, and the initial fusion information obtained may include:

[0090] S701: Input word feature information and knowledge point feature information into the association analysis layer for association analysis to obtain target association information.

[0091] Specifically, the target association information can represent the association between word feature information and knowledge point feature information.

[0092] In a specific embodiment, the association analysis layer may include an association calculation function, which inputs word feature information and knowledge point feature information into the association calculation function to perform association calculation to obtain target association information. Optionally, the association calculation function may be: S(x i ,y)=f(W h ·x i +W v y+b), optionally, f() can be a hyperbolic tangent function,

[0093] Correspondingly, when the word feature information is a word feature vector and the knowledge point feature information is a knowledge point feature vector, the target association information can be: S(h i ,g)=f(W h ·h i +W v g+b), where h i represents the word feature vector of the i-th word segmentation data in multiple word segmentation data, g represents the knowledge point feature vector, W h represents the first alignment matrix, W v represents the second alignment matrix, b represents the offset vector, and f() represents the activation function. In practical applications, W h 、W v and b can be obtained after the training of the second preset fusion network is completed.

[0094] S702: Input the target association information into a normalization layer for normalization processing to obtain the association weight of the word feature information.

[0095] In a specific embodiment, the normalization layer may include a normalization function, and the target association information is input into the normalization function for normalization calculation to obtain the association weight. Optionally, the normalization function may be a softmax function (normalized exponential function).

[0096] S703: Input the association weight and word feature information into the weighting layer for weighted processing to obtain initial fusion information.

[0097] In a specific embodiment, the product of the word feature vector of each word in the plurality of word segmentation data and the corresponding associated weight is added together to obtain an initial fusion vector, and the initial fusion vector is used as the initial fusion information.

[0098] It can be seen from the above embodiments that the fusion network including the association analysis layer, the normalization layer and the weighting layer captures the association relationship between the test text data and the knowledge point structure tree features, thereby effectively embedding the structured semantic features of the knowledge point structure tree into the test text expression, thereby improving the accuracy of the representation of the test text features.

[0099] S602: Input the initial fusion information, text feature information and knowledge point feature information into the splicing network for splicing processing to obtain target fusion information.

[0100] In a specific embodiment, the splicing network can be obtained by splicing and training a preset splicing network. Specifically, when the initial fusion information is an initial fusion vector, the text feature information is a text feature vector, and the knowledge point feature information is a knowledge point feature vector, the splicing network can include a weight matrix and an activation function. The initial fusion vector, the text feature vector, and the knowledge point feature vector are spliced ​​together to obtain a splicing matrix. The product of the weight matrix and the splicing matrix is ​​input into the activation function to obtain target fusion information, and the target fusion vector is used as the target fusion information. Optionally, the activation function can be a hyperbolic tangent function.

[0101] It can be seen from the above embodiments that the initial fusion information, text feature information and knowledge point feature information are spliced ​​and processed, and the features of the test text data and the knowledge point structure tree are further fused to obtain the target fusion information, so as to facilitate the subsequent use of the deep semantic association between the test text and the knowledge points in the target fusion information and the hierarchical dependency relationship between the knowledge points to perform knowledge point annotation, thereby greatly improving the accuracy of knowledge point annotation for test questions.

[0102] S205: Input the target fusion information into the knowledge point recognition network to perform knowledge point annotation, and obtain the annotated knowledge point information corresponding to the test question text data.

[0103] In an embodiment of the present application, the annotated knowledge point information may be identification information of a target knowledge point corresponding to the test question text data. Specifically, the target knowledge point may be at least one knowledge point corresponding to the test question text data among multiple knowledge points in the knowledge point structure tree.

[0104] In a specific embodiment, the annotated knowledge point information can be a target knowledge point tag, and the target knowledge point tag can be at least one knowledge point tag in a preset knowledge point tag set corresponding to the test question text data, wherein the preset knowledge point tag set can be set in combination with multiple knowledge points in the knowledge point structure tree in actual applications.

[0105] In a specific embodiment, the knowledge point recognition network can be obtained by performing knowledge point annotation training on a preset knowledge point recognition network. Specifically, the preset knowledge point recognition network can include a fully connected layer and an output layer.

[0106] Specifically, the fully connected layer can serve as a connection layer between the nodes of the upper and lower layers, establishing a connection relationship between the node data obtained from the upper and lower layers. The fully connected layer can perform feature aggregation processing on the target fusion information to obtain the information of the test questions to be labeled.

[0107] In a specific embodiment, the information of the test questions to be marked can be expressed in the form of a test question vector to be marked, the dimension of the test question vector to be marked is the same as the number of knowledge point labels in the preset knowledge point label set, and the multiple dimensions of the test question vector to be marked correspond one-to-one to the multiple knowledge point labels in the preset knowledge point label set. Each dimension of the test question vector to be marked represents the correlation between the test question text data and the corresponding knowledge point label.

[0108] In the aforementioned pre-set knowledge point recognition network, each layer outputs a linear function of the previous layer's input. However, given that data in real applications is often not linearly separable, a nonlinear factor can be introduced by adding an activation function. This involves adding a linear correction layer, using the activation function Relu (rectified linear unit) as an example. Specifically, the Relu layer corrects the previous layer's data output, converting all inputs less than 0 to 0 before outputting them, while leaving any outputs greater than 0 unchanged.

[0109] Specifically, the output layer can perform knowledge point labeling on the question vector to be labeled and output the corresponding target knowledge point label. In a specific embodiment, the output layer can use a Sigmoid function to output the target knowledge point label. The Sigmoid function contains a nonlinear classifier to perform knowledge point labeling on the question vector to be labeled. Specifically, the output layer can perform probability prediction on each dimension of the question vector to be labeled and obtain the probability value of each dimension. The knowledge point label corresponding to the dimension whose probability value is greater than the preset probability threshold is used as the target knowledge point label.

[0110] In addition, it should be noted that the knowledge point recognition network described in the embodiment of the present application is not limited to the above-mentioned preset knowledge point recognition network. In practical applications, it can also include other machine learning networks, such as decision tree machine learning networks, etc. The embodiment of the present application is not limited to the above-mentioned machine learning networks.

[0111] In an optional embodiment, the annotated knowledge point information may be the final-level knowledge point information corresponding to any terminal node knowledge point in the knowledge point structure tree. After inputting the target fusion information into the knowledge point recognition network for knowledge point annotating and obtaining the annotated knowledge point information corresponding to the test question text data, the method may further include:

[0112] Acquire knowledge point mapping information, where the knowledge point mapping information represents a mapping relationship between final-level knowledge point information and multi-level knowledge point information corresponding to the final-level knowledge point information;

[0113] Based on the knowledge point mapping information, target multi-level knowledge point information corresponding to the annotated knowledge point information is determined.

[0114] In a specific embodiment, the final-level knowledge point information may include a final-level knowledge point label, and the multi-level knowledge point information corresponding to the final-level knowledge point information may include a multi-level knowledge point label corresponding to the final-level knowledge point label. Accordingly, the knowledge point mapping information may represent a mapping relationship between the final-level knowledge point label and the multi-level knowledge point label corresponding to the final-level knowledge point label. Specifically, the final-level knowledge point label may be a knowledge point label corresponding to any final node knowledge point in the knowledge point structure tree, and the multi-level knowledge point label may be a multi-level knowledge point label corresponding to the knowledge point path from the corresponding final node knowledge point to the root node knowledge point.

[0115] In a specific embodiment, the knowledge point mapping information may include but is not limited to a knowledge point mapping table and a knowledge point mapping dictionary. Figure 3 Taking the knowledge point structure tree shown in FIG. 1 as an example, a knowledge point mapping table as shown in Table 1 is generated.

[0116] Final knowledge point label Multi-level knowledge point labels Finding the arithmetic mean Sequence - Arithmetic Sequence - Mean of Arithmetic Sequence - Finding the Arithmetic Mean Application of Arithmetic Mean Sequence-Arithmetic Sequence-Mean of Arithmetic Sequence-Application of Arithmetic Mean Properties of arithmetic progression Sequence-Arithmetic Sequence-Properties of Arithmetic Sequence Definition of the limit of a sequence Sequence - Limit of a Sequence - Definition of the Limit of a Sequence Limit of a convergent sequence Sequence-Limit of Sequence-Limit of Convergent Sequence Finding the Mean of a Geometric Proportion Sequence - Geometric Sequence - Mean of Geometric Sequence - Finding the Geometric Mean Application of Geometric Mean Sequence-Geometric Sequence-Mean of Geometric Sequence-Application of Geometric Mean Properties of geometric sequences Sequence-Geometric Sequence-Properties of Geometric Sequence

[0117] Table 1

[0118] In an optional embodiment, when the annotated knowledge point information includes multiple final-level knowledge point tags, the above-mentioned determination of the target multi-level knowledge point information corresponding to the annotated knowledge point information based on the knowledge point mapping information may include: determining multiple target multi-level knowledge point tags corresponding to multiple final-level knowledge point tags based on the knowledge point mapping information; after determining multiple target multi-level knowledge point tags corresponding to multiple final-level knowledge point tags based on the knowledge point mapping information, a target knowledge point tag tree may also be generated based on the multiple target multi-level knowledge point tags.

[0119] It can be seen from the above embodiments that the final-level knowledge points, i.e., the most refined knowledge points, of the test paper text data are annotated using the knowledge point recognition network, and then the target multi-level knowledge point information corresponding to the annotated knowledge point information is determined based on the mapping relationship between the final-level knowledge point information and the multi-level knowledge point information corresponding to the final-level knowledge point information, thereby improving the accuracy of knowledge point annotation.

[0120] In an embodiment of the present application, the above-mentioned semantic recognition network, the above-mentioned feature recognition network, the above-mentioned fusion network, the above-mentioned splicing network and the above-mentioned knowledge point recognition network can be obtained by jointly training the preset semantic recognition network, the preset feature recognition network, the preset fusion network, the preset splicing network and the preset knowledge point recognition network.

[0121] In a specific embodiment, Figure 8 As shown, Figure 8 This is a flowchart of a network joint training method provided by an embodiment of the present application, which may specifically include:

[0122] S801, obtaining sample test question text data and preset annotated knowledge point information corresponding to the sample test question text data.

[0123] In practical applications, before conducting network joint training, training data may be determined first. Specifically, in an embodiment of the present application, sample test text data containing preset annotated knowledge point information may be obtained as training data.

[0124] Specifically, the preset annotated knowledge point information may be preset knowledge point labels that are pre-annotated on the sample test question text data.

[0125] S802 , inputting the sample test question text data into a preset semantic recognition network for semantic extraction, and obtaining sample text feature information of the sample test question text data and sample word feature information of a plurality of sample word segmentation data of the sample test question text data.

[0126] S803: Based on a preset feature recognition network, feature extraction is performed on the knowledge point structure tree to obtain feature information of sample knowledge points of the knowledge point structure tree.

[0127] S804: Input the sample knowledge point feature information and the sample word feature information into a preset fusion network for fusion processing to obtain sample initial fusion information.

[0128] S805 , inputting the sample initial fusion information, the sample text feature information, and the sample knowledge point feature information into a preset splicing network for splicing processing to obtain sample target fusion information.

[0129] S806: Input the sample target fusion information into a preset knowledge point recognition network to perform knowledge point annotation, and obtain sample annotated knowledge point information corresponding to the sample test question text data.

[0130] S807: Determine target loss information based on the preset labeled knowledge point information and the sample labeled knowledge point information.

[0131] S808, based on the target loss information, train a preset semantic recognition network, a preset feature recognition network, a preset fusion network, a preset splicing network and a preset knowledge point recognition network to obtain a semantic recognition network, a feature recognition network, a fusion network, a splicing network and a knowledge point recognition network.

[0132] In an optional embodiment, the sample annotated knowledge point information may include sample knowledge point labels of sample test question text data; correspondingly, the target loss information may include knowledge point label loss;

[0133] Accordingly, the above-mentioned determination of target loss information based on the preset annotated knowledge point information and the sample annotated knowledge point information may include:

[0134] Determine the knowledge point label loss based on the preset knowledge point labels and the sample knowledge point labels.

[0135] In a specific embodiment, determining the knowledge point label loss based on the preset knowledge point label and the sample knowledge point label may include determining the knowledge point label loss between the preset knowledge point label and the sample knowledge point label based on a preset loss function.

[0136] In a specific embodiment, the knowledge point label loss may represent the difference between the preset knowledge point label and the sample knowledge point label.

[0137] In a specific embodiment, the preset loss function may include but is not limited to a cross entropy loss function, a logistic loss function, an exponential loss function, and the like.

[0138] In an optional embodiment, based on the target loss information, training a preset semantic recognition network, a preset feature recognition network, a preset fusion network, a preset splicing network and a preset knowledge point recognition network to obtain a semantic recognition network, a feature recognition network, a fusion network, a splicing network and a knowledge point recognition network may include: updating the network parameters of the preset semantic recognition network, the preset feature recognition network, the preset fusion network, the preset splicing network and the preset knowledge point recognition network based on the target loss information; repeating step S802 to updating the preset semantic recognition network, the preset feature recognition network, the preset fusion network, the preset splicing network and the preset knowledge point recognition network based on the updated preset semantic recognition network, the preset feature recognition network, the preset fusion network, the preset splicing network and the preset knowledge point recognition network based on the target loss information. The preset semantic recognition network, preset feature recognition network, preset fusion network, preset splicing network and preset knowledge point recognition network are trained iteratively for knowledge point labeling, and the network parameters of the preset semantic recognition network, preset feature recognition network, preset fusion network, preset splicing network and preset knowledge point recognition network are updated based on the target loss information, until the knowledge point labeling convergence conditions are reached; the preset semantic recognition network, preset feature recognition network, preset fusion network, preset splicing network and preset knowledge point recognition network obtained when the knowledge point labeling convergence conditions are reached are used as the semantic recognition network, feature recognition network, fusion network, splicing network and knowledge point recognition network.

[0139] In an optional embodiment, the above-mentioned updating of the network parameters of the preset semantic recognition network, the preset feature recognition network, the preset fusion network, the preset splicing network and the preset knowledge point recognition network based on the target loss information can be performed by setting independent adaptive learning rates for different network parameters by calculating the first-order moment estimation and the second-order moment estimation of the gradient. Optionally, the initial learning rate can be set to 0.000125, and the size of each batch of training samples can be 32.

[0140] In an optional embodiment, the above-mentioned condition for reaching knowledge point labeling convergence can be that the number of training iterations reaches a preset number of training times. Alternatively, the condition for reaching knowledge point labeling convergence can also be that the target loss information is less than a specified threshold. In the embodiments of this specification, the preset number of training times and the specified threshold can be pre-set based on the actual application requirements for network training speed and accuracy.

[0141] It can be seen from the above embodiments that by jointly training the preset semantic recognition network, the preset feature recognition network, the preset fusion network, the preset splicing network and the preset knowledge point recognition network, the training efficiency can be improved while the accuracy of the network's knowledge point annotation can be better improved.

[0142] In a specific embodiment, Figure 9As shown, a knowledge point annotation network is established which includes the above-mentioned semantic recognition network, the above-mentioned feature recognition network, the above-mentioned fusion network, the above-mentioned preset splicing network and the above-mentioned knowledge point recognition network. The test text data and the knowledge point structure tree are input into the knowledge point annotation network for knowledge point annotation to obtain the annotated knowledge point information corresponding to the test text data.

[0143] It can be seen from the technical solutions provided by the above embodiments of the present application that, in the scenario of knowledge point annotation of test question text, the present application, on the one hand, performs contextual semantic recognition on the test question text data through a semantic recognition network including a word segmentation layer, a semantic recognition layer and a semantic fusion layer, realizes the fusion of contextual semantic features of the test question text data, and thus can improve the accuracy of the representation of the test question semantic features; on the other hand, performs feature extraction on the knowledge point structure tree through a feature extraction network including multiple feature aggregation layers corresponding one-to-one to multiple nodes in the knowledge point structure tree, thereby aggregating the structural hierarchical relationship and semantic features of the knowledge point structure tree from the end node to the root node, and improving the accuracy of the representation of the knowledge point structure tree features; on the other hand, the features of the knowledge point structure tree and the test question text data are aggregated twice through the fusion network and the splicing network, thereby further improving the accuracy of the representation of the test question features by the target fusion information; on the other hand, the deep semantic association between the test question text and the knowledge points in the target fusion information and the hierarchical dependency relationship between the knowledge points are used to identify the identification points, which greatly improves the accuracy and effectiveness of knowledge point annotation of the test questions.

[0144] The present application also provides a knowledge point marking device, such as Figure 10 As shown, the knowledge point marking device may include:

[0145] The test question text data acquisition module 1010 is used to acquire the test question text data and the knowledge point structure tree;

[0146] Semantic recognition module 1020, used to input the test question text data into the semantic recognition network for semantic recognition, and obtain text feature information of the test question text data and word feature information of multiple word segmentation data of the test question text data;

[0147] A feature recognition module 1030 is configured to perform feature recognition on the knowledge point structure tree based on a feature recognition network to obtain knowledge point feature information of the knowledge point structure tree, wherein the knowledge point feature information represents the structural semantic features of the knowledge point structure tree;

[0148] A fusion processing module 1040 is used to fuse the knowledge point feature information, word feature information and text feature information to obtain target fusion information;

[0149] The knowledge point annotation module 1050 is used to input the target fusion information into the knowledge point recognition network to perform knowledge point annotation, and obtain the annotated knowledge point information corresponding to the test question text data.

[0150] In a specific embodiment, the semantic recognition network may include a word segmentation layer, a semantic recognition layer, and a semantic fusion layer. The semantic recognition module 1020 may include:

[0151] The word segmentation unit is used to input the test text data into the word segmentation layer for word segmentation preprocessing to obtain multiple word segmentation data;

[0152] The first semantic recognition unit is used to input the multiple word segmentation data into the semantic recognition layer for semantic recognition, and obtain the initial semantic information, position information and sentence information corresponding to the multiple word segmentation data;

[0153] The semantic fusion unit is used to input the initial semantic information, position information and sentence information into the semantic fusion layer to perform contextual semantic fusion on multiple word segmentation data to obtain text feature information and word feature information.

[0154] In a specific embodiment, the feature recognition network may include multiple feature aggregation layers corresponding one-to-one to multiple nodes in the knowledge point structure tree. The feature recognition module 1030 may include:

[0155] The second semantic recognition is used to perform semantic recognition on the knowledge point data of each node in the multiple nodes to obtain the node semantic information corresponding to each node;

[0156] The first feature aggregation unit is used to input the node semantic information corresponding to the end node of the knowledge point structure tree into the feature aggregation layer corresponding to the end node to obtain the node aggregation feature information corresponding to the end node;

[0157] The traversal unit is used to traverse multiple nodes starting from the parent node of the last node;

[0158] The second feature aggregation unit is used to input the node aggregation feature information corresponding to the child nodes of the currently traversed node and the node semantic information corresponding to the currently traversed node into the feature aggregation layer corresponding to the currently traversed node to obtain the node aggregation feature information corresponding to the currently traversed node;

[0159] The knowledge point feature information unit is used to aggregate feature information of nodes corresponding to the root node in the knowledge point structure tree as knowledge point feature information after traversing multiple nodes.

[0160] In a specific embodiment, the fusion processing module 1040 may include:

[0161] A fusion processing unit, configured to input the knowledge point feature information and the word feature information into a fusion network for fusion processing to obtain initial fusion information;

[0162] The splicing processing unit is used to input the initial fusion information, text feature information and knowledge point feature information into the splicing network for splicing processing to obtain target fusion information.

[0163] In a specific embodiment, the fusion network may include an association analysis layer, a normalization layer, and a weighting layer. The fusion processing unit may include:

[0164] The association analysis unit is used to input word feature information and knowledge point feature information into the association analysis layer for association analysis to obtain target association information;

[0165] The normalization unit is used to input the target association information into the normalization layer for normalization processing to obtain the association weight of the word feature information;

[0166] The weighting unit is used to input the association weight and word feature information into the weighting layer for weighted processing to obtain initial fusion information.

[0167] In a specific embodiment, the above-mentioned knowledge point marking device may further include:

[0168] A sample acquisition module is used to obtain sample test question text data and preset annotated knowledge point information corresponding to the sample test question text data;

[0169] A sample semantic extraction module is used to input the sample test question text data into a preset semantic recognition network for semantic extraction, and obtain sample text feature information of the sample test question text data and sample word feature information of multiple sample word segmentation data of the sample test question text data;

[0170] The sample feature recognition module is used to extract features from the knowledge point structure tree based on a preset feature recognition network to obtain sample knowledge point feature information of the knowledge point structure tree;

[0171] The sample fusion processing module is used to input the sample knowledge point feature information and the sample word feature information into the preset fusion network for fusion processing to obtain the sample initial fusion information;

[0172] The sample splicing processing module is used to input the sample initial fusion information, sample text feature information and sample knowledge point feature information into the preset splicing network for splicing processing to obtain sample target fusion information;

[0173] The sample knowledge point annotation module is used to input the sample target fusion information into the preset knowledge point recognition network for knowledge point annotation, and obtain the sample annotated knowledge point information corresponding to the sample test question text data;

[0174] A target loss information determination module is used to determine target loss information based on preset annotated knowledge point information and sample annotated knowledge point information;

[0175] The network training module is used to train a preset semantic recognition network, a preset feature recognition network, a preset fusion network, a preset splicing network and a preset knowledge point recognition network based on target loss information to obtain a semantic recognition network, a feature recognition network, a fusion network, a splicing network and a knowledge point recognition network.

[0176] It should be noted that the device and method embodiments in the device embodiment are based on the same inventive concept.

[0177] An embodiment of the present application provides a knowledge point annotation device, which includes a processor and a memory, wherein the memory stores at least one instruction or at least one program, and the at least one instruction or the at least one program is loaded and executed by the processor to implement the knowledge point annotation method provided in the above method embodiment.

[0178] Furthermore, Figure 11 A schematic diagram of the hardware structure of a knowledge point annotation device for implementing the knowledge point annotation method provided in the embodiment of the present application is shown. The knowledge point annotation device can participate in or include the knowledge point annotation apparatus provided in the embodiment of the present application. Figure 11 As shown, the knowledge point annotation device 110 may include one or more (1102a, 1102b, ..., 1102n are used to illustrate) processors 1102 (the processor 1102 may include but is not limited to a processing device such as a microprocessor MCU or a programmable logic device FPGA), a memory 1104 for storing data, and a transmission device 1106 for communication functions. In addition, it may also include: a display, an input / output interface (I / O interface), a universal serial bus (USB) port (which may be included as one of the ports of the I / O interface), a network interface, a power supply and / or a camera. It will be understood by those skilled in the art that Figure 11 The structure shown is only for illustration and does not limit the structure of the above electronic device. Figure 11 More or fewer components than shown, or with Figure 11 Different configurations shown.

[0179] It should be noted that the one or more processors 1102 and / or other data processing circuits described above may generally be referred to herein as "data processing circuits". The data processing circuit may be embodied in whole or in part as software, hardware, firmware, or any other combination thereof. In addition, the data processing circuit may be a single independent processing module, or may be incorporated in whole or in part into any of the other components in the knowledge point annotation device 110 (or mobile device). As described in the embodiments of the present application, the data processing circuit serves as a processor control (e.g., selection of a variable resistor terminal path connected to an interface).

[0180] The memory 1104 can be used to store software programs and modules of application software, such as the program instructions / data storage device corresponding to the knowledge point annotation method described in the embodiment of the present application. The processor 1102 executes various functional applications and data processing by running the software programs and modules stored in the memory 1104, that is, realizes the above-mentioned knowledge point annotation method. The memory 1104 may include a high-speed random access memory, and may also include a non-volatile memory, such as one or more magnetic storage devices, flash memory, or other non-volatile solid-state memory. In some instances, the memory 1104 may further include a memory remotely arranged relative to the processor 1102, and these remote memories can be connected to the knowledge point annotation device 110 via a network. Examples of the above-mentioned network include but are not limited to the Internet, corporate intranet, local area network, mobile communication network and combinations thereof.

[0181] The transmission device 1106 is used to receive or send data via a network. Specific examples of the aforementioned network may include a wireless network provided by a communication provider of the knowledge point annotation device 110. In one embodiment, the transmission device 1106 includes a network adapter (Network Interface Controller, NIC), which can be connected to other network devices via a base station to communicate with the Internet. In one embodiment, the transmission device 1106 can be a radio frequency (RF) module, which is used to communicate with the Internet wirelessly.

[0182] The display may be, for example, a touch screen liquid crystal display (LCD), which may enable a user to interact with a user interface of the knowledge point annotation device 110 (or mobile device).

[0183] An embodiment of the present application also provides a computer-readable storage medium, which can be set in a knowledge point annotation device to store at least one instruction or at least one program related to the knowledge point annotation method in the method embodiment. The at least one instruction or the at least one program is loaded and executed by the processor to implement the knowledge point annotation method provided by the above method embodiment.

[0184] Optionally, in this embodiment, the storage medium may be located in at least one of a plurality of network servers in a computer network. Optionally, in this embodiment, the storage medium may include, but is not limited to, various media capable of storing program code, such as a USB flash drive, a read-only memory (ROM), a random access memory (RAM), a mobile hard disk, a magnetic disk, or an optical disk.

[0185] The embodiments of the present application also provide a computer program product or computer program, which includes computer instructions stored in a computer-readable storage medium. A processor of a computer device reads the computer instructions from the computer-readable storage medium and executes the computer instructions, causing the computer device to perform the knowledge point annotation method provided in the method embodiment. Optionally, the computer program product may include, but is not limited to, a recording and arrangement system for question bank construction, a test question recommendation system, and a personalized learning system.

[0186] It should be noted that the order of the embodiments of the present application described above is for descriptive purposes only and does not represent the superiority or inferiority of the embodiments. The above description is of specific embodiments of the present application. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps described in the claims can be performed in an order different from that in the embodiments and still achieve the desired results. In addition, the processes depicted in the accompanying drawings do not necessarily require the specific order or continuous order shown to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0187] The various embodiments in this application are described in a progressive manner. Similar parts between the various embodiments can be referred to in conjunction with each other. Each embodiment focuses on the differences between the other embodiments. In particular, the device and apparatus embodiments are generally similar to the method embodiments, so the description is relatively simple. For relevant parts, refer to the partial description of the method embodiments.

[0188] Those skilled in the art will understand that all or part of the steps to implement the above embodiments may be accomplished by hardware, or by a program to instruct the relevant hardware, and the program may be stored in a computer-readable storage medium, which may be a read-only memory, a disk, or an optical disk, etc.

[0189] The above description is only a preferred embodiment of the present application and is not intended to limit the present application. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principles of the present application should be included in the scope of protection of the present application.

Claims

1. A knowledge point annotation method, characterized in that: The method comprises: Obtaining the test question text data corresponding to the target subject and the knowledge point structure tree matching the target subject; Inputting the test question text data into a semantic recognition network for semantic recognition, and obtaining text feature information of the test question text data and word feature information of a plurality of word segmentation data of the test question text data; Based on a feature recognition network, feature recognition is performed on the knowledge point structure tree to obtain knowledge point feature information of the knowledge point structure tree, wherein the knowledge point feature information represents the structured semantic features of the knowledge point structure tree of the target subject; Fusing the knowledge point feature information, the word feature information, and the text feature information to obtain target fusion information; The target fusion information is input into the knowledge point recognition network for knowledge point annotation to obtain the annotated knowledge point information corresponding to the test question text data. The annotated knowledge point information is used to identify at least one knowledge point corresponding to the test question text data in the knowledge point structure tree.

2. The method according to claim 1, characterized in that The feature recognition network includes multiple feature aggregation layers corresponding to multiple nodes in the knowledge point structure tree. The feature recognition of the knowledge point structure tree based on the feature recognition network is performed to obtain knowledge point feature information of the knowledge point structure tree, including: Performing semantic recognition on the knowledge point data of each of the multiple nodes to obtain node semantic information corresponding to each node; Inputting the node semantic information corresponding to the end node of the knowledge point structure tree into the feature aggregation layer corresponding to the end node to obtain the node aggregation feature information corresponding to the end node; Starting from the parent node of the last node, traverse the multiple nodes; Inputting node aggregation feature information corresponding to the child nodes of the currently traversed node and node semantic information corresponding to the currently traversed node into the feature aggregation layer corresponding to the currently traversed node to obtain node aggregation feature information corresponding to the currently traversed node; After traversing the plurality of nodes, the node aggregation feature information corresponding to the root node in the knowledge point structure tree is used as the knowledge point feature information.

3. The method according to claim 1, characterized in that The semantic recognition network includes a word segmentation layer, a semantic recognition layer, and a semantic fusion layer. The test question text data is input into the semantic recognition network for semantic recognition, and the text feature information of the test question text data and the word feature information of the multiple word segmentation data of the test question text data are obtained. Inputting the test question text data into the word segmentation layer for word segmentation preprocessing to obtain the plurality of word segmentation data; Inputting the plurality of segmentation data into the semantic recognition layer for semantic recognition, and obtaining initial semantic information, position information, and sentence information corresponding to the plurality of segmentation data; The initial semantic information, the position information and the sentence information are input into the semantic fusion layer to perform contextual semantic fusion on the multiple word segmentation data to obtain the text feature information and the word feature information.

4. The method according to claim 1, wherein The fusing of the knowledge point feature information, the word feature information and the text feature information to obtain target fusion information includes: Inputting the knowledge point feature information and the word feature information into a fusion network for fusion processing to obtain initial fusion information; The initial fusion information, the text feature information and the knowledge point feature information are input into a splicing network for splicing processing to obtain the target fusion information.

5. The method according to claim 4, characterized in that The fusion network includes an association analysis layer, a normalization layer, and a weighting layer. The knowledge point feature information and the word feature information are input into the fusion network for fusion processing to obtain initial fusion information including: Inputting the word feature information and the knowledge point feature information into the association analysis layer for association analysis to obtain target association information; Inputting the target association information into the normalization layer for normalization processing to obtain the association weight of the word feature information; The association weight and the word feature information are input into the weighting layer for weighted processing to obtain the initial fusion information.

6. The method according to claim 4, characterized in that The method further comprises: Obtaining sample test question text data and preset annotated knowledge point information corresponding to the sample test question text data; Inputting the sample test question text data into a preset semantic recognition network for semantic extraction to obtain sample text feature information of the sample test question text data and sample word feature information of a plurality of sample word segmentation data of the sample test question text data; Based on a preset feature recognition network, feature extraction is performed on the knowledge point structure tree to obtain sample knowledge point feature information of the knowledge point structure tree; Inputting the sample knowledge point feature information and the sample word feature information into a preset fusion network for fusion processing to obtain sample initial fusion information; Inputting the sample initial fusion information, the sample text feature information and the sample knowledge point feature information into a preset splicing network for splicing processing to obtain sample target fusion information; Inputting the sample target fusion information into a preset knowledge point recognition network for knowledge point annotation to obtain sample annotated knowledge point information corresponding to the sample test question text data; Determining target loss information based on the preset annotated knowledge point information and the sample annotated knowledge point information; Based on the target loss information, the preset semantic recognition network, the preset feature recognition network, the preset fusion network, the preset splicing network and the preset knowledge point recognition network are trained to obtain the semantic recognition network, the feature recognition network, the fusion network, the splicing network and the knowledge point recognition network.

7. The method according to any one of claims 1 to 5, characterized in that: The annotated knowledge point information is the final-level knowledge point information corresponding to any terminal node knowledge point in the knowledge point structure tree. After inputting the target fusion information into the knowledge point recognition network for knowledge point annotating to obtain the annotated knowledge point information corresponding to the test question text data, the method further includes: Acquire knowledge point mapping information, wherein the knowledge point mapping information represents a mapping relationship between the final-level knowledge point information and the multi-level knowledge point information corresponding to the final-level knowledge point information; Based on the knowledge point mapping information, target multi-level knowledge point information corresponding to the annotated knowledge point information is determined.

8. A knowledge point annotation device, characterized in that the device comprises: A test question text data acquisition module is used to acquire the test question text data corresponding to the target subject and the knowledge point structure tree matching the target subject; A semantic recognition module is used to input the test question text data into a semantic recognition network for semantic recognition, and obtain text feature information of the test question text data and word feature information of a plurality of word segmentation data of the test question text data; A feature recognition module is used to perform feature recognition on the knowledge point structure tree based on a feature recognition network to obtain knowledge point feature information of the knowledge point structure tree, wherein the knowledge point feature information represents the structured semantic features of the knowledge point structure tree of the target subject; A fusion processing module, configured to fuse the knowledge point feature information, the word feature information, and the text feature information to obtain target fusion information; The knowledge point annotation module is used to input the target fusion information into the knowledge point recognition network for knowledge point annotation to obtain the annotated knowledge point information corresponding to the test question text data. The annotated knowledge point information is used to identify at least one knowledge point corresponding to the test question text data in the knowledge point structure tree.

9. The device according to claim 8, characterized in that The feature recognition network includes multiple feature aggregation layers corresponding one-to-one to multiple nodes in the knowledge point structure tree, and the feature recognition module includes: Second semantic recognition, for performing semantic recognition on the knowledge point data of each of the multiple nodes to obtain node semantic information corresponding to each node; A first feature aggregation unit is configured to input node semantic information corresponding to the end node of the knowledge point structure tree into a feature aggregation layer corresponding to the end node to obtain node aggregation feature information corresponding to the end node; A traversal unit, configured to traverse the plurality of nodes starting from the parent node of the last node; A second feature aggregation unit is configured to input node aggregation feature information corresponding to a child node of a currently traversed node and node semantic information corresponding to the currently traversed node into a feature aggregation layer corresponding to the currently traversed node to obtain node aggregation feature information corresponding to the currently traversed node; The knowledge point feature information unit is configured to, after traversing the plurality of nodes, aggregate feature information of the nodes corresponding to the root node in the knowledge point structure tree as the knowledge point feature information.

10. The device according to claim 8, characterized in that The semantic recognition network includes a word segmentation layer, a semantic recognition layer and a semantic fusion layer, and the semantic recognition module includes: A word segmentation unit, configured to input the test question text data into the word segmentation layer for word segmentation preprocessing to obtain the plurality of word segmentation data; A first semantic recognition unit is configured to input the plurality of segmented word data into the semantic recognition layer for semantic recognition, and obtain initial semantic information, position information, and sentence information corresponding to the plurality of segmented word data; The semantic fusion unit is used to input the initial semantic information, the position information and the sentence information into the semantic fusion layer to perform contextual semantic fusion on the multiple word segmentation data to obtain the text feature information and the word feature information.

11. The device according to claim 8, characterized in that The fusion processing module includes: a fusion processing unit, configured to input the knowledge point feature information and the word feature information into a fusion network for fusion processing to obtain initial fusion information; The splicing processing unit is used to input the initial fusion information, the text feature information and the knowledge point feature information into a splicing network for splicing processing to obtain the target fusion information.

12. The device according to claim 11, characterized in that The fusion network includes an association analysis layer, a normalization layer, and a weighting layer, and the fusion processing unit includes: An association analysis unit, configured to input the word feature information and the knowledge point feature information into the association analysis layer for association analysis to obtain target association information; a normalization unit, configured to input the target association information into the normalization layer for normalization processing to obtain an association weight of the word feature information; The weighting unit is used to input the association weight and the word feature information into the weighting layer for weighted processing to obtain the initial fusion information.

13. The device according to claim 11, characterized in that The device further comprises: A sample acquisition module is used to obtain sample test question text data and preset annotated knowledge point information corresponding to the sample test question text data; A sample semantic extraction module is used to input the sample test question text data into a preset semantic recognition network for semantic extraction, and obtain sample text feature information of the sample test question text data and sample word feature information of a plurality of sample word segmentation data of the sample test question text data; A sample feature recognition module is used to extract features from the knowledge point structure tree based on a preset feature recognition network to obtain sample knowledge point feature information of the knowledge point structure tree; A sample fusion processing module is used to input the sample knowledge point feature information and the sample word feature information into a preset fusion network for fusion processing to obtain sample initial fusion information; A sample splicing processing module is used to input the sample initial fusion information, the sample text feature information and the sample knowledge point feature information into a preset splicing network for splicing processing to obtain sample target fusion information; A sample knowledge point annotation module is used to input the sample target fusion information into a preset knowledge point recognition network for knowledge point annotation, and obtain sample annotated knowledge point information corresponding to the sample test question text data; a target loss information determination module, configured to determine target loss information based on the preset annotated knowledge point information and the sample annotated knowledge point information; A network training module is used to train the preset semantic recognition network, the preset feature recognition network, the preset fusion network, the preset splicing network and the preset knowledge point recognition network based on the target loss information to obtain the semantic recognition network, the feature recognition network, the fusion network, the splicing network and the knowledge point recognition network.

14. The device according to any one of claims 8 to 12, characterized in that The annotated knowledge point information is the last-level knowledge point information corresponding to any last-node knowledge point in the knowledge point structure tree, and the device further includes: A knowledge point mapping information acquisition module is used to acquire knowledge point mapping information, wherein the knowledge point mapping information represents a mapping relationship between the final-level knowledge point information and the multi-level knowledge point information corresponding to the final-level knowledge point information; The target multi-level knowledge point information determination module is used to determine the target multi-level knowledge point information corresponding to the annotated knowledge point information based on the knowledge point mapping information.

15. A knowledge point annotation device, characterized in that: The device includes a processor and a memory, wherein the memory stores at least one instruction or at least one program, and the at least one instruction or the at least one program is loaded and executed by the processor to implement the knowledge point annotation method according to any one of claims 1 to 7.

16. A computer-readable storage medium, characterized in that The storage medium stores at least one instruction or at least one program, and the at least one instruction or the at least one program is loaded and executed by the processor to implement the knowledge point annotation method according to any one of claims 1 to 7.

17. A computer program product, characterized in that The computer program product includes at least one instruction or at least one program segment, and the at least one instruction or the at least one program segment is loaded and executed by a processor to implement the knowledge point labeling method according to any one of claims 1 to 7.

Images