Computer-implemented method

Abstract

Provided in the embodiments of the present disclosure are a computer-implemented method, a computing system, a computer-readable storage medium and a computer program product. The method comprises: constructing a reference item bank, wherein the reference item bank utilizes a hypergraph structure, the hypergraph comprises nodes and hyperedges, the nodes represent reference items, the attributes of the nodes include questions and reference answers of corresponding reference items, and the hyperedges represent knowledge points corresponding to the reference items represented by the connected nodes. The solution can address the actual requirements of a user related to test items and knowledge points, and provides the user with a "request-response" mechanism based on the reference item bank.

Description

A computer-implemented method Technical Field

[0001] This disclosure relates to the field of artificial intelligence technology, and more particularly to a computer-implemented method, computing system, computer program product, and computer storage medium. Background Technology

[0002] Learning science subjects (including mathematics, physics, chemistry, biology, etc.), engineering, and even humanities subjects (such as law) usually requires students to do a certain number of exercises to master the knowledge of the subject. Therefore, it is necessary to establish a question bank suitable for students to use and meet their various learning requirements for these subjects.

[0003] With the development of computer, network, and artificial intelligence (AI) technologies, and the increasing abundance of e-learning resources, it is necessary to establish a question bank based on computer, network, and AI technologies, and develop corresponding applications. Such methods and systems can provide a new learning model, meet students' personalized learning needs, and promote student autonomy in learning. Summary of the Invention

[0004] Embodiments of this disclosure provide a computer system, corresponding method, computer-readable storage medium, and computer program product based on a reference test question bank.

[0005] In a first aspect of this disclosure, a computer-implemented method is provided, comprising: constructing a reference question bank, wherein the structure of the reference question bank adopts a hypergraph, the hypergraph includes nodes and hyperedges, the nodes represent reference questions, the attributes of the nodes include the question and the reference answer of the corresponding reference question, and the hyperedges represent the knowledge points corresponding to the reference questions represented by the connected nodes.

[0006] In embodiments of this disclosure, the attributes of the node further include at least one of the following: the identifier of the node, the type of the corresponding reference question, the address of the question image for the reference question with graphics, and the address of the answer image for the reference question with graphics.

[0007] In embodiments of this disclosure, the method further includes: receiving a first request, the first request instructing the generation of a new question set based on an input question set; obtaining the input question set, wherein the input question set includes at least one question and a corresponding answer; extracting knowledge points corresponding to the input question set; retrieving from the reference question bank the hyperedges representing the extracted knowledge points and the nodes they contain; and generating the new question set based on the questions and reference answers of the reference questions represented by the retrieved nodes.

[0008] In embodiments of this disclosure, the method further includes: for each new question in the new question set, extracting the knowledge point corresponding to the new question; and in response to the knowledge point corresponding to the new question being different from the knowledge point corresponding to the input question set, removing the new question from the new question set.

[0009] In embodiments of this disclosure, the method further includes: presenting the new set of test questions.

[0010] In embodiments of this disclosure, the method further includes adding the new set of test questions to the reference test question bank.

[0011] In embodiments of this disclosure, the method further includes: receiving a second request, the second request instructing the recommendation of a reference question set from the reference question bank having the same knowledge points as the knowledge points corresponding to the input question set; obtaining the input question set, the input question set including at least one question and a corresponding answer; extracting the knowledge points corresponding to the input question set; retrieving from the reference question bank a hyperedge representing the extracted knowledge points and the nodes contained therein; and recommending the reference question set using the reference questions represented by the retrieved nodes.

[0012] In embodiments of this disclosure, generating the reference question set using reference questions represented by retrieved nodes includes: determining the semantic similarity between the reference questions represented by retrieved nodes and the input question set; and recommending the reference question set using reference questions represented by a predetermined number of nodes with the highest semantic similarity.

[0013] In embodiments of this disclosure, the method further includes: receiving a third request, the third request instructing a check of the correctness of an input question set; obtaining the input question set, the input question set including at least one question and a corresponding answer; extracting knowledge points corresponding to the input question set; retrieving from the reference question bank the node corresponding to the input question set and its superedge; and for each answer in the input question set, checking whether the answer is correct.

[0014] In embodiments of this disclosure, the method further includes: in response to the answer being correct, presenting a reference answer and the corresponding knowledge point for the question corresponding to the answer; and in response to the answer being incorrect: determining knowledge points represented by the retrieved hyperedge that are different from the knowledge points extracted that correspond to the answer; and presenting a reference answer and the different knowledge point for the question corresponding to the answer.

[0015] In embodiments of this disclosure, the method further includes: constructing a knowledge point graph based on the knowledge points in the reference question bank, wherein the knowledge point graph is structured as a graph, the graph includes nodes and edges, the nodes of the graph represent knowledge points in the reference question bank, the attributes of the nodes of the graph include the content of the corresponding knowledge points, and the edges of the graph represent the relationships between the connected knowledge points.

[0016] In embodiments of this disclosure, the attributes of the nodes of the graph further include at least one of the following: the identifier of the node, and the level of the corresponding knowledge point.

[0017] In embodiments of this disclosure, the method further includes: receiving a fourth request, the fourth request instructing the generation of a new question set having knowledge points different from those corresponding to the input question set; obtaining the input question set, wherein the input question set includes at least one question and a corresponding answer; extracting knowledge points corresponding to the input question set; retrieving nodes representing the extracted knowledge points and their neighboring nodes from the knowledge point graph; retrieving hyperedges representing the knowledge points represented by the neighboring nodes and their contained nodes from the reference question library; and generating a new question set having knowledge points different from those corresponding to the input question set based on the questions and reference answers of the reference questions represented by the retrieved nodes.

[0018] In embodiments of this disclosure, the method further includes: receiving a fifth request, the fifth request instructing the recommendation of a reference question set from the reference question bank having knowledge points different from those corresponding to the input question set; obtaining the input question set, the input question set including at least one question and a corresponding answer; extracting knowledge points corresponding to the input question set; retrieving nodes representing the extracted knowledge points and their neighboring nodes from the knowledge point graph, retrieving hyperedges representing the knowledge points represented by the neighboring nodes and their contained nodes from the reference question bank; and recommending the reference question set using reference questions represented by the nodes retrieved from the reference question bank.

[0019] In embodiments of this disclosure, generating the reference question set using reference questions represented by nodes retrieved from the reference question bank includes: determining the semantic similarity between the reference questions represented by the retrieved nodes and the input question set; and recommending the reference question set using reference questions represented by a predetermined number of nodes with the highest semantic similarity.

[0020] In embodiments of this disclosure, the method further includes: receiving a sixth request, the sixth request indicating a recommended learning path based on an input question set; obtaining the input question set, the input question set including at least one question and a corresponding answer; extracting knowledge points corresponding to the input question set; retrieving from the reference question bank the node corresponding to the input question set and its hyperedge; for each answer in the input question set, checking whether the answer is correct; generating an incorrect question set including incorrect questions with incorrect answers and corresponding knowledge points; for each knowledge point in the knowledge points corresponding to the incorrect question set, retrieving from the knowledge point graph a subgraph containing the node representing the knowledge point and its neighboring nodes, the relationship between the neighboring nodes and the corresponding node being a preorder or postorder relationship; and merging subgraphs with the same nodes to generate a recommended learning path.

[0021] According to a second aspect of this disclosure, a computing system is provided, comprising: a reference question bank, wherein the structure of the reference question bank adopts a hypergraph, the hypergraph including nodes and hyperedges, the nodes representing reference questions, the attributes of the nodes including the question and reference answer of the corresponding reference question, and the hyperedges representing knowledge points corresponding to the reference questions represented by the connected nodes; and at least one processor configured to perform the method according to the first aspect of this disclosure on the reference question bank.

[0022] According to a third aspect of this disclosure, a computer-readable storage medium is provided. Computer program instructions are stored on the computer-readable storage medium, wherein, when executed by a processor, the computer program instructions cause the processor to perform the method according to a first aspect of this disclosure.

[0023] According to a fourth aspect of this disclosure, a computer program product is provided. The computer program product includes computer program instructions, wherein, when executed by a processor, the computer program instructions cause the processor to perform the method according to a first aspect of this disclosure.

[0024] Further aspects and scope of adaptation become apparent from the description provided herein. It should be understood that various aspects of this disclosure may be implemented individually or in combination with one or more other aspects. It should also be understood that the descriptions and specific embodiments in this disclosure are for illustrative purposes and are not intended to limit the scope of this application. Attached Figure Description

[0025] The accompanying drawings described herein are for illustrative purposes only, and do not represent all possible implementations, and are not intended to limit the scope of this application, wherein:

[0026] Figure 1 shows a schematic flowchart of a process for training a knowledge point generation model according to an embodiment of the present disclosure;

[0027] Figure 2 shows a schematic flowchart of a process for constructing a reference test question bank according to an embodiment of the present disclosure;

[0028] Figure 3 shows a schematic diagram of a partial reference test bank constructed according to an embodiment of the present disclosure;

[0029] Figure 4 shows a schematic diagram of a partial knowledge point diagram constructed according to an embodiment of the present disclosure;

[0030] Figure 5 shows a schematic block diagram of the internal structure of the knowledge point generation model;

[0031] Figure 6 shows a schematic block diagram of a computer system based on a reference test question bank according to an embodiment of the present disclosure;

[0032] Figure 7 shows a schematic flowchart of a computer-implemented method based on a reference test question bank according to an embodiment of the present disclosure;

[0033] Figure 8 shows a schematic flowchart of a processing method for providing a response to a first request according to an embodiment of the present disclosure;

[0034] Figure 9 shows a schematic flowchart of a processing method for providing a response to a fourth request according to an embodiment of the present disclosure;

[0035] Figure 10 shows a schematic flowchart of a processing method for providing a response to a second request according to an embodiment of the present disclosure;

[0036] Figure 11 shows a schematic flowchart of a processing method for providing a response to a fifth request according to an embodiment of the present disclosure;

[0037] Figure 12 shows a schematic flowchart of a processing method for providing a response to a third request according to an embodiment of the present disclosure;

[0038] Figure 13 shows a schematic flowchart of a processing method for providing a response to a sixth request according to an embodiment of the present disclosure;

[0039] Figure 14 schematically illustrates a diagram of learning path generation; and

[0040] Figure 15 shows a schematic block diagram of a processing apparatus for a reference test bank according to an embodiment of the present disclosure. Detailed Implementation

[0041] To make the objectives, technical solutions, and advantages of the embodiments of this disclosure clearer, the technical solutions of the embodiments of this disclosure will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this disclosure, and not all of them. Based on the described embodiments of this disclosure, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this disclosure. The embodiments of this disclosure will be described in detail below with reference to the accompanying drawings and embodiments. It should be noted that, unless otherwise specified, the features in the embodiments of this disclosure can be combined with each other.

[0042] Currently, some question bank products exist on the market, but most of these products are based on manually constructed knowledge graphs, and then the relationships between the questions in the question bank and the knowledge graph are manually established. This solution requires a lot of manpower, and secondly, users can only learn from the questions in the question bank. If the original question bank has limited questions for certain knowledge points, users cannot get effective training.

[0043] This disclosure proposes a reference question bank specifically for science and engineering exams, designed to organize and manage exam questions in both science, engineering, and humanities fields. The reference question bank employs a hypergraph structure, comprising nodes and hyperedges. Nodes represent reference questions, and their attributes include at least the reference answer for that question. Hyperedges represent knowledge points related to the reference questions. The knowledge points in the hypergraph can be generated using a trained AI generative pre-trained language model, significantly reducing the manual work required for question bank creation. Furthermore, based on this reference question bank, existing large-scale AI models can be used to generate new questions that share or are related to the knowledge points in the reference question bank, thus addressing the deficiency of insufficient questions on relevant knowledge points. Additionally, the reference question bank can be used to address users' practical needs, providing a "request-response" method and system for processing reference question bank questions.

[0044] The relevant concepts of this invention will be introduced below:

[0045] Exam questions and corresponding answers: These are exam questions in science, engineering, or even humanities (such as law). The exam questions include text descriptions, may contain special symbols and formulas, and may also contain images (such as images in geometry exam questions). The corresponding answers are the answers to the specific exam questions. In the following description, "exam question" refers to the question itself. "Exam question" and "the question of the exam question" have the same meaning and can be used interchangeably.

[0046] Reference questions and answers: These are the questions and their corresponding answers stored in the reference question bank.

[0047] Input questions and input answers: These are the questions from the "Input Question Set" included in the request, along with the answers to those questions.

[0048] Generating test questions and answers: This refers to generating test questions and corresponding answers based on requirements using a large AI model.

[0049] Different reference question banks can be built for different science, engineering, or humanities subjects, such as mathematics, physics, chemistry, engineering mechanics, etc. Even for the same science, engineering, or humanities subject, different levels of reference question banks can be built, such as elementary school mathematics reference question banks and middle school mathematics reference question banks. Those skilled in the art will understand that the processing method is similar for any reference question bank in science, engineering, or humanities.

[0050] A reference question bank can be constructed from the questions and corresponding answers in a regular question bank. Before construction, it is necessary to determine the multiple knowledge points contained in each question. Existing knowledge graphs are essentially knowledge points and their relationships. However, existing knowledge graphs are usually manually input by humans, which requires a lot of human work. In one embodiment of this disclosure, existing AI generative pre-trained language models, such as Transformer, BART, and T5 models, can be used for training. The trained model is then used as a knowledge point generation model, which automatically generates the knowledge points for the questions, thereby reducing human work. An AI generative pre-trained language model is a large-scale language model and an important framework for generative AI. It can use artificial neural networks, be pre-trained on large labeled / unlabeled text datasets, and generate text similar to human natural language. A knowledge point generation model can be trained using any existing AI generative pre-trained language model, and then the trained knowledge point generation model can be used to generate the knowledge points corresponding to all the questions in the regular question bank.

[0051] Figure 1 shows a schematic flowchart of a process 100 for training a knowledge point generation model according to an embodiment of the present disclosure. According to Figure 1, in step S110, a large number of questions, corresponding answers, and all corresponding knowledge points from a general question bank can be obtained as a backup training data set. The questions and corresponding answers obtained at this time can be represented in various ways, such as a mixture of text, symbols, and / or images.

[0052] In step S120, each question and its corresponding answer can be converted into text data, which serves as input data for an AI generative pre-trained language model. Since questions and / or answers may contain not only text descriptions but also symbols, formulas, and images, this disclosure provides a method for converting questions and / or answers into text data, so that subsequent questions and / or answers can be processed by a reference question bank and the AI ​​model. In the process of converting questions and / or answers into text data, if the questions and / or answers contain symbols and / or formulas, the text data of the questions may contain, in addition to the original text, symbols and / or formulas converted to LaTeX format; if the questions and / or answers contain images, the text data of the questions may contain, in addition to the original text, the location where the images of the questions and / or answers are stored, which can be a local storage location or a network storage location. The textification of knowledge points is similar. In this way, questions and / or answers, as well as knowledge points, are converted into text data. The above-described textual processing of test questions and corresponding answers, as well as knowledge points, is a standard procedure in this disclosure and is frequently used in other methods of this disclosure.

[0053] In step S130, by connecting all the knowledge points corresponding to each question using connectors, labeled data for the AI ​​generative pre-trained language model can be formed, thus creating a training dataset. Knowledge points can also be described using text data. A knowledge point can be related to multiple questions, and a question can contain multiple knowledge points. During the labeling process, if a question contains multiple knowledge points, they can be connected using connectors such as commas or slashes. The training dataset is then formed through the processing in steps S120 and S130.

[0054] In step S140, after inputting the training dataset into the AI ​​generative pre-trained language model, the parameters of the AI ​​generative pre-trained language model can be fine-tuned using fine-tuning methods.

[0055] In step S150, once the training conditions are met (e.g., a certain number of training iterations are reached), the model can terminate training. At this point, the trained AI generative pre-trained language model becomes a knowledge point generation model.

[0056] It should be understood that the flowchart of process 100 in Figure 1 is not intended to indicate that the operations of process 100 will be performed in any particular order, or that all operations of process 100 will be included in every case. For example, step S110 uses questions and answers from a general question bank, but examples and answers from textbooks may also be used, for example, to build a secondary school mathematics reference question bank. The order of steps S120 and S130 may also be interchanged, etc.

[0057] In one implementation, the training dataset includes multiple training samples, each of which may include text data consisting of test questions and corresponding answers, and text data consisting of labeled knowledge points related to the test questions. The labeled knowledge points can be manually annotated or obtained through other means, such as from other question banks. The labeled knowledge points enable supervised training of the AI-generative pre-trained language model. During training, the text data consisting of test questions and corresponding answers can serve as input data for the AI-generative pre-trained language model, and the corresponding knowledge point text data can serve as labels. Since the AI-generative pre-trained language model essentially transforms a multi-label classification task into a text generation task, the training dataset needs to contain as many knowledge points as possible.

[0058] In one implementation, during the training of an AI-generative pre-trained language model, after inputting the training dataset into the model, fine-tuning of the model parameters can be performed to improve training speed. Various methods can be used for fine-tuning, such as LoRA, Adapter, P-tuning, and Prefix Tuning. Among these, LoRA is an efficient fine-tuning method that optimizes only a small number of additional model parameters to achieve fine-tuning of the pre-trained model, ensuring that the fine-tuned model can well integrate the knowledge of the subject domain without changing a large number of parameters.

[0059] In one implementation, once the AI ​​generative pre-trained language model is trained into a knowledge point generation model, the knowledge point generation model can be used to implement the knowledge point extraction function. That is, when the knowledge point generation model is input with text data consisting of a test question and its corresponding answer, the knowledge point generation model can output text data consisting of knowledge points related to the test question, and the output text data also uses connectors to connect different knowledge points.

[0060] In one implementation, after obtaining the knowledge points of questions and corresponding answers contained in a general question bank using a knowledge point generation model, the general question bank can be constructed into a reference question bank. The reference question bank uses a hypergraph data structure, which connects the relationships between questions and knowledge points. While a graph data structure can represent the relationship between two objects, an edge in a graph can only connect two nodes. A hypergraph data structure can be seen as an extension of a graph data structure, as hyperedges on a hypergraph can connect multiple nodes, making it more suitable for representing many-to-many relationships between questions and knowledge points; that is, one knowledge point can contain multiple questions, and one question can also contain multiple knowledge points. Therefore, this disclosure constructs a reference question bank by organically integrating a large number of questions and corresponding answers, as well as various corresponding knowledge points, from a general question bank using a hypergraph data structure.

[0061] In one implementation, FIG2 shows a schematic flowchart of a process 200 for constructing a reference test question bank according to an embodiment of the present disclosure. According to FIG1, in step S210, test questions in a general test question bank can be converted into text data of test questions using the method of step S120.

[0062] In step S220, the knowledge point extraction function of the knowledge point generation model can be used to input the text data of the transformed test question (including the corresponding answer) into the knowledge point generation model to generate the text data of the knowledge point corresponding to the test question.

[0063] In step S230, nodes for the reference question bank can be constructed. The data structure of the reference question bank adopts a hypergraph, which includes nodes and hyperedges. Questions from the text data can be constructed as nodes in the hypergraph. For ease of description, the questions in the reference question bank are referred to as reference questions, and the corresponding answers to the reference questions are referred to as reference answers. The reference answers to the reference questions can be used as an attribute of the node. Optionally, nodes can also include other attributes, such as node ID, question type (multiple choice, fill-in-the-blank, application question, etc.), and difficulty level, etc., to facilitate subsequent retrieval of the reference question bank based on the node's attributes.

[0064] In step S240, hyperedges of the reference question bank can be constructed. The knowledge points of the generated questions can be constructed as hyperedges of the hypergraph. The construction rules used in this disclosure may include: questions with the same knowledge points belong to the same hyperedge; a question can reside within different hyperedges. Once the questions and corresponding knowledge points in the ordinary question bank are converted into nodes and hyperedges in the reference question bank, the basic framework of the reference question bank is successfully established.

[0065] It should be understood that the flowchart of process 200 in Figure 2 is not intended to indicate that the operations of process 200 will be performed in any particular order, or that all operations of process 200 will be included in every case. For example, the order of steps S230 and S240 may also be interchanged, etc.

[0066] Figure 3 illustrates a schematic diagram of a partial reference question bank constructed according to an embodiment of the present disclosure. Each small gray circle in Figure 3 represents a question; for example, 305 and 306 represent node 1 (question 1) and node 2 (question 2) of the hypergraph, respectively. The irregular shapes surrounding different numbers of nodes in Figure 3 represent knowledge points, i.e., hyperedges; for example, 301, 302, 303, and 304 represent hyperedge 1 (i.e., knowledge point 1), hyperedge 2 (i.e., knowledge point 2), hyperedge 3 (i.e., knowledge point 3), and hyperedge 4 (i.e., knowledge point 4), respectively. As shown in Figure 3, the hypergraph includes four knowledge points: knowledge point 1, knowledge point 2, knowledge point 3, and knowledge point 4, and 16 questions. Question 1 includes knowledge point 1 and knowledge point 2 (i.e., node 1 is in hyperedge 1 and hyperedge 2), and question 2 includes knowledge point 1 (i.e., node 2 is in hyperedge 1). Both question 1 and question 2 have knowledge point 1 (i.e., both node 1 and node 2 are in hyperedge 1). As can be seen, a hyperedge can connect multiple nodes, and a node can belong to multiple hyperedges. In other words, a knowledge point can be included in multiple test questions, and a test question can contain multiple knowledge points.

[0067] In one implementation, the attributes of the nodes of the hypergraph also include at least one of the following: a node identifier, a corresponding type of reference question, a question image address for a reference question with a graphic, and an answer image address for a reference question with a graphic, etc.

[0068] In one implementation, as shown in FIG2, process 200 of FIG2 may optionally include step S250, in which a knowledge point graph can be constructed based on the knowledge points in the reference test question bank. The knowledge point graph adopts a graph structure, where nodes represent knowledge points, the attributes of nodes include the content of the corresponding knowledge points, and the edges of the graph structure represent the relationship between two knowledge points connected by the edge. The relationship between knowledge points includes inclusion, preorder, or postorder relationships. FIG4 shows a schematic diagram of a partial knowledge point graph constructed according to an embodiment of the present disclosure. In FIG4, each gray circle represents a knowledge point, and the edge connecting two gray circles represents the relationship between knowledge points. In this diagram, thin black edges indicate an inclusion relationship between two connected knowledge points. For example, knowledge point 2 includes knowledge point 4, meaning knowledge point 2 is the parent knowledge point and knowledge point 4 is the child knowledge point. Thick black edges indicate a preorder relationship between two connected knowledge points. For example, knowledge point 3 is the predecessor of knowledge point 2, meaning knowledge point 3 is the foundation of knowledge point 2; knowledge point 3 should be learned before knowledge point 2 can be learned. Dashed edges indicate a successor relationship between two connected knowledge points. For example, knowledge point 1 is the successor of knowledge point 2, meaning knowledge point 2 is the foundation of knowledge point 1; knowledge point 2 should be learned before knowledge point 1 can be learned. Preorder and successor relationships are reciprocal; either convenient relationship can be used in the knowledge point diagram, and the processing results are the same.

[0069] In one implementation, the preorder, postorder, and containment relationships of knowledge points can form a set of triples t, which can be represented as (knowledge point 1, preorder, knowledge point 2), (knowledge point 3, postorder, knowledge point 4), etc. In this way, the knowledge point graph can be represented using nodes and multiple triples. This transforms graph searching into searching within a data structure.

[0070] In one implementation, the attributes of nodes (knowledge points) in the knowledge point graph may also include the node's identifier (ID), the corresponding knowledge point level, and the knowledge point content.

[0071] In one implementation, if the reference questions and answers corresponding to nodes in the reference question bank contain images, such as images of geometric figures in geometry questions, the reference question image address and / or reference answer image address can be added to the attributes of the nodes in the reference question bank. In another implementation, an object detection model can be used to detect objects in the reference questions and answers, cropping the image portions from the output detection boxes, and storing these image portions in the reference questions and / or answers at the reference question image address and reference answer image address. Then, the nodes corresponding to the reference questions are added with these two attributes. In one implementation, the object detection model can use AI models, such as the YOLO series (YOLOv1, YOLOv2, YOLOv3...), Fast-RCNN models, and faster-RCNN models. The YOLO model is commonly used due to its speed. In another implementation, existing image processing object detection models can also be used.

[0072] In one implementation, the node's embedding vector can be stored as an attribute of the node in a reference test question bank.

[0073] In one implementation, the node's embedding vector can utilize the knowledge point generation model described above in this disclosure. Figure 5 schematically illustrates a block diagram of the internal structure of the knowledge point generation model 500. The internal structure of the knowledge point generation model 500 can be a common structure of existing AI generative pre-trained language models. According to Figure 5, the knowledge point generation model 500 may include an Encoder part 502 and a Decoder part 504. After inputting the text data consisting of the reference test question and reference answer corresponding to the node into the Encoder part 502 of the knowledge point generation model 500, the encoder part of the knowledge point generation model 500 outputs a vector 503, which serves as the input vector for the Decoder part 504. After processing by the Decoder part, the knowledge point 505 is obtained. In the embodiments of this disclosure, the output vector 503 can be used as the embedding vector of the node. Those skilled in the art will know that the output vector can also be obtained from other parts of the knowledge point generation model 500 as the embedding vector of the node. The main purpose of the embedding vector is to calculate the semantic similarity of subsequent test questions or answers.

[0074] In one implementation, the node embedding vector can also be calculated using AI hypergraph neural network models familiar to those skilled in the art, such as HGNN, DeepHGNN, and HyperGCN. AI hypergraph neural networks can aggregate information about all neighboring nodes and hyperedges around a given node, where neighboring nodes are all nodes connected to that node. If two nodes are connected by an edge, it indicates a relationship between them. Graphs or hypergraphs can connect various nodes, showing the relationships between them. AI hypergraph neural networks incorporate these relationships into the node information, so each node is no longer an isolated entity. The above implementation directly uses the output vector of the encoder part as the embedding vector of the node corresponding to the input question. Such embedding vectors only contain the semantic information of this node (question) and lack the information about the relationships between nodes. Using an AI hypergraph neural network model not only considers the semantic information of the node (question) but also the information about the relationships between nodes, and can be used to calculate the embedding vector of each node and hyperedge in the hypergraph based on the initial embedding vectors of each node and hyperedge. In the embodiments of this disclosure, although the subsequent processing only needs to calculate the embedding vector of the node and does not need to calculate the embedding vector of the hyperedge, the intermediate calculation process requires the participation of the initial embedding vector of the hyperedge.

[0075] In one implementation, both the initial embedding vectors of nodes and hyperedges can be random vectors. However, calculating the node embedding vector in this way is time-consuming and has poor subsequent utilization. In another implementation, the initial embedding vectors of hyperedges are still random vectors, but the initial embedding vectors of nodes can be calculated using the knowledge point generation model described above. Specifically, after inputting the text data consisting of the reference questions and answers corresponding to a node into the Encoder part 502 of the knowledge point generation model 500, the output vector 503 of the encoder part of the knowledge point generation model 500 can be used as the initial embedding vector of that node. In this way, using the initial embedding vectors of nodes and hyperedges in the reference question bank as input to the AI ​​hypergraph neural network model, the node embedding vector can be obtained by reaching a specified number of iterations through the operation of the AI ​​hypergraph neural network model. The calculated node embedding vector can be stored as an attribute of the node in the reference question bank for subsequent processing.

[0076] The established reference question bank can be used to support various processing methods related to the reference question bank, such as retrieval and request / response functions. For simplicity, this paper will subsequently use "question" instead of the text data of the question itself, "corresponding answer" instead of the text data of the corresponding answer, and "knowledge point" instead of the text data of the knowledge point. Correspondingly, "reference question" will be used instead of the text data of the reference question, "reference answer" instead of the text data of the reference answer, "input question" instead of the text data of the input question, "input answer" instead of the text data of the input answer, "generated question" corresponding to the text data of the generated question, and "generated answer" corresponding to the text data of the generated answer, etc.

[0077] In one embodiment, based on the reference test question bank constructed as disclosed above, this disclosure discloses a computer system based on the reference test question bank. Figure 6 shows a schematic structural block diagram of a computer system 600 based on the reference test question bank according to an embodiment of this disclosure. According to Figure 600, the system 600 includes: a reference test question bank 602 and a processor 604. The reference test question bank 602 can be configured to adopt a hypergraph structure, where nodes of the hypergraph represent reference test questions, and the attributes of the nodes include at least the corresponding answer to the reference test question; the hyperedges of the hypergraph represent knowledge points related to the reference test questions; the processor 604 can be configured to perform operations on the reference test question bank 602.

[0078] It should be understood that the block diagram of FIG6 is not intended to indicate that system 600 includes all the components shown in FIG6. Rather, system 600 may include fewer or additional components (e.g., additional data or additional results, etc.) not shown in FIG6.

[0079] In one implementation, the attributes of the nodes of the hypergraph of the reference question bank 602 further include at least one of the following: a node identifier, a corresponding reference question type, a question image address for a reference question with a graphic, and an answer image address for a reference question with a graphic.

[0080] In one embodiment, the processor 604 of FIG. 6 can be configured to receive a request 601, wherein the request 601 may include an input question set, or the input question set may not be included in the request but may be obtained by the processor 604 using a separate step. In one embodiment, the input questions in the input question set may be related to reference questions in a reference question bank 606. The processor 604 can also be configured to, in response to request 601, retrieve information related to the input question set from the reference question bank 602. The reference question bank 606 can also be configured to generate a response to the request based on the request and the retrieved information.

[0081] The system 600 shown in FIG6 can support various computer-implemented methods based on a reference test question bank, such as the processing methods shown in FIG7-FIG13 described below. FIG7 shows a schematic flowchart of a processing method 700 according to an embodiment of the present disclosure. According to FIG7, in step S710, a request may be received. In step S720, in response to the request, information may be retrieved from the reference test question bank. In step S730, a response to the request is generated based on the request and the retrieved information.

[0082] It should be understood that the flowchart in Figure 7 is not intended to indicate that method 700 includes all the steps shown in Figure 7. Rather, method 700 may include fewer or additional steps not shown in Figure 7 (e.g., establishing a connection with a reference test bank, etc.).

[0083] In one embodiment, the request in step S710 includes an input question set, which includes at least one question and its corresponding answer. In another embodiment, step S710 further includes obtaining the input question set, which includes at least one question and its corresponding answer. In yet another embodiment, the input questions in the input question set are related to reference questions in a reference question bank, and step 720 includes retrieving information related to the input question set from the reference question bank.

[0084] The hypergraph technology described above allows for the reconstruction of existing question banks and knowledge points. The constructed reference question bank not only stores information such as questions and corresponding answers but also preserves the relationships between questions and knowledge points. Using the system shown in Figure 6 and the method shown in Figure 7, combined with the constructed reference question bank, users can retrieve questions and knowledge points from the reference question bank. Furthermore, based on user performance or common mistakes, a large AI model can generate new questions not currently in the question bank without retraining. In addition, the system can recommend questions related to relevant knowledge points to users, grade user questions, and use the hypergraph to analyze users' mastery of various knowledge points, recommending learning paths, and so on.

[0085] Since the reference question bank contains reference questions and knowledge points based on hypergraphs, as well as knowledge point graphs based on graphs, these two graph data structures themselves can realize retrieval functions. Users or other applications can complete the search for questions and knowledge points as needed. Therefore, the request shown in Figure 7 can include single and compound searches based on hypergraph data structures, single searches based on graph data structures, compound searches based on graph data structures, and joint searches based on hypergraph and graph data structures. In a single search based on a hypergraph data structure, you can search based on question type attributes to return all questions of a specified type; you can search based on hyperedges to return all questions corresponding to a specified knowledge point. In a compound search based on a hypergraph data structure, you can specify both question type and knowledge point to search, returning all questions that meet the requirements, such as "find all multiple-choice questions related to rational numbers". In a single search based on a graph data structure, you can search the knowledge point graph based on the relationship between nodes to return the preceding, succeeding, related, and / or containing knowledge points corresponding to a specified knowledge point, and understand the relationships between various knowledge points. In composite retrieval based on graph data structures, both grade level and knowledge point relationships can be specified simultaneously, returning all questions that meet the requirements, such as "find all second-grade successor knowledge points related to rational numbers." In joint retrieval based on hypergraphs and graphs, more complex searches can be achieved by jointly querying the content of the reference question bank and the knowledge point graph. Since the knowledge point content can link information from both the graph and hypergraph data structures, to achieve more complex question queries, one can first filter out the set of knowledge points that meet the conditions from the graph structure, and then use these knowledge points to filter out the set of questions that meet the conditions from the hypergraph data structure. For example, to query all questions contained in the predecessor (successor, related, contained) knowledge points of a specified knowledge point, or questions of a specified type, one needs to first find the set of predecessor knowledge points corresponding to the specified knowledge point in the knowledge point graph, and then find all questions or questions of a specified type contained in the hyperedges of these predecessor knowledge points in the reference question bank. For example, if you want to query all the test questions corresponding to all knowledge points of a specified grade or a specific test question, you would first search for all the knowledge point sets of the specified grade in the graph structure, and then search for all the test questions or test questions of the specified type contained in the hyperedges of these knowledge points in the reference test question bank, and so on.

[0086] In one implementation, the request shown in Figure 7 can be described as "generating generated questions and answers that are identical to the knowledge points in the input question set, and the generated questions are different from the reference questions in the reference question bank," hereinafter referred to as the first request. Such a first request, to generate new questions and corresponding answers, requires the use of large AI models used in the AI ​​field. Large AI models, short for Large Language Models (LLMs), are trained using massive amounts of data and employ billions of parameters to generate raw outputs for tasks such as answering questions, translating languages, and completing sentences. Currently, large AI models have been widely used in text generation, question-answering systems, and other fields, all thanks to their generative capabilities. However, because these large AI models are typically trained using publicly available, general-purpose data, they may not be able to meet specific practical needs in certain domains. However, retraining with domain-specific data incurs high costs. Therefore, many large AI models currently employ Retrieval Augmentation (RAG) technology to enhance their generation capabilities. RAG retrieves relevant knowledge and integrates the search results into the AI ​​model's instructions (commonly known as prompt instructions), providing the model with more reasonable outputs. Typically, RAG's retrieval process is based on document data and a vector database of segmented documents. It compares the user's question with various text blocks in the vector database, returning the result with the highest similarity. Since the data involved in this disclosure is not merely plain text but includes text data containing numerous symbols and formulas, direct segmentation might destroy the semantic information of the entire question, generating meaningless text blocks. Therefore, unlike traditional RAG, this disclosure does not segment the data to build a vector database but instead uses a constructed hypergraph to achieve the question retrieval function. Currently, multiple companies have developed various large AI models to form question-answering systems. This invention can utilize any suitable large AI model to generate new questions and answers.

[0087] In one implementation, processor 604 may be configured to: receive a first request instructing the generation of a new question set based on an input question set. Processor 604 may be further configured to: acquire the input question set, wherein the input question set includes at least one question and a corresponding answer; extract knowledge points corresponding to the input question set; retrieve hyperedges representing the extracted knowledge points and their contained nodes from a reference question bank; and generate a new question set based on the questions and answers of reference questions represented by the retrieved nodes.

[0088] In one implementation, the processor 604 may be further configured to: for each new question in the new question set, extract the knowledge point corresponding to the new question; and in response to the knowledge point corresponding to the new question being different from the knowledge point corresponding to the input question set, remove the new question from the new question set.

[0089] Figure 8 shows a schematic flowchart of a processing method 800 for providing a response to a first request according to an embodiment of the present disclosure. Generally, Figure 8 is a refinement of the method of Figure 7 in processing the first request; therefore, Figure 8 includes the three steps S710, S720, and S730 of Figure 7. Referring to Figure 8, step S810 is the same as step S710, i.e., a first request can be received, which instructs the generation of a new set of questions with the same knowledge points based on an input question set. This input question set can be obtained using a separate step, and the input question set includes at least one question and a corresponding answer, or the input question set can be directly included in the first request, and the input question set includes at least one question and a corresponding answer. The input questions in the input question set can be in a different form than the reference questions in a reference question bank. The step corresponding to step S720, "retrieve information related to the input question set from the reference question bank," in Figure 8 includes steps S820 and S830. In step S820, by inputting the text data consisting of input questions and answers from the input question set into the knowledge point generation model, a knowledge point set corresponding to the input question set can be obtained. This utilizes the previously trained knowledge point generation model, which can input not only reference questions and answers but also the aforementioned input questions and answers, thereby obtaining the knowledge points corresponding to the input questions. In step S830, the knowledge points in the obtained knowledge point set are used as hyperedges. The set of reference questions and answers corresponding to these hyperedges in the reference question bank can be retrieved and used as the retrieval information. The basic retrieval functions of the aforementioned reference question bank can be used here.

[0090] Referring again to Figure 8, the steps corresponding to step S730, "Generate a response to the request based on the request and the retrieved information," include steps S840, S850, S860, S870, S880, and S890. In step S840, based on the request and the retrieved set of reference questions and answers, an AI large-scale model instruction can be constructed. This instruction instructs the AI ​​large-scale model to generate a set of generated questions and answers similar to the retrieved set of reference questions and answers, but the generated questions in this set differ from the reference questions in the reference question bank. Currently, the AI ​​large-scale model instruction typically uses the `prompt` instruction. As a prompt word, it is primarily used to guide the AI ​​large-scale model in generating high-quality output results, clearly informing the AI ​​large-scale model of the tasks it needs to complete, the necessary background knowledge, and the specific output format. For example, the format of the prompt instruction required by method 800 can be expressed as follows: "As a question generator, please generate questions and reference answers similar to those in the input question set, and the generated questions cannot be the same as the content of the reference questions in the reference question bank. Please output the results in the format (question; reference answer)." In step S850, the set of generated questions and answers can be obtained by inputting the constructed AI large model instruction into the AI ​​large model. In step S860, the knowledge point generation model can be used to obtain a new set of knowledge points corresponding to the generated questions and answers in the set of generated questions and answers. This step still utilizes the knowledge point generation model to generate knowledge points. In step S870, the new set of knowledge points can be compared with the above-mentioned set of knowledge points. This comparison compares each knowledge point in the new knowledge point set with each knowledge point in the existing knowledge point set one by one. If a knowledge point in the new knowledge point set is different from any knowledge point in the existing knowledge point set, then that knowledge point in the new knowledge point set is determined to be different from any knowledge point in the existing knowledge point set; if that knowledge point in the new knowledge point set is the same as any knowledge point in the existing knowledge point set, then that knowledge point in the new knowledge point set is determined to be the same as any knowledge point in the existing knowledge point set. In step S880, in response to the fact that a specific knowledge point in the new knowledge point set is different from a knowledge point in the existing knowledge point set, the generated test question and generated answer corresponding to that specific knowledge point can be deleted from the set of generated test questions and generated answers, thereby updating the set of generated test questions and generated answers. In step S890, the updated set of generated test questions and generated answers can be used as the response to the request.

[0091] In one implementation, processor 604 can present a new set of test questions.

[0092] In one implementation, processor 604 can add a new set of test questions to a reference test question bank.

[0093] It should be understood that the flowchart in Figure 8 is not intended to indicate that method 800 includes all the steps shown in Figure 8. Rather, method 800 may include fewer or additional steps not shown in Figure 8.

[0094] In one implementation, processor 604 is also configured to perform operations on the knowledge point graph.

[0095] In one implementation, the request shown in Figure 7 can be "to generate questions and answers related to knowledge points in the input question set, and the generated questions are different from the reference questions in the reference question bank." Since this request is related to the first request mentioned above, it will be referred to as the fourth request below. Such a fourth request, to generate new questions and corresponding answers, still requires the use of large AI models used in the AI ​​field.

[0096] In one embodiment, the processor 604 is further configured to: receive a fourth request, the fourth request instructing the generation of a new set of questions with knowledge points different from those corresponding to the input set of questions, based on the input set of questions. The processor 604 is further configured to: acquire the input set of questions, wherein the input set of questions includes at least one question and a corresponding answer; extract knowledge points corresponding to the input set of questions; retrieve nodes representing the extracted knowledge points and their neighboring nodes from the knowledge point graph; retrieve hyperedges representing the knowledge points represented by the neighboring nodes and their contained nodes from the reference question bank; and generate a new set of questions with knowledge points different from those corresponding to the input set of questions, based on the questions and reference answers of the reference questions represented by the retrieved nodes.

[0097] Figure 9 shows a schematic flowchart of a processing method 900 for providing a response to a fourth request according to an embodiment of the present disclosure. Generally, Figure 9 is a refinement of the method of Figure 7 in processing a fourth request, and therefore Figure 9 also includes the three steps S710, S720, and S730 of Figure 7. Referring to Figure 9, step S910 is the same as step S710, i.e., a fourth request can be received, which instructs the generation of a new set of knowledge-point-related questions based on an input question set. This input question set can be obtained using a separate step, and the input question set includes at least one question and a corresponding answer, or the input question set can be directly included in the fourth request, and the input question set includes at least one question and a corresponding answer. The input questions in the input question set can be in a different form than the reference questions in a reference question bank. The steps corresponding to step S720, "retrieve information related to the input question set from the reference question bank," in Figure 9 include steps S920, S930, and S940. In step S920, by inputting the text data consisting of input questions and answers from the input question set into the knowledge point generation model, the knowledge point set corresponding to the input question set can be obtained. In step S930, the set of first-order neighbor knowledge points in the knowledge point graph corresponding to the knowledge points in the knowledge point set can be obtained. In the fourth request, the meaning of knowledge point related can be interpreted as the existence of at least one of the inclusion, preorder, or postorder relationships between the knowledge points of the generated question and the input question. If there is a direct inclusion, preorder, or postorder relationship between two knowledge points, then these two knowledge points are first-order neighbor knowledge points. Obtaining the first-order neighbor knowledge points in the knowledge point graph corresponding to the inclusion, preorder, and postorder relationships can be obtained by searching the knowledge point graph or by retrieving the head / tail entity of each element in the triple set t of the knowledge point graph. In step S940, by using the knowledge points in the obtained set of first-order neighbor knowledge points as hyperedges, the set of reference questions and reference answers corresponding to the hyperedges in the reference question library is retrieved and used as the retrieved information.

[0098] Referring again to Figure 9, the steps corresponding to step S730, "Generate a response to the request based on the request and the retrieved information," include steps S950, S960, S970, S980, S990, and S995. In step S950, based on the request and the retrieved set of reference questions and answers, an AI large-scale model instruction can be constructed. This instruction instructs the AI ​​large-scale model to generate a set of generated questions and answers similar to the retrieved set of reference questions and answers, but the generated questions in this set differ from the reference questions in the reference question bank. The `prompt` instruction can still be used here. For example, the format of the prompt instruction required by method 900 can be expressed as follows: "As a question generator, please generate questions and reference answers similar to those in the input question set, and the generated questions cannot be the same as the content of the reference questions in the reference question bank. Please output the results in the format (question; reference answer)." In step S960, the constructed AI large model instructions can be input into the AI ​​large model to obtain a set of generated questions and answers. In step S970, the knowledge point generation model can be used to obtain a new set of knowledge points corresponding to the generated questions and answers in the set of generated questions and answers. This step still utilizes the knowledge point generation model's function of generating knowledge points. In step S980, the new set of knowledge points can be compared with the above-mentioned first-order neighbor knowledge point set. This comparison compares each knowledge point in the new knowledge point set with each knowledge point in the aforementioned first-order neighbor knowledge point set. If a knowledge point in the new knowledge point set is different from any knowledge point in the aforementioned first-order neighbor knowledge point set, then that knowledge point in the new knowledge point set is determined to be different from any knowledge point in the aforementioned first-order neighbor knowledge point set. If that knowledge point in the new knowledge point set is the same as any knowledge point in the aforementioned first-order neighbor knowledge point set, then that knowledge point in the new knowledge point set is determined to be the same as any knowledge point in the aforementioned first-order neighbor knowledge point set. In step S880, in response to the fact that a specific knowledge point in the new knowledge point set is different from a knowledge point in the aforementioned first-order neighbor knowledge point set, the generated test question and generated answer corresponding to that specific knowledge point can be deleted from the set of generated test questions and generated answers, thereby updating the set of generated test questions and generated answers. In step S890, the updated set of generated test questions and generated answers can be used as the response to the request.

[0099] It should be understood that the flowchart in Figure 9 is not intended to indicate that method 900 includes all the steps shown in Figure 9. Rather, method 900 may include fewer or additional steps not shown in Figure 9.

[0100] In the methods for generating new test questions shown in Figures 8 and 9, the AI ​​large model can be an LLAMA model, an LLAVA model, a GLM model, or a Qwen model, etc. If the input test questions and answers in the input test question set do not contain images, the single-modal AI large model mentioned above can be used to generate new test questions (where the LLAVA model does not have a single-modal model). If the input test questions and answers in the input test question set contain images, then the multi-modal AI large model mentioned above needs to be used to generate new test questions.

[0101] In one implementation, optionally, Figures 8 and 9 may further include the following steps (not shown in Figures 8 and 9): If all generated questions in the set of generated questions and answers have been completed, the generated questions in the set of generated questions and answers are added as reference questions to the reference question bank according to the hypergraph data structure. This increases the number of questions in the reference question bank, giving users more choices.

[0102] In one implementation, the request shown in FIG7 can be "recommend k1 reference questions and corresponding answers from a reference question bank that have the same knowledge points as the input question set", hereinafter referred to as the second request. In this implementation, the processor 604 can be configured to: receive the second request, which instructs to recommend a reference question set from the reference question bank that has the same knowledge points as the knowledge points corresponding to the input question set. The processor 604 can be further configured to: obtain the input question set, which includes at least one question and a corresponding answer; extract the knowledge points corresponding to the input question set; retrieve the hyperedges representing the extracted knowledge points and the nodes they contain from the reference question bank 602; and recommend the reference question set using the reference questions represented by the retrieved nodes.

[0103] In one trial mode, processor 604 can be further configured to generate the reference question set using the reference questions represented by the retrieved nodes by: determining the semantic similarity between the reference questions represented by the retrieved nodes and the input question set; and recommending the reference question set using the reference questions represented by a predetermined number of nodes with the highest semantic similarity.

[0104] Figure 10 shows a schematic flowchart of a processing method 1000 for providing a response to a second request according to an embodiment of the present disclosure. Generally, Figure 10 is a refinement of the method of Figure 7 in processing a second request; therefore, Figure 10 includes the three steps S710, S720, and S730 of Figure 7. Referring to Figure 10, step S1010 is the same as step S710, i.e., a second request can be received, which instructs the recommendation of a set of reference questions from a reference question bank having the same knowledge points as the input question set. The input question set can be obtained using a separate step, and the input question set includes at least one question and a corresponding answer, or the input question set can be directly included in the second request, and the input question set includes at least one question and a corresponding answer. The input questions in the input question set may be in a different form than the reference questions in the reference question bank. In Figure 10, steps S1020 and S1030, corresponding to step S720 "retrieving information related to the input question set from the reference question bank," are the same as steps S820 and S830 in Figure 8, and will not be repeated here. Referring again to Figure 10, the steps corresponding to step S730 "generating a response to the request based on the request and the retrieved information" include steps S1040 and S1050. In step S1040, the semantic similarity s1 between the reference questions in the retrieved set of reference questions and answers and the input questions in the input question set can be calculated. In step S1050, the k1 retrieved reference questions and answers with the highest semantic similarity can be selected from the retrieved set of reference questions and answers and used as the response to the second request.

[0105] It should be understood that the flowchart in FIG10 is not intended to indicate that method 1000 includes all the steps shown in FIG10. Rather, method 1000 may include fewer or additional steps not shown in FIG10.

[0106] In one implementation, the request shown in Figure 7 can be "recommend k2 reference questions and corresponding answers from a reference question bank related to the knowledge points in the input question set", hereinafter referred to as the fifth request.

[0107] In one implementation, processor 604 is configured to: receive a fifth request instructing the recommendation of a reference question set from a reference question bank that has knowledge points different from those corresponding to the input question set. Processor 604 is further configured to: acquire the input question set, which includes at least one question and a corresponding answer; extract knowledge points corresponding to the input question set; retrieve nodes representing the extracted knowledge points and their neighboring nodes from a knowledge point graph; retrieve hyperedges representing the knowledge points represented by the neighboring nodes and their contained nodes from the reference question bank; and recommend the reference question set using reference questions represented by the nodes retrieved from the reference question bank.

[0108] In one implementation, the processor 604 is further configured to generate the reference question set using reference questions represented by nodes retrieved from the reference question bank 602 by: determining the semantic similarity between the reference questions represented by the retrieved nodes and the input question set; and recommending the reference question set using reference questions represented by a predetermined number of nodes with the highest semantic similarity.

[0109] Figure 11 shows a schematic flowchart of a processing method 1100 for providing a response to a fifth request according to an embodiment of the present disclosure. Generally, Figure 11 is a refinement of the method of Figure 7 in processing a fifth request, and therefore Figure 11 includes the three steps S710, S720, and S730 of Figure 7. Referring to Figure 11, step S1110 is the same as step S710, i.e., a fifth request can be received, which instructs the recommendation of a set of questions from a reference question bank related to knowledge points based on an input question set. This input question set can be obtained using a separate step, and the input question set includes at least one question and a corresponding answer, or the input question set can be directly included in the fifth request, and the input question set includes at least one question and a corresponding answer. The input questions in this input question set can be in a different form than the reference questions in the reference question bank. In Figure 11, steps S1120, S1130, and S1140, corresponding to step S720 "retrieving information related to the input question set from the reference question bank," are the same as steps S920, S930, and S940 in Figure 9, and will not be repeated here. Referring again to Figure 11, the steps corresponding to step S730 "generating a response to the request based on the request and the retrieved information" include steps S1150 and S1160. In step S1150, the semantic similarity s1 between the reference questions in the retrieved set of reference questions and answers and the input questions in the input question set can be calculated. In step S1160, the k2 retrieved reference questions and answers with the highest semantic similarity can be selected from the retrieved set of reference questions and answers and used as the response to the fifth request.

[0110] It should be understood that the flowchart in Figure 11 is not intended to indicate that method 1100 includes all the steps shown in Figure 11. Rather, method 1100 may include fewer or additional steps not shown in Figure 11.

[0111] In one implementation, the semantic similarity s1 between the reference test question and the input test question in steps S1040 and S1150 can be calculated using the following formula:

[0112] Where q1 e1 h1 is the embedding vector of the reference question. e1 This is the embedding vector of the input question. Those skilled in the art can also define other methods for calculating semantic similarity s1.

[0113] In one implementation, q1 e1 The value can be the embedding vector of the corresponding node of the reference question in the reference question bank, h1 e1 The value can be the output vector of the encoder part of the knowledge point generation model after the text data consisting of the input test question and the input answer is input into the knowledge point generation model.

[0114] In one implementation, the selection of the k1 / k2 retrieved reference questions and answers with the highest semantic similarity in steps S1050 and S1160 can be achieved by first calculating multiple semantic similarities s1 between the retrieved specific reference question and each input question in the input question set, and then using the highest value among the multiple s1 as the semantic similarity s1 between the specific reference question and the input questions in the input question set. max Then from multiple s1 max Select s1 max The highest k1 / k2 questions retrieved are the reference questions and answers, which are then used as the response to the request.

[0115] In one implementation, the request shown in FIG7 may be "grading the input question set", hereinafter referred to as the third request. In one implementation, the processor 604 may be configured to: receive the third request, which instructs to check the correctness of the input question set; obtain the input question set, which includes at least one question and its corresponding answer; extract the knowledge points corresponding to the input question set; retrieve the node corresponding to the input question set and its hyperedge from the reference question bank 602; and for each answer in the input question set, check whether the answer is correct.

[0116] In one implementation, the processor 604 may be further configured to: in response to a correct answer, present the reference answer and the corresponding knowledge point of the question corresponding to the answer; and in response to an incorrect answer: determine the knowledge points represented by the retrieved hyperedge that are different from the knowledge points extracted that correspond to the answer; and present the reference answer and the different knowledge points of the question corresponding to the answer.

[0117] Figure 12 shows a schematic flowchart of a processing method 1200 for providing a response to a third request according to an embodiment of the present disclosure. Generally, Figure 12 is a refinement of the method of Figure 7 in processing a third request, and therefore Figure 12 also includes the three steps S710, S720, and S730 of Figure 7. Referring to Figure 12, step S1210 is the same as step S710, i.e., a third request can be received, which instructs the user to check the correctness of an input question set. The input question set can be obtained using a separate step, and the input question set includes at least one question and a corresponding answer, or the input question set can be directly included in the third request, and the input question set includes at least one question and a corresponding answer. The input questions in the input question set can be questions that the user has already completed, and the questions can be in a different form than reference questions in a reference question bank. In Figure 12, step S720, "Retrieving information related to the input question set from the reference question bank," includes step S1220, which retrieves the reference answer of the reference question corresponding to the input question in the input question set from the reference question bank. Referring again to Figure 12, step S730, "Generating a response to the request based on the request and the retrieved information," includes: for at least one input question in the input question set, executing steps S1230, S1240, S1250, S1260, S1270, S1280, and S1290. In step S1230, the semantic similarity s2 between the corresponding answer of the reference question and the input answer of the input question is calculated. In step S1240, the edit distance lev between the reference answer and the input answer is calculated. a,b In step S1250, based on semantic similarity s2 and edit distance lev a,b Calculate the score for the input answer using the following formula, where score = α·1 / lev a,b +(1-α)·s2

[0118] And here, α represents a set proportional coefficient. In step S1260, it is determined whether the score is greater than m. Here, m is a set threshold, which can be set by the user or by the system 600. In step S1270, in response to score>m, the correction result of the input answer is determined to be correct. In step S1280, in response to score≤m, the correction result of the input answer is determined to be incorrect. In step S1290, the correction result of at least one input question in the input question set is used as a response to the request. In this way, the correction result of each input question in the input question set can be provided to the user or other applications. In addition, in one embodiment, in response to the correction result of an input question being incorrect, the reference answer corresponding to the input question can be added to the correction result as a response to the request.

[0119] It should be understood that the flowchart in Figure 12 is not intended to indicate that method 1200 includes all the steps shown in Figure 12. Rather, method 1200 may include fewer or additional steps not shown in Figure 12.

[0120] In one implementation, the semantic similarity s2 can be calculated using the following formula:

[0121] Among them, q2 e1 h2 is the embedding vector of the reference answer. e1 This is the embedding vector of the input answer. Those skilled in the art can also define other methods for calculating semantic similarity s².

[0122] In one implementation, q2 e1 The value of h2 can be the embedding vector of the node corresponding to the reference question for that answer. e1 The value can be the output vector of the encoder part of the knowledge point generation model after the text data consisting of the input test question and the input answer is input into the knowledge point generation model.

[0123] In one implementation, the edit distance lev a,b The following formula can be used to calculate:

[0124] Where 'a' represents the text data consisting of the corresponding answer, 'b' represents the text data consisting of the input answer, and 'i' and 'j' represent the character indices of the text data in 'a' and 'b', respectively. Those skilled in the art can also define other methods for calculating edit distance.

[0125] In one implementation, the test question recommendations in Figures 10 and 11 can be expanded by combining the error correction shown in Figure 12. For example, by examining the test questions a user has completed, we can assess their understanding of the knowledge points. If a user has a weak grasp of certain knowledge points, we can recommend various test questions corresponding to those knowledge points. If a user has completed many test questions with almost no errors, it indicates that they have mastered these knowledge points and can proceed to learning subsequent knowledge. Therefore, we can recommend corresponding test questions for subsequent knowledge points to help the user further advance their learning process. The specific implementation process can be as follows:

[0126] 1. Randomly sample n reference questions from the reference question bank and return them to the user. After the user completes the reference questions, the questions are graded, and for the wrong questions, the knowledge points that the user has not mastered are returned.

[0127] 2. If a user's number of incorrect answers is greater than w1 (w1 can be a percentage or absolute number defined by the user or the system), then based on the knowledge points corresponding to these incorrect answers, test questions corresponding to these knowledge points will be recommended to the user.

[0128] 3. If the number of incorrect answers is less than w2 (w2 can be a percentage or absolute number defined by the user or the system), it means that the user has mastered these knowledge points well. Therefore, the subsequent nodes corresponding to these knowledge points can be retrieved, and the corresponding reference questions can be returned to the user, indicating that the user has learned the current knowledge point and can continue to learn the subsequent content.

[0129] In one implementation, the request shown in Figure 7 can be "recommend a learning path based on the input question set", hereinafter referred to as the sixth request.

[0130] In one implementation, processor 604 is configured to: receive a sixth request indicating a recommended learning path based on an input question set. Processor 604 is further configured to: acquire the input question set, which includes at least one question and its corresponding answer; extract knowledge points corresponding to the input question set; retrieve nodes and their hyperedges corresponding to the input question set from a reference question bank; for each answer in the input question set, check if the answer is correct; generate an incorrect question set including incorrect answers and their corresponding knowledge points; for each knowledge point in the knowledge points corresponding to the incorrect question set, retrieve a subgraph from the knowledge point graph containing the node representing the knowledge point and its neighboring nodes, wherein the relationship between the neighboring nodes and the corresponding nodes is a preorder or postorder relationship; and merge subgraphs with the same nodes to generate a recommended learning path.

[0131] Figure 13 shows a schematic flowchart of a processing method 1300 for providing a response to a sixth request according to an embodiment of the present disclosure. Generally, Figure 13 is a refinement of the method of Figure 7 in processing the sixth request; therefore, Figure 13 includes the three steps S710, S720, and S730 of Figure 7. Referring to Figure 13, step S1310 is the same as step S710, i.e., the sixth request can be received, the fifth related request indicating a recommended learning path based on an input question set. The input question set can be obtained using a separate step, and the input question set includes at least one question and a corresponding answer, or the input question set can be directly included in the sixth request, and the input question set includes at least one question and a corresponding answer. The input questions in the input question set can be in a different form than the reference questions in the reference question bank. The steps corresponding to step S720 "retrieve information related to the input question set from the reference question bank" in Figure 13 include steps S1320, S1330, and S1340. In step S1320, the request to "grade the input question set" (i.e., execute the third request) is executed, obtaining the set of incorrect questions. In step S1330, the set of knowledge points corresponding to the incorrect questions in the incorrect question set is retrieved from the reference question bank. In step S1340, the set of first-order neighbor knowledge points corresponding to the knowledge points in the knowledge point set in the knowledge point graph is obtained. The steps corresponding to step S720 "retrieve information related to the input question set from the reference question bank" in Figure 13 include steps S1350, S1360, S1370, and S1380. In step S1350, for the knowledge points in the above knowledge point set and the knowledge points in the above first-order neighbor knowledge point set, multiple candidate knowledge point paths are obtained according to the preorder and postorder relationships in the knowledge point graph. In step S1360, the multiple candidate knowledge point paths are iteratively merged. During the merging process, if in any two candidate knowledge point paths, the subsequent knowledge point of one path is the same as the preceding knowledge point of the other, then these two paths are merged into a new candidate knowledge point path. In step S1370, in response to the completion of the merging, the remaining knowledge point path after merging is used as the recommended learning path. In step S1380, the recommended learning path is used as the response to the request.

[0132] It should be understood that the flowchart in Figure 13 is not intended to indicate that method 1300 includes all the steps shown in Figure 13. Rather, method 1300 may include fewer or additional steps not shown in Figure 13.

[0133] In one implementation, each node in the recommended learning path represents various knowledge points. Knowledge points in the set of knowledge points corresponding to incorrect test questions can be marked as important nodes, which indicate knowledge points that the user has not mastered. Users can determine which aspects they need to consolidate based on the marked important nodes and the recommended learning path, thereby strengthening their learning process.

[0134] Figure 14 schematically illustrates the learning path generation diagram. The leftmost part of Figure 14 shows a partial knowledge point diagram. Black circles represent knowledge points p2, p5, and p7 corresponding to incorrect test questions, i.e., important nodes.

[0135] In one implementation, in step S1350, multiple candidate knowledge point paths can be obtained using triples in the knowledge point graph. Nodes and edges in the knowledge point graph can be represented by triples t, for example, t1 = (knowledge point 1, predecessor, knowledge point 2) indicates that knowledge point 1 is the predecessor node of knowledge point 2, t2 = (knowledge point 2, predecessor, knowledge point 3) indicates that knowledge point 2 is the predecessor node of knowledge point 3, and so on. This allows obtaining a set tr of all candidate triples in the knowledge point graph for the knowledge points in the aforementioned knowledge point set and the knowledge points in the aforementioned first-order neighbor knowledge point set. Then, all triples in the candidate triple set tr can be traversed. If any two triples ti = (ei1, ri, ei2) and tj = (ej2, rj, ej3), the tail entity of ti and the head entity of tj are the same (i.e., ei2 = ej2), and the relation type ri = rj, then these two triples can be connected to obtain a candidate path. When ri = rj = "preorder", the candidate path is represented as ei1->ei2->ej3; when ri = rj = "postorder", the candidate path is represented as ei3->ei2->ej1. These candidate paths are stored in the candidate knowledge point path set r. For example, since the tail entity of t1 and the head entity of t2 overlap, the three knowledge points can be connected through the preorder relationship to form a shorter learning path, such as knowledge point 1->knowledge point 2->knowledge point 3. Here, each point in the learning path can also be called a node of the learning path, and the preceding node in the path is the preorder part of the knowledge point. The meaning of the learning path knowledge point 1->knowledge point 2->knowledge point 3 is to learn knowledge point 1 first, then learn knowledge point 2, and finally learn knowledge point 3. Those skilled in the art will know that multiple triples t represented by nodes and edges in the knowledge point graph can be used to quickly obtain the candidate knowledge point path set r, but the candidate knowledge point path set r can also be obtained by direct graph search. In Figure 14, three candidate knowledge point paths can be obtained using the above method: p6->p7->p8, p5->p6, and p1<-p2<-p3. These are shown as candidate paths in the middle of Figure 14.

[0136] In one implementation, in step S1360, to merge multiple candidate knowledge point paths, all elements in the candidate path set r can be traversed. If the tail node of any two candidate paths is the same as the head node of the other path, then these two paths are merged to form a new candidate path. This process is repeated until there are no more paths in set r that can be merged. At this point, all the merged paths in set r are the learning paths to be recommended to the user. In Figure 14, two recommended learning paths can be obtained using the above method: p5->p6->p7->p8 and p1<-p2<-p3. These are shown as the recommended learning paths on the right side of Figure 14.

[0137] The method of this disclosure can be implemented in the system shown in Figure 6. Those skilled in the art will understand that, based on the above-described responses to the first through sixth requests, other simpler or more complex requests, such as retrieval requests, question-and-answer requests, etc., can be designed for system 600. Users can operate system 600 by providing request 601 to system 600, thereby obtaining response 603. Furthermore, system 600 of this disclosure can also be connected to other systems to receive request 601 from other systems and provide response 603.

[0138] In one embodiment, a computer-implemented method is also disclosed, comprising: constructing a reference question bank; wherein the structure of the reference question bank adopts a hypergraph, the hypergraph includes nodes and hyperedges, the nodes represent reference questions, the attributes of the nodes include the question and the reference answer of the corresponding reference question, and the hyperedges represent the knowledge points corresponding to the reference questions represented by the connected nodes. This computer-implemented method can also implement all the steps of the above methods executed by the processor 604.

[0139] Figure 15 shows a schematic block diagram of a processing device 1500 for a reference test item bank according to an embodiment of the present disclosure. The device 1500 includes one or more processors 1502 and a memory 1504. The memory 1504 is coupled to the processor 1502 via a bus and to an I / O interface 1506, and stores instructions executable by the processor 1502. When the instructions are executed by the processor 1502, the device 1500 can perform the steps of the method for processing a reference test item bank according to any of the above embodiments.

[0140] The memory 1504 may include a readable medium in the form of volatile memory, such as random access memory (RAM) and / or cache memory, and may further include read-only memory (ROM).

[0141] The memory 1504 may also include a program / utility having a set (at least one) of program modules, including but not limited to: an operating system, one or more application programs, other program modules, and program data, each or some combination of these examples may include an implementation of a network environment.

[0142] A bus can represent one or more of several types of bus structures, including a memory bus or memory controller, a peripheral bus, a graphics acceleration port, a processing unit, or a local bus that uses any of the various bus structures.

[0143] Device 1500 can also communicate with one or more external devices (e.g., keyboards, pointing devices, Bluetooth devices, etc.), one or more devices that enable a user to interact with device 1500, and / or any device that enables device 1500 to communicate with one or more other computing devices (e.g., routers, modems, etc.). This communication can be performed via input / output (I / O) interface 1506. Furthermore, device 1500 can also communicate with one or more networks (e.g., local area networks (LANs), wide area networks (WANs), and / or public networks, such as the Internet) via a network adapter. The network adapter can communicate with other modules of device 1500 via a bus. It should be understood that, although not shown in the figures, other hardware and / or software modules can be used in conjunction with device 1500, including but not limited to: microcode, device drivers, redundant processing units, external disk drive arrays, RAID systems, tape drives, and data backup storage systems.

[0144] In embodiments of this disclosure, a computer-readable storage medium is also provided, on which computer program instructions are stored, which, when executed by, for example, a processor, can implement the steps of the computer-implemented method based on the reference test question bank in any of the above embodiments. In some possible implementations, various aspects of this disclosure can also be implemented as a program product comprising program code that, when run on a terminal device, causes the terminal device to perform the steps described in this specification for processing the reference test question bank according to various exemplary embodiments of this disclosure.

[0145] In embodiments of this disclosure, a computer program product is also provided. This computer program product includes computer program instructions, which, when executed by a processor, can implement the steps of the computer-implemented method based on the reference test question bank in any of the above embodiments.

[0146] The program product for implementing the above-described method according to embodiments of this disclosure may employ a portable compact disc read-only memory (CD-ROM) and include program code, and may run on a terminal device, such as a personal computer. However, the program product of this disclosure is not limited thereto. In this document, a readable storage medium may be any tangible medium that contains or stores a program that may be used by or in conjunction with an instruction execution system, apparatus, or device.

[0147] The program product may employ any combination of one or more readable media. A readable medium may be a readable signal medium or a readable storage medium. A readable storage medium may be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples of readable storage media (a non-exhaustive list) include: an electrical connection having one or more wires, a portable disk, a hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof.

[0148] The computer-readable storage medium may include data signals propagated in baseband or as part of a carrier wave, carrying readable program code. Such propagated data signals may take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. The readable storage medium may also be any readable medium other than a readable storage medium, capable of transmitting, propagating, or transmitting programs for use by or in connection with an instruction execution system, apparatus, or device. The program code contained on the readable storage medium may be transmitted using any suitable medium, including but not limited to wireless, wired, optical fiber, RF, etc., or any suitable combination thereof.

[0149] Program code for performing the operations of this disclosure can be written in any combination of one or more programming languages, including object-oriented programming languages ​​such as Java and C++, and conventional procedural programming languages ​​such as C or similar languages. The program code can execute entirely on the user's computing device, partially on the user's computing device, as a standalone software package, partially on the user's computing device and partially on a remote computing device, or entirely on a remote computing device or server. In cases involving remote computing devices, the remote computing device can be connected to the user's computing device via any type of network, including a local area network (LAN) or a wide area network (WAN), or it can be connected to an external computing device (e.g., via the Internet using an Internet service provider).

[0150] Those skilled in the art will understand that various aspects of this disclosure can be implemented as a system, method, or program product. Therefore, various aspects of this disclosure can be specifically implemented in the following forms: a completely hardware implementation, a completely software implementation (including firmware, microcode, etc.), or a combination of hardware and software aspects, collectively referred to herein as a "circuit," "module," or "system."

[0151] The above description is merely a preferred embodiment of this disclosure and is not intended to limit this disclosure. Various modifications and variations can be made to this disclosure by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this disclosure should be included within the scope of protection of this disclosure.

Claims

1. A computer-implemented method comprising: constructing a reference question library; wherein a structure of the reference question library adopts a hypergraph, the hypergraph comprising nodes and hyperedges, the nodes representing reference questions, attributes of the nodes including titles and reference answers of the corresponding reference questions, the hyperedges representing knowledge points corresponding to the reference questions represented by the connected nodes.

2. The method of claim 1, wherein, the attributes of the nodes further including at least one of the following: an identifier of the node, a type of the corresponding reference question, a title picture address for a reference question with graphics, and an answer picture address for a reference question with graphics.

3. The method of claim 1 or 2, further comprising: receiving a first request indicating to generate a new question set based on an input question set; obtaining the input question set, wherein the input question set comprises at least one title and a corresponding answer; extracting knowledge points corresponding to the input question set; retrieving, from the reference question library, hyperedges representing the extracted knowledge points and nodes contained therein; and generating the new question set based on the titles and reference answers of the reference questions represented by the retrieved nodes.

4. The method of claim 3, further comprising: for each new question in the new question set, extracting knowledge points corresponding to the new question; and in response to the knowledge points corresponding to the new question being different from the knowledge points corresponding to the input question set, removing the new question from the new question set.

5. The method of claim 3 or 4, further comprising: presenting the new question set.

6. The method of any of claims 3-5, further comprising: adding the new question set to the reference question library.

7. The method of any one of claims 1 to 6, further comprising: receiving a second request indicating to recommend a reference question set having the same knowledge points as an input question set from the reference question library; obtaining the input question set, the input question set comprising at least one title and a corresponding answer; extracting knowledge points corresponding to the input question set; retrieving, from the reference question library, hyperedges representing the extracted knowledge points and nodes contained therein; and using the reference questions represented by the retrieved nodes to recommend the reference question set. using the reference questions represented by the retrieved nodes to generate the reference question set comprises:

8. The method of claim 7, wherein, determining semantic similarity of the reference questions represented by the retrieved nodes to the input question set; and using reference questions represented by a predetermined number of nodes with the highest semantic similarity to recommend the reference question set.

9. The method of any one of claims 1 to 8, further comprising: receiving a third request indicating to check correctness of an input question set; obtaining the input question set, the input question set comprising at least one title and corresponding answer; extracting knowledge points corresponding to the input question set; retrieving, from reference question library, nodes corresponding to the input question set and hyperedges in which the nodes are located; and for each answer in the input question set, checking whether the answer is correct.

10. The method of claim 9, further comprising: ​ ​ in response to the answer being correct, presenting a reference answer of the question corresponding to the answer and a corresponding knowledge point; and in response to the answer being incorrect: determining a knowledge point represented by the retrieved hyperedge that is different from the knowledge point corresponding to the answer among the extracted knowledge points; and presenting a reference answer of the question corresponding to the answer and the different knowledge point. 11.The method of any one of claims 1-10, further comprising: constructing a knowledge point graph based on the knowledge points in the reference question bank, wherein a structure of the knowledge point graph adopts a graph, the graph comprises nodes of the graph and edges of the graph, the nodes of the graph represent the knowledge points in the reference question bank, and attributes of the nodes of the graph comprise content of the corresponding knowledge points, and the edges of the graph represent relationships between the connected knowledge points.

12. The method of claim 11, wherein, The attributes of the nodes of the graph further comprise at least one of the following: an identifier of the node, and a level of the corresponding knowledge point. 13.The method of claim 11 or 12, further comprising: receiving a fourth request indicating to generate a new question set having knowledge points different from knowledge points corresponding to an input question set based on the input question set; obtaining the input question set, wherein the input question set comprises at least one question and a corresponding answer; extracting knowledge points corresponding to the input question set; retrieving, from the knowledge point graph, nodes representing the extracted knowledge points and neighbor nodes thereof; retrieving, from the reference question bank, hyperedges representing knowledge points represented by the neighbor nodes and nodes contained in the hyperedges; and generating the new question set having knowledge points different from knowledge points corresponding to the input question set based on questions and reference answers of reference questions represented by the retrieved nodes. 14.The method of any one of claims 11-13, further comprising: receiving a fifth request indicating to recommend a reference question set having knowledge points different from knowledge points corresponding to an input question set from the reference question bank; obtaining the input question set, wherein the input question set comprises at least one question and a corresponding answer; extracting knowledge points corresponding to the input question set; retrieving, from the knowledge point graph, nodes representing the extracted knowledge points and neighbor nodes thereof, retrieving, from the reference question bank, hyperedges representing knowledge points represented by the neighbor node and nodes contained in the hyperedges; and using reference questions represented by the retrieved nodes from the reference question bank to recommend the reference question set. Using reference questions represented by the retrieved nodes from the reference question bank to generate the reference question set comprises:

15. The method of claim 14, wherein, determining semantic similarity of reference questions represented by the retrieved nodes to the input question set; and using reference questions represented by a predetermined number of nodes with the highest semantic similarity to recommend the reference question set. 16.The method of any one of claims 11-15, further comprising: receiving a sixth request indicating to recommend a learning path based on an input question set; ​ obtaining the input question set, the input question set including at least one question and corresponding answer; extracting knowledge points corresponding to the input question set; retrieving nodes corresponding to the input question set and hyper-edges in which the nodes are located from the reference question library; checking whether the answer is correct for each answer in the input question set; generating an error question set including error questions with incorrect answers and corresponding knowledge points; retrieving a subgraph including a node representing the knowledge point and neighbor nodes of the node for each knowledge point corresponding to the error question set from the knowledge point graph, the neighbor nodes having a pre-order relationship or a post-order relationship with the corresponding node; and merging subgraphs with the same node to generate a recommended learning path.

17. A computing system comprising: a reference question library, wherein a structure of the reference question library adopts a hypergraph, the hypergraph including nodes and hyper-edges, the nodes representing reference questions, attributes of the nodes including questions and reference answers of corresponding reference questions, the hyper-edges representing knowledge points corresponding to reference questions represented by connected nodes; and at least one processor configured to perform the method of any one of claims 1-16 for the reference question library.

18. A computer readable storage medium having stored thereon computer program instructions, wherein, The computer program instructions, when executed by the processor, cause the processor to perform the method according to any one of claims 1 to 16.

19. A computer program product comprising computer program instructions, wherein, The computer program instructions, when executed by the processor, cause the processor to perform the method according to any one of claims 1 to 16. The computer program instructions, when executed by the processor, cause the processor to perform the method according to any one of claims 1 to 16.

Images