Training method of cognitive diagnosis model, question recommendation method, device and equipment

Abstract

The application provides a cognitive diagnosis model training method, a question recommendation method, a device and equipment, and belongs to the technical field of artificial intelligence. The method comprises the following steps: embedding a first object, a plurality of nodes in a knowledge graph and a first question based on a cognitive diagnosis model of the i th iteration to obtain object characteristics of the first object, a plurality of node characteristics of the plurality of nodes and question characteristics of the first question; fusing the object characteristics and the plurality of node characteristics based on the cognitive diagnosis model of the i th iteration to obtain knowledge state characteristics; and training the cognitive diagnosis model of the i th iteration based on the question characteristics and the knowledge state characteristics, taking the answer result of the first object to the first question as supervision information. The above scheme can determine the mastery degree of each knowledge of the object based on the answering situation of the object, and then train the cognitive diagnosis model, so that the trained cognitive diagnosis model has high accuracy.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of artificial intelligence, in particular to a cognitive diagnosis model training method, a question recommendation method, a device and equipment. BACKGROUND

[0002] Cognitive diagnosis plays a very important role in intelligent education. The goal of cognitive diagnosis is to dig into the proficiency of students on all knowledge. How to improve the accuracy of cognitive diagnosis is a research direction.

[0003] At present, when cognitive diagnosis modeling is performed, the relationship between students and questions can be analogized to users and goods in a recommendation system, and a matrix decomposition method is used to complete the record matrix of the students' answers, that is, to predict the scores of the students on the unanswered questions, so as to obtain the proficiency of the students on all knowledge.

[0004] Since there is a dependency relationship between the knowledge involved in the questions, such as one-digit multiplication being the precursor knowledge of two-digit multiplication and two-digit multiplication being the successor knowledge of one-digit multiplication, the above scheme does not accurately model the complex relationship between students and knowledge, resulting in a predicted result that cannot accurately reflect the proficiency of students on knowledge. SUMMARY

[0005] The embodiments of the present application provide a cognitive diagnosis model training method, a question recommendation method, a device and equipment, which can determine the mastery of objects on each knowledge based on the answering situation of the objects, and then train the cognitive diagnosis model, so that the trained cognitive diagnosis model has high accuracy. The technical solution is as follows:

[0006] In one aspect, a cognitive diagnosis model training method is provided, the method comprising:

[0007] Based on the cognitive diagnosis model of the i-th iteration, the first object, a plurality of nodes in the knowledge graph and the first question are embedded to obtain an object feature of the first object, a plurality of node features of the plurality of nodes and a question feature of the first question, the object feature is used to represent the mastery of the first object on the knowledge in at least one second question, the nodes in the knowledge graph represent knowledge, the edges in the knowledge graph represent the dependency relationship between knowledge, and the node features include the features of the corresponding nodes and the features of the nodes connected to the corresponding nodes;

[0008] Based on the cognitive diagnosis model of the i-th iteration, the object feature and the plurality of node features are fused to obtain a knowledge state feature, the knowledge state feature is used to represent the mastery of the first object on the knowledge represented by the plurality of nodes;

[0009] The answer result of the first object to the first question is taken as supervision information, and the cognitive diagnosis model of the i th iteration is trained based on the question feature and the knowledge state feature, the cognitive diagnosis model being used for prediction based on input objects and questions to obtain a probability that the object correctly answers the question.

[0010] In another aspect, a question recommendation method is provided, the method comprising:

[0011] performing cognitive diagnosis on a second object based on a target cognitive diagnosis model to obtain a target knowledge state feature of the second object, the target cognitive diagnosis model being trained based on node features of multiple nodes in a knowledge graph, the nodes in the knowledge graph representing knowledge, edges in the knowledge graph representing dependency relationships between knowledge, the node features including features of corresponding nodes and features of nodes connected to the corresponding nodes, the target knowledge state feature being used to represent a mastery degree of the second object to knowledge represented by the multiple nodes;

[0012] determining at least one target question to be recommended based on the target knowledge state feature;

[0013] recommending the at least one target question to the second object.

[0014] In another aspect, a training device of a cognitive diagnosis model is provided, the device comprising:

[0015] an embedding module configured to perform embedding processing on a first object, multiple nodes in a knowledge graph, and a first question based on a cognitive diagnosis model of an i th iteration to obtain an object feature of the first object, multiple node features of the multiple nodes, and a question feature of the first question, the object feature being used to represent a mastery of the first object to knowledge in at least one second question, the nodes in the knowledge graph representing knowledge, edges in the knowledge graph representing dependency relationships between knowledge, the node features including features of corresponding nodes and features of nodes connected to the corresponding nodes;

[0016] a fusion module configured to fuse the object feature and the multiple node features based on the cognitive diagnosis model of the i th iteration to obtain a knowledge state feature, the knowledge state feature being used to represent a mastery of the first object to knowledge represented by the multiple nodes;

[0017] a training module configured to take an answer result of the first object to the first question as supervision information, and train the cognitive diagnosis model of the i th iteration based on the question feature and the knowledge state feature, the cognitive diagnosis model being used for prediction based on input objects and questions to obtain a probability that the object correctly answers the question.

[0018] In some embodiments, the fusion module is configured to, for any node, concatenate the object feature, a node feature of the node, and a feature of a predecessor node of the node based on the cognitive diagnosis model of the i-th iteration, to obtain a concatenated feature of the node, the predecessor node representing a predecessor knowledge of a knowledge represented by the node; and perform full connection on the concatenated features of the plurality of nodes based on the cognitive diagnosis model of the i-th iteration, to obtain the knowledge state feature.

[0019] In some embodiments, the embedding module is configured to determine an object vector of the first object based on an object identifier of the first object; obtain an object embedding matrix of the first object based on the object identifier; and multiply the object vector and the object embedding matrix to obtain the object feature of the first object.

[0020] In some embodiments, the embedding module is configured to, for any node in the knowledge graph, obtain at least one neighbor node connected to the node; and perform weighted summation on features of the at least one neighbor node and a feature of the node based on node weights of the at least one neighbor node, to obtain a node feature of the node, the node weights being used to represent importance degrees of the corresponding neighbor nodes to the node.

[0021] In some embodiments, the question feature of the first question comprises a discrimination feature, a question knowledge feature, and a difficulty feature, the question knowledge feature being used to represent knowledge contained in the first question.

[0022] The embedding module is configured to determine a question vector of the first question based on a question identifier of the first question; obtain a discrimination embedding matrix, a question knowledge embedding matrix, and a difficulty embedding matrix of the first question based on the question identifier; and determine the discrimination feature, the question knowledge feature, and the difficulty feature based on the question vector, the discrimination embedding matrix, the question knowledge embedding matrix, and the difficulty embedding matrix.

[0023] In some embodiments, the embedding module is configured to multiply the question vector and the discrimination embedding matrix to obtain the discrimination feature; multiply the question vector and the question knowledge embedding matrix to obtain the question knowledge feature; and multiply the question vector and the difficulty embedding matrix to obtain the difficulty feature.

[0024] In some embodiments, the training module is configured to: obtain a feature difference value by subtracting the difficulty feature from the knowledge state feature; obtain a Hadamard product by multiplying the question knowledge feature and the feature difference value bit by bit; obtain an object answering feature by multiplying the Hadamard product and the discrimination feature; obtain a prediction result information by predicting the object answering feature based on the cognitive diagnosis model of the i th iteration, the prediction result information being used to indicate a probability that the first object correctly answers the first question; and adjust a model parameter of the cognitive diagnosis model of the i th iteration based on a difference between the prediction result information and an answer result of the first object to the first question.

[0025] In some embodiments, the apparatus further includes:

[0026] an obtaining module configured to obtain, based on the cognitive diagnosis model of the i th iteration, an edge matrix of the knowledge graph, an element in the edge matrix being used to represent an edge between the plurality of nodes;

[0027] a prediction module configured to predict, based on the cognitive diagnosis model of the i th iteration, the edge matrix to obtain edge probability information, the edge probability information being used to represent a probability that an edge exists between any two nodes in the plurality of nodes;

[0028] a graph processing module configured to add an edge in the knowledge graph based on the edge probability information.

[0029] In some embodiments, the apparatus further includes:

[0030] a parameter adjusting module configured to adjust a model parameter of the cognitive diagnosis model of the i th iteration based on a difference between the edge probability information and an existing edge in the knowledge graph.

[0031] In another aspect, a question recommendation apparatus is provided, the apparatus including:

[0032] an object processing module configured to perform cognitive diagnosis on a second object based on a target cognitive diagnosis model to obtain a target knowledge state feature of the second object, the target cognitive diagnosis model being trained based on node features of a plurality of nodes in a knowledge graph, a node in the knowledge graph representing knowledge, an edge in the knowledge graph representing a dependency relationship between knowledge, the node features including a feature of a corresponding node and features of nodes connected to the corresponding node, the target knowledge state feature being used to represent a mastery degree of the second object to knowledge represented by the plurality of nodes;

[0033] a determination module configured to determine at least one question to be recommended based on the target knowledge state feature;

[0034] recommendation module configured to recommend the at least one subject title to the second subject.

[0035] In another aspect, a computer device is provided, which includes a processor and a memory, the memory being configured to store at least one piece of computer program, the at least one piece of computer program being loaded and executed by the processor to implement the operations performed by the training method of the cognitive diagnosis model in the embodiments of the present application, or the at least one piece of computer program being loaded and executed by the processor to implement the operations performed by the item recommendation method in the embodiments of the present application.

[0036] In another aspect, a computer readable storage medium is provided, which stores at least one piece of computer program, the at least one piece of computer program being loaded and executed by a processor to implement the training method of the cognitive diagnosis model in the embodiments of the present application, or the at least one piece of computer program being loaded and executed by the processor to implement the item recommendation method in the embodiments of the present application.

[0037] In another aspect, a computer program product or computer program is provided, which includes computer program code stored in a computer readable storage medium, a processor of a computer device reading the computer program code from the computer readable storage medium, the processor executing the computer program code to cause the computer device to perform the training method of the cognitive diagnosis model provided in various optional implementation manners of the aspects described above, or to cause the computer device to perform the item recommendation method provided in various optional implementation manners of the aspects described above.

[0038] The embodiments of the present application provide a training scheme of a cognitive diagnosis model. Since the mastery of knowledge by a subject is influenced by the knowledge itself and other knowledge dependent on the knowledge, and whether the subject can correctly answer a question is influenced by the knowledge contained in the question, the dependency relationship between the knowledge is introduced into the training of the cognitive diagnosis model, so that the mastery of the knowledge by the subject can be determined based on the answering of the subject, and then the cognitive diagnosis model is trained, so that the trained cognitive diagnosis model has high accuracy. BRIEF DESCRIPTION OF DRAWINGS

[0039] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings needed in the embodiment description will be briefly introduced. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor.

[0040] Figure 1is a training method of a cognitive diagnosis model according to an embodiment of the present application;

[0041] Figure 2 is a flowchart of a training method of a cognitive diagnosis model according to an embodiment of the present application;

[0042] Figure 3 is a flowchart of another training method of a cognitive diagnosis model according to an embodiment of the present application;

[0043] Figure 4 is a schematic diagram of embedding processing of a knowledge graph according to an embodiment of the present application;

[0044] Figure 5 is a schematic diagram of feature fusion according to an embodiment of the present application;

[0045] Figure 6 is a model framework diagram of a cognitive diagnosis model according to an embodiment of the present application;

[0046] Figure 7 is an architecture schematic diagram of an application scenario according to an embodiment of the present application;

[0047] Figure 8 is a schematic diagram of a learning situation report according to an embodiment of the present application;

[0048] Figure 9 is a flowchart of a question recommendation method according to an embodiment of the present application;

[0049] Figure 10 is a schematic diagram of a question recommendation interface according to an embodiment of the present application;

[0050] Figure 11 is a block diagram of a training device of a cognitive diagnosis model according to an embodiment of the present application;

[0051] Figure 12 is a block diagram of another training device of a cognitive diagnosis model according to an embodiment of the present application;

[0052] Figure 13 is a block diagram of a question recommendation device according to an embodiment of the present application;

[0053] Figure 14 is a structural block diagram of a terminal according to an embodiment of the present application;

[0054] Figure 15 is a structural schematic diagram of a server according to an embodiment of the present application. DETAILED DESCRIPTION

[0055] In order to make the purposes, technical solutions and advantages of the present application clearer, the embodiments of the present application will be further described in detail below with reference to the drawings.

[0056] The terms "first", "second" and the like in the present application are used to distinguish the same items or similar items with substantially the same function and action, and it should be understood that there is no logical or time sequence relationship between "first", "second", "nth", and the number and execution order are not limited.

[0057] The term "at least one" in the present application means one or more, and the meaning of "a plurality of" is two or more.

[0058] Hereinafter, the terms involved in the present application are explained.

[0059] Artificial intelligence (AI) is to use digital computers or digital computer controlled machines to simulate, extend and expand human intelligence, perceive environment, acquire knowledge and use knowledge to obtain the best results. In other words, artificial intelligence is a comprehensive technology of computer science, which tries to understand the essence of intelligence and produce a new intelligent machine that can react in a similar way to human intelligence. Artificial intelligence is to study the design principles and implementation methods of various intelligent machines, so that machines have the functions of perception, reasoning and decision-making.

[0060] Artificial intelligence technology is a comprehensive discipline, involving a wide range of fields, both hardware and software technologies. Artificial intelligence basic technologies generally include technologies such as sensors, special artificial intelligence chips, cloud computing, distributed storage, big data processing technology, operation / interaction system, mechatronics, etc. Artificial intelligence software technology mainly includes computer vision technology, speech processing technology, natural language processing technology, and machine learning / deep learning, automatic driving, intelligent transportation and other major directions.

[0061] Natural language processing (NLP) is an important direction in the field of computer science and artificial intelligence. It studies various theories and methods that can realize effective communication between people and computers using natural language. Natural language processing is a science that integrates linguistics, computer science and mathematics. Therefore, the research in this field will involve natural language, i.e. the language used in daily life, so it has a close relationship with the study of linguistics. Natural language processing technology usually includes text processing, semantic understanding, machine translation, robot question and answer, knowledge graph and other technologies.

[0062] Machine Learning (ML) is a multi-disciplinary subject that involves probability theory, statistics, approximation theory, convex analysis, algorithmic complexity theory, etc. It is a specialized study of how computers simulate or implement human learning behavior to acquire new knowledge or skills, reorganize existing knowledge structure, and continuously improve their performance. Machine learning is the core of artificial intelligence and the fundamental approach to making computers intelligent. Its applications are widespread in various fields of artificial intelligence. Machine learning and deep learning generally include artificial neural networks, belief networks, reinforcement learning, transfer learning, inductive learning, and teaching learning.

[0063] With the research and progress of artificial intelligence technology, artificial intelligence technology has been researched and applied in many fields, such as common smart home, smart wearable devices, virtual assistants, smart speakers, smart marketing, unmanned vehicles, autonomous vehicles, drones, robots, smart medical care, smart customer service, Internet of Vehicles, autonomous driving, intelligent transportation, etc. It is believed that with the development of technology, artificial intelligence technology will be applied in more fields and play an increasingly important role.

[0064] Knowledge precedence relationship refers to the predecessor-successor dependency relationship between knowledge in a discipline. For example, in mathematics, one-digit multiplication is the predecessor of two-digit multiplication. A student needs to master one-digit multiplication before he can master two-digit multiplication.

[0065] Item response theory (IRT) is also known as latent trait theory or latent trait model. It is a modern psychometric theory that can guide item selection and test development. IRT assumes that the subject has a "latent trait", which is a statistical construct based on observed and analyzed test responses. In the test, latent trait generally refers to latent ability, and test total score is often used to estimate this potential. IRT believes that the subject's response and performance on test items have a special relationship with their latent trait. The item parameters established by IRT have the characteristics of permanence, meaning that scores from different measurement scales can be unified. IRT integrates various item analysis data through item response curves, allowing us to comprehensively and intuitively see item difficulty, discrimination, and other item analysis characteristics, thereby guiding item selection, test development, and score comparison.

[0066] The sigmoid function is a commonly used S-shaped function in biology, also known as the S-shaped growth curve.

[0067] The training method of the cognitive diagnosis model provided in the embodiments of the present application can be executed by a computer device, which is a terminal or a server. The following takes the computer device as a server as an example to introduce the implementation environment of the training method of the cognitive diagnosis model provided in the embodiments of the present application. Figure 1 is a schematic diagram of an implementation environment of a training method of a cognitive diagnosis model according to the embodiments of the present application. Referring to FIG. 1, Figure 1 the implementation environment includes a terminal 101 and a server 102.

[0068] The terminal 101 and the server 102 can be directly or indirectly connected through wired or wireless communication, which is not limited in the present application.

[0069] In some embodiments, the terminal 101 is a smart phone, a tablet computer, a notebook computer, a desktop computer, a smart speaker, a smart watch, etc., but is not limited thereto. The terminal 101 is installed and runs an application program, such as an education application program. The user account logged in by the terminal 101 can learn knowledge, answer questions, and view the answering situation based on the application program.

[0070] In some embodiments, the server 102 is a standalone physical server, and can also be a server cluster or a distributed system composed of multiple physical servers, and can also be a cloud server providing basic cloud computing services such as cloud service, cloud database, cloud computing, cloud function, cloud storage, network service, cloud communication, middleware service, domain name service, security service, CDN (Content Delivery Network), and big data and artificial intelligence platform. The server 102 is used to provide background service for the above-mentioned application program installed in the terminal 101. In some embodiments, the server 102 undertakes the main computing work, and the terminal 101 undertakes the secondary computing work; or the server 102 undertakes the secondary computing work, and the terminal 101 undertakes the main computing work; or the server 102 and the terminal 101 cooperatively compute in a distributed computing architecture. Those skilled in the art can know that the number of the above-mentioned terminals can be more or less. For example, the above-mentioned terminal can be only one, or the above-mentioned terminal can be dozens or hundreds, or more. The number and the type of the terminal are not limited in the embodiments of the present application.

[0071] Figure 2 is a flowchart of a training method of a cognitive diagnosis model according to the embodiments of the present application, referring to FIG. 2, Figure 2 in the embodiments of the present application, the training method of the cognitive diagnosis model is taken as an example executed by the server. The training method of the cognitive diagnosis model includes the following steps:

[0072] 201. The server performs embedding processing on the first object, the plurality of nodes in the knowledge graph, and the first question based on the cognitive diagnosis model of the i th iteration, to obtain object features of the first object, a plurality of node features of the plurality of nodes, and question features of the first question.

[0073] In the embodiments of the present application, i is a positive integer, and the i th iteration is any iteration in the model training process. The cognitive diagnosis model is used to predict the first object and the first question input, to obtain the probability that the first object correctly answers the first question. The first object is a sample object. The nodes in the knowledge graph represent knowledge, such as mathematical knowledge, Chinese knowledge, and geographical knowledge, etc. The edges in the knowledge graph represent the dependency relationship between knowledge, such as the dependency relationship between one-digit addition and two-digit addition in mathematical knowledge, that is, one-digit addition must be mastered before two-digit addition can be mastered. The first question is a sample question answered by the first object. The object features are used to represent the characteristics of the first object in answering at least one second question, and the second question is a sample question answered by the first object. The node features include the features of the corresponding node and the features of the nodes connected to the corresponding node. The process of embedding processing of the cognitive diagnosis model on the first object, the plurality of nodes in the knowledge graph, and the first question is the process of extracting object features, node features, and question features.

[0074] 202. The server fuses the object features and the plurality of node features based on the cognitive diagnosis model of the i th iteration, to obtain knowledge state features.

[0075] In the embodiments of the present application, the server can splice the object features with each node feature respectively, and then perform full connection on the plurality of features obtained by splicing, to obtain knowledge state features. The knowledge state features can represent the mastery degree of the first object to the knowledge represented by the plurality of nodes.

[0076] 203. The server trains the cognitive diagnosis model of the i th iteration based on the question features and the knowledge state features, with the answer result of the first object to the first question as supervision information.

[0077] In the embodiments of the present application, the answer result of the first object to the first question is the actual result of the first object answering the first question, and the answer result is correct or incorrect. The server predicts the answer result of the first object answering the first question based on the question features and the knowledge state features, and trains the model based on the difference between the predicted answer result and the actual result with the actual result as supervision information.

[0078] The training scheme of the cognitive diagnosis model is provided in the embodiments of the present application. Since the mastering degree of the object on knowledge is influenced by the knowledge itself and other knowledge dependent on the knowledge, and whether the object can correctly answer the question is influenced by the knowledge contained in the question, the dependency relationship between the knowledge is added in the training by introducing the knowledge graph in the training of the cognitive diagnosis model, so that the mastering degree of the object on the knowledge can be determined based on the answering condition of the object, and then the cognitive diagnosis model is trained, so that the cognitive diagnosis model obtained by the training has high accuracy.

[0079] Figure 2 The main process of the training method of the cognitive diagnosis model provided in the embodiments of the present application is exemplarily shown, and the training method of the cognitive diagnosis model is described in detail as follows, Figure 3 is a flowchart of another training method of a cognitive diagnosis model provided in the embodiments of the present application, and the training method of the cognitive diagnosis model is described in detail as follows, Figure 3 It is described in the embodiments of the present application by taking the execution by the server as an example. It should be noted that the server can train the cognitive diagnosis model in a multi-round iteration manner, and the process of each round of iteration is similar. The present application takes the i-th round of iteration as an example, and i is a positive integer. The training method of the cognitive diagnosis model includes the following steps:

[0080] 301. The server performs embedding processing on the first object based on the cognitive diagnosis model of the i-th round of iteration to obtain object features of the first object, and the object features are used to represent the characteristics of the first object in answering at least one second question.

[0081] In the embodiments of the present application, the cognitive diagnosis model of the i-th round of iteration is the cognitive diagnosis model after parameter adjustment in the i-1-th round of iteration, or the cognitive diagnosis model of the i-th round of iteration is an initial model without training. The first object is a sample object, such as a student object. The first object answers a plurality of sample questions, such as mathematical questions. For the convenience of description, the sample questions not participating in the model training in the plurality of sample questions are referred to as first questions, and the sample questions participating in the model training are referred to as second questions. The server inputs the object identifier of the first object into the cognitive diagnosis model of the i-th round of iteration, and performs embedding processing on the first object by the embedding layer in the cognitive diagnosis model to obtain the mastering degree of the first object on the knowledge in at least one second question. That is, before the i-th round of iteration, the cognitive diagnosis model can determine the mastering degree of the first object on the knowledge in at least one second question based on the answering condition of the first object on at least one second question, and the object features of the first object can be obtained by embedding processing on the first object.

[0082] In some embodiments, the server can determine an object vector of the first object based on the object identifier of the first object, the object vector being a one-hot vector, i.e., the server uniquely encodes the object identifier of the first object to obtain the object vector of the first object. The server can obtain an object embedding matrix of the first object based on the object identifier, the object embedding matrix including the mastery of the first object on the knowledge in the at least one second question. The server can multiply the object vector and the object embedding matrix to obtain the object feature of the first object.

[0083] For example, the first object is a student s, the one-hot vector of the first object is x = [1, 0, 0, 0, 0], and the object embedding matrix is A. The object feature of the first object is represented as U-x T A. Wherein T represents matrix transposition.

[0084] It should be noted that the object embedding matrix of the first object refers to the object embedding matrix common to the plurality of sample objects participating in model training, which is a parameter of the cognitive diagnosis model. The object embedding matrix includes the relevant features of the plurality of objects participating in training, and the object feature of each object can be obtained by multiplying the object vector of each object and the object embedding matrix.

[0085] 302. The server performs embedding processing on a plurality of nodes in a knowledge graph based on the cognitive diagnosis model of the i-th iteration to obtain a plurality of node features of the plurality of nodes, wherein the nodes in the knowledge graph represent knowledge, the edges in the knowledge graph represent the dependency relationship between knowledge, and the node features include the features of the corresponding nodes and the features of the nodes connected to the corresponding nodes.

[0086] In the embodiments of the present application, the knowledge graph includes a plurality of nodes, each node representing a knowledge, and the edges between the nodes representing the dependency relationship between the knowledge. Wherein the knowledge graph is a directed graph, and the edges in the knowledge graph are directed edges. The knowledge represented by the node pointing to the directed edge is the predecessor knowledge of the knowledge represented by the node pointed to by the directed edge. In other words, the knowledge represented by the node pointed to by the directed edge is the successor knowledge of the knowledge represented by the node pointing to the directed edge. It should be noted that since the edges in the knowledge graph can represent the knowledge precedence relationship, the knowledge graph can be referred to as a knowledge precedence relationship graph.

[0087] The server inputs the knowledge graph into the cognitive diagnosis model of the i-th iteration, and the embedding layer in the cognitive diagnosis model performs embedding processing on the plurality of nodes in the knowledge graph to obtain the node feature of each node in the knowledge graph. Wherein the node feature of each node is fused with the features of at least one neighbor node on the basis of the features of the node itself, and the neighbor node is a node connected to the node by an edge in the knowledge graph.

[0088] In some embodiments, taking any node in the knowledge graph as an example, the server can fuse the features of the neighbor nodes of the node with the features of the node to obtain the node features of the node. First, the server obtains at least one neighbor node connected to the node, and then based on the node weights of the at least one neighbor node, the features of the at least one neighbor node are weighted and summed with the features of the node to obtain the node features of the node. Wherein, the node weight is used to represent the importance of the corresponding neighbor node to the node. By fusing the features of the neighbor nodes with the features of the current node based on the node weight, the fusion process of knowledge with different influence degrees on the knowledge represented by the current node can be modeled, thereby improving the accuracy of the node features of the node.

[0089] In some embodiments, the embedding layer of the cognitive diagnosis model includes a graph attention network (GAT), and based on the graph attention network, the node features of each node in the knowledge graph can be obtained. First, the server inputs the knowledge graph and the initial node matrix of each node in the knowledge graph into the embedding layer of the cognitive diagnosis model, wherein the initial node matrix of a node is used to represent the features of the node itself, that is, the features before fusing the features of the neighbor nodes. Then, for each node, the node matrix of the node is updated by fusing the features of the neighbor nodes of the node to obtain the node features of the node. It should be noted that the neighbor nodes of a node include first-order neighbor nodes, that is, nodes directly connected to the node through edges; the neighbor nodes of the node also include second-order neighbor nodes, that is, nodes connected to the first-order neighbor nodes of the node through edges; the neighbor nodes of the node also include M-order neighbor nodes, that is, nodes connected to the M-1 order neighbor nodes of the node through edges. M is a positive integer greater than 1. The embodiments of the present application take the neighbor nodes of a node including first-order neighbor nodes and second-order neighbor nodes as an example.

[0090] Wherein, since the knowledge graph is a directed graph, for any node, the neighbor nodes of the node are divided into two categories: predecessor neighbor nodes and successor neighbor nodes. The knowledge represented by the predecessor neighbor nodes is the predecessor knowledge of the knowledge represented by the node, and the knowledge represented by the successor neighbor nodes is the successor knowledge of the knowledge represented by the node. However, the influence degree of the predecessor knowledge and the successor knowledge on the knowledge is different, and the influence degree of different predecessor knowledge on the knowledge is also different. For example, "multiplication" and "first-order derivative" are both predecessor knowledge of "high-order derivative", but the influence degree on "high-order derivative" is obviously different. The embodiments of the present application model the fusion process of knowledge with different influence degrees by setting one or more graph attention layers in the embedding layer of the cognitive diagnosis model, thereby obtaining the node features of the node.

[0091] The following example, using a two-layer graph attention layer, illustrates the process of embedding multiple nodes.

[0092] First, obtain the adjacency matrix of the knowledge graph.

[0093] Secondly, each graph attention layer is used to update the features of the input nodes. Taking the first graph attention layer as an example, its input is the initial set of node features. in, Let represent the initial feature of the k-th node, F represent the number of features included in the initial feature set, and K represent the number of nodes. k, K, and F are positive integers. The output is... in, Let F' represent the updated feature of the k-th node, and let F' represent the number of features included in the updated feature. F' is a positive integer. F and F' can be the same or different. In this embodiment, F and F' are the same.

[0094] Furthermore, to calculate the weight of each neighboring node, a shared weight matrix W based on F×F′ is applied to each node. This shared weight matrix W represents the relationship between the node's input features and its output features:

[0095] Secondly, for each node, a self-attention mechanism 'a' is executed, where 'a' represents R. F′ ×R F′ The mapping to R yields the attention coefficients. e ij The value represents the importance of the knowledge represented by node j to the knowledge represented by node i, where node j is a neighbor node of node i.

[0096] Furthermore, to facilitate the calculation and comparison of attention coefficients, the attention coefficients of each neighbor node are normalized. The normalization method is shown in Formula (1), and the output node i is shown in Formula (2).

[0097]

[0098]

[0099] Where, α ij The node weight of the j-th neighbor of node i is represented by _{j}_; softmax(·) represents the normalization function; exp(·) represents the exponential function; N i Let n represent the set of neighboring nodes of node i; n represents N. i The nth neighbor node; e ij This indicates the importance of the knowledge represented by node j to the knowledge represented by node i; e in This indicates the importance of the knowledge represented by node n to the knowledge represented by node i; denotes the updated feature of node i, and σ (·) denotes a nonlinear function; j denotes the jth neighbor node of N i denotes the jth neighbor node in the ith round of iteration; W denotes a shared weight matrix; denotes the updated feature of node i, and σ (·) denotes a nonlinear function; j denotes the jth neighbor node of N

[0100] In some embodiments, in order to stabilize the learning process of self-attention, a multi-head attention mechanism is used, and the above formula (2) is adjusted. The adjusted formula is shown in formula (3):

[0101]

[0102] wherein, denotes the updated feature of node i, and σ (·) denotes a nonlinear function; T denotes the number of heads of multi-head attention, t and T are positive integers; N i denotes the set of neighbor nodes of node i; a ij denotes the node weight of the jth neighbor node of node i; W denotes a shared weight matrix; denotes the updated feature of node i, and σ (·) denotes a nonlinear function; j denotes the jth neighbor node of N

[0103] Finally, the node features of multiple nodes are represented in the form of a matrix: H e R K×d wherein, K denotes the number of nodes, and d denotes the embedding dimension of node features.

[0104] For example, Figure 4 is a schematic diagram for embedding processing of a knowledge graph according to an embodiment of the present application. As shown in Figure 4 , the cognitive diagnosis model performs embedding processing on the knowledge graph based on the graph attention network in the embedding layer, and obtains a node feature matrix representing the features of multiple nodes in the knowledge graph.

[0105] 303. The server performs embedding processing on the first question based on the cognitive diagnosis model of the ith round of iteration, and obtains question features of the first question.

[0106] In the embodiment of the present application, the first question is a sample question answered by the first object. The server inputs the question identification of the first question into the cognitive diagnosis model of the ith round of iteration, and performs embedding processing on the first question by the embedding layer in the cognitive diagnosis model to obtain the question features of the first question.

[0107] In some embodiments, the question characteristics of the first question include a discrimination characteristic, a question knowledge characteristic, and a difficulty characteristic, the discrimination characteristic is used to represent discrimination of the first question, the question knowledge characteristic is used to represent knowledge contained in the first question, and the difficulty characteristic is used to represent difficulty of the first question. The server can determine a question vector of the first question based on the question identification of the first question, the question vector being a one-hot vector. The server can obtain a discrimination embedding matrix, a question knowledge embedding matrix, and a difficulty embedding matrix of the first question based on the question identification. The server can determine the discrimination characteristic, the question knowledge characteristic, and the difficulty characteristic based on the question vector, the discrimination embedding matrix, the question knowledge embedding matrix, and the difficulty embedding matrix. The server can multiply the question vector and the discrimination embedding matrix to obtain the discrimination characteristic. The server can multiply the question vector and the question knowledge embedding matrix to obtain the question knowledge characteristic. The server can multiply the question vector and the difficulty embedding matrix to obtain the difficulty characteristic. The discrimination embedding matrix represents discrimination of a question; the question knowledge embedding matrix represents an association relationship between a question and knowledge; and the difficulty matrix represents difficulty of a question. In some embodiments, the association relationship between a question and knowledge is marked by a domain expert or a technical personnel in the domain.

[0108] For example, the first question is E, the one-hot vector of the first question is x [0, 0, 1, 0, 0], the discrimination embedding matrix is D, the question knowledge embedding matrix is Q, and the difficulty embedding matrix is B. The discrimination characteristic of the first question is represented as DISC = x T D; the question knowledge characteristic of the first question is represented as Q E = x T Q; and the difficulty characteristic of the first question is represented as DIFF = x T B. Wherein, T represents matrix transposition.

[0109] It should be noted that the discrimination embedding matrix, the question knowledge embedding matrix, and the difficulty embedding matrix of the first question refer to the discrimination embedding matrix, the question knowledge embedding matrix, and the difficulty embedding matrix common to a plurality of sample questions participating in model training, which are parameters of the cognitive diagnosis model. By multiplying the question vector of each question and the discrimination embedding matrix, the question knowledge embedding matrix, and the difficulty embedding matrix respectively, the discrimination characteristic, the question knowledge characteristic, and the difficulty characteristic of each question can be obtained.

[0110] It should be noted that the steps 301 to 303 are implemented by the server based on an embedding layer in the cognitive diagnosis model, and the main purpose of the embedding layer is to obtain the object feature, the plurality of node features of the plurality of nodes in the knowledge graph, and the question feature. When obtaining the plurality of node features, a graph attention network, a graph convolutional neural network, or other graph neural network model can be used, and the embodiments of the present application do not limit this. In some embodiments, the server can also construct a unified "object-question-knowledge" interaction heterogeneous graph based on the object answer record, the question-knowledge relationship matrix, and the knowledge graph and the like, and then use a related graph embedding model such as HAN (Hierarchical Attention Network, hierarchical attention network) to learn the object feature, the question feature, and the node feature. Then, according to the question feature, the fitting ability of the neural network can be used to map the question feature to the difficulty feature, the discrimination feature, and the question knowledge feature. The above object feature, node feature, difficulty feature, discrimination feature, and question knowledge feature are input into the fusion layer of the cognitive diagnosis model for further processing.

[0111] 304. The server fuses the object feature and the plurality of node features based on the cognitive diagnosis model of the i th round of iteration to obtain a knowledge state feature, and the knowledge state feature is used to represent the mastery degree of the first object to the knowledge represented by the plurality of nodes.

[0112] In the embodiments of the present application, the server can input the above object feature and the plurality of node features into the fusion layer of the cognitive diagnosis model, and output the mastery degree of the first object to the knowledge represented by the plurality of nodes. The mastery degree of the object to a certain knowledge is affected by the prerequisite knowledge of the knowledge, so when determining the mastery degree of the first object to a certain knowledge, the node feature of the knowledge and the node feature of the prerequisite knowledge of the knowledge need to be considered. The prerequisite knowledge of a knowledge refers to the prerequisite or premise for mastering the knowledge, for example, the prerequisite knowledge of two-digit addition is one-digit addition, and the prerequisite knowledge of multiplication is addition.

[0113] In some embodiments, the server splices the object feature and each node feature respectively, and then fully connects the plurality of features spliced to obtain the above knowledge state feature. For any node, the server can splice the object feature, the node feature of the node, and the feature of the predecessor node of the node based on the cognitive diagnosis model of the i th round of iteration to obtain the spliced feature of the node, and the knowledge represented by the predecessor node is the prerequisite knowledge of the knowledge represented by the node. Then the server fully connects the spliced features of the plurality of nodes based on the cognitive diagnosis model of the i th round of iteration to obtain the knowledge state feature.

[0114] It should be noted that, due to the number of pre-knowledge of each knowledge is different, that is, the number of predecessor nodes of each node is different, resulting in different dimensions of the spliced features, the embodiment of the application fills the spliced features with zero vectors of the same dimension as the object features to fill the spliced features to the same dimension. It should be noted that the dimension of the object feature is the same as the dimension of the node feature.

[0115] For example, referring to Figure 5 , Figure 5 is a schematic diagram of feature fusion provided by an embodiment of the application. Taking a knowledge graph including four nodes a, b, c and d as an example, the node features of the four nodes are represented as , and The object feature is represented as When the object feature is spliced with the four node features respectively, since node a has no predecessor node, the spliced and The predecessor node of node b is node a, so the splicing is performed in the order of and Node c has no predecessor node, so the splicing is performed in the order of Then, the zero vector is used to fill the spliced features to obtain the spliced features of the multiple nodes. Finally, the spliced features of the multiple nodes are input into a fully connected neural network to obtain the knowledge state feature.

[0116] For example, taking node d as an example, the first object s represents the mastery degree of the knowledge represented by node d, referring to formulas (4) and (5) shown as follows:

[0117]

[0118] KS d = sigmoid(W1xX+b1) (5);

[0119] Wherein, X represents the spliced feature of node d; represents the object feature of the first object; represents vector splicing; represents the node feature of node d; represents the node feature of node a; represents the node feature of node b; represents the node feature of node c; KS d represents the mastery degree of the knowledge represented by node d by the first object s; sigmoid(·) represents the sigmoid function; W1 and b1 represent the parameters of the fully connected layer.

[0120] It should be noted that this step is implemented by the server based on the fusion layer in the cognitive diagnosis model, and the main purpose of the fusion layer is to obtain the mastery degree of the object on the knowledge represented by each node according to the object features and the node features. The above splicing operation of the features can be replaced by the way of bit-by-bit multiplication of vectors. For example, when calculating the mastery degree of the first object on the knowledge represented by node d, formula (4) is replaced by: wherein, represents bit-by-bit multiplication of vectors. Then the mastery degree of the first object on the knowledge represented by node d is obtained through the neural network. For other nodes, the dimensions of the features X that can be calculated are the same, so that the step of filling the zero vector can be omitted. If the values of each dimension of X obtained by using the way of bit-by-bit multiplication of vectors are relatively small, the values of each dimension of X can be scaled to a suitable range in a standardization manner, and the embodiments of the present application do not limit this.

[0121] 305、The server trains the cognitive diagnosis model of the i-th iteration based on the question features and the knowledge state features, taking the answer result of the first object to the first question as supervision information.

[0122] In the embodiments of the present application, the server can input the above question features and knowledge state features into the diagnosis layer of the cognitive diagnosis model, and output the probability that the first object can correctly answer the first question. Taking the answer result of the first object to the first question as supervision information, the model parameters of the cognitive diagnosis model are adjusted according to the difference between the predicted probability and the answer result. Wherein, the answer result is correct answer or wrong answer.

[0123] In some embodiments, the server can obtain a feature difference value by subtracting the knowledge state features from the difficulty features, then multiply the question knowledge features with the feature difference value bit by bit to obtain a Hadamard product, multiply the Hadamard product with the discrimination features to obtain the object answering features. Then the server predicts the object answering features based on the cognitive diagnosis model of the i-th iteration to obtain the prediction result information, which is used to indicate the probability that the first object correctly answers the first question. Then the server adjusts the model parameters of the cognitive diagnosis model of the i-th iteration based on the difference between the prediction result information and the answer result of the first object to the first question.

[0124] Wherein, the diagnosis layer of the cognitive diagnosis model can predict the performance of the sample object on the sample question based on the item response theory and the neural network. The following takes the prediction of the performance of the first object on the first question as an example for description: the one-hot vector of the first question E is x[0, 0, 1, 0, 0], the difficulty feature is DIFF=x T B, the discrimination feature is DISC=x T D, the question knowledge feature is Q E =xT Q. Then the first object correctly answers the first question probability of the subject is fitted based on three full connection layers in the diagnosis layer. Wherein, the input of the first full connection layer is shown in formula (6):

[0125] Y=Q E o (KS-DIFF) x DISC (6);

[0126] Wherein, Y represents the object answer characteristics; Q E represents the question knowledge characteristics; o represents the vector bit multiplication; KS represents the knowledge state characteristics; DIFF represents the difficulty characteristics; DISC represents the discrimination characteristics.

[0127] The intermediate output of the full connection layer and the final output of the prediction result information are shown in formula (7)-formula (9):

[0128] f1=sigmoid(W2xY+b2) (7);

[0129] f2=sigmoid(W3xf1+b3) (8);

[0130] y=sigmoid(W4xf2+b4) (9);

[0131] Wherein, f1 represents the output of the first full connection layer; sigmoid(·) represents the sigmoid function; Y represents the object answer characteristics; W2 and b2 represent the parameters of the first full connection layer; f2 represents the output of the second full connection layer; W3 and b3 represent the parameters of the second full connection layer; y represents the prediction result information; W4 and b4 represent the parameters of the third full connection layer.

[0132] In the embodiment of the application, for the above self-supervised learning process, the model is trained by using cross entropy loss, which is the loss of the predicted sample object correctly answering the sample test question. For convenience of description, the cross entropy loss is called the first loss. The calculation method of the first loss is shown in formula (10):

[0133]

[0134] Wherein, loss1 represents the first loss; r sE represents the true result of the object s answering the question E; y sE represents the probability of the model predicting the object s correctly answering the question E.

[0135] It should be noted that the present step is implemented by the server based on the diagnosis layer in the cognitive diagnosis model, which utilizes the true results of the object answering the questions and the predicted results obtained by the neural network to realize the training of the model. In some embodiments, other methods can also be combined to obtain the final predicted results, such as IRT (Item Response Theory) and MIRT (Multidimensional Item Response Theory).

[0136] wherein for IRT, the formula is Z = σ ((h-diff) x disc), wherein σ (·) represents a nonlinear function; h, diff, and disc are all scalars, representing the object ability value, the question difficulty, and the discrimination, respectively. By using one or more layers of neural networks, the knowledge state feature, the difficulty feature, and the discrimination feature in the present application can be mapped into a one-dimensional vector to replace the above-mentioned scalars.

[0137] wherein for MIRT, the formula is z = σ (Q E x (h s -h E )), wherein σ (·) represents a nonlinear function; Q E represents the question knowledge feature; h s represents the latent feature vector of the object; h E represents the latent feature vector of the question. By using KS representing the knowledge state feature and DIFF representing the difficulty feature in the present application, direct replacement is achieved.

[0138] The edges in the knowledge graph are obtained by manual annotation. Since the workload of manual annotation is large, the edges in the knowledge graph are not perfect, that is, the edges in the knowledge graph cannot reflect the knowledge precedence relationship between the knowledge represented by all nodes. Therefore, the server can complete the knowledge graph based on the completion layer in the cognitive diagnosis model of the i-th iteration. The server first obtains the edge matrix of the knowledge graph based on the cognitive diagnosis model of the i-th iteration, and the elements in the edge matrix are used to represent the edges between multiple nodes in the knowledge graph. Then the server predicts the edge matrix based on the cognitive diagnosis model of the i-th iteration to obtain edge probability information, which is used to represent the probability of the existence of an edge between any two nodes in the multiple nodes. Finally, the server adds edges to the above knowledge graph based on the edge probability information. By completing the knowledge graph, the cost of manually annotating the knowledge graph can be reduced, the knowledge precedence relationship in the knowledge graph can be improved, and the knowledge state of the student can be better diagnosed.

[0139] The input to the completion layer is the edge matrix of the knowledge graph, which has a size of K×K×2d. The elements in the matrix represent the edge embedding between any two nodes in the knowledge graph. The output of the completion layer is the adjacency matrix of the completed knowledge graph, which has a size of K×K, where K represents the number of nodes in the knowledge graph and K is a positive integer.

[0140] The following example uses the prediction of the edge between node a and node b. Taking the probability of existence as an example, we can illustrate this by showing that the knowledge represented by node a is the precursor knowledge of the knowledge represented by node b. The node feature of node a is represented by H. a The node characteristics of node b are represented as H. b ,Will Represented as This involves concatenating the node features of nodes a and b. By concatenating the features of any two nodes in the knowledge graph, an edge matrix is ​​obtained. An edge exists between nodes a and b. For the calculation of probability, please refer to formula (11):

[0141]

[0142] Among them, P ab Represents the edge between node a and node b. The probability of existence; sigmoid(·) represents the sigmoid function; W5 and b5 represent the parameters of the fully connected layer; H a Represents the node characteristics of node a; H represents vector concatenation; b This represents the node characteristics of node b.

[0143] In some embodiments, the server can adjust the model parameters of the cognitive diagnostic model in the i-th iteration based on the difference between the aforementioned edge probability information and the existing edges in the knowledge graph. That is, in this embodiment, for the aforementioned process of completing the knowledge graph, the cognitive diagnostic model is trained by taking the maximum likelihood of all tag node pairs as the objective function. Here, the tag node pair is used to represent the existing edges in the knowledge graph. The objective function is shown in the following formula (12):

[0144]

[0145] Where there are edges in the knowledge graph Then r ab =1; P ab Represents the edge between node a and node b. The probability of its existence.

[0146] For ease of description, the loss in the above completion process is called the second loss. The calculation method for this second loss is shown in formula (13):

[0147]

[0148] wherein, loss = represents the second loss; if there is an edge between nodes a and b in the knowledge graph then r ab = 1; P ab represents the probability of the existence of an edge between nodes a and b.

[0149] It should be noted that, for the above process of completing the knowledge graph, if the number of labeled node pairs is small, an additional target can be set for the unlabeled node pairs and an additional loss is calculated. Then the additional loss and the loss of the labeled node pairs are weighted and added, as the final loss of the completion task, and the model is trained based on the final loss.

[0150] For example, the possibility of an edge between the unlabeled node pairs is limited to be less than that of the artificially labeled node pairs, i.e. P ab <P ij , wherein ab is an unlabeled node pair, and ij is a labeled node pair. The above relationship is limited by the following formula (14):

[0151]

[0152] wherein, P(P ij >P ab ) is a target function; P ab represents the probability of the existence of an edge between the unlabeled node pairs ab; P ij represents the probability of the existence of an edge between the labeled node pairs ij, and λ represents a hyperparameter.

[0153] The loss function is shown in formula (15):

[0154]

[0155] wherein, loss3 represents the loss of the unlabeled node pairs, P ab represents the probability of the existence of an edge between the unlabeled node pairs ab; P ij represents the probability of the existence of an edge between the labeled node pairs ij, and λ represents a hyperparameter.

[0156] The loss loss3 of the above unlabeled node pairs and the loss loss2 of the labeled node pairs are weighted and added as the final loss of the completion task.

[0157] ​It should be noted that the steps 301 to 305 above exemplarily show the process of the i-th iteration in the multiple iterations. The following takes the first object as a student object as an example to introduce the preparation process of the training data. Before the training starts, first, the answering logs of multiple students are obtained, and then the answering logs are divided into training data, validation data and test data according to the ratio of 8:1:1, the training set is constructed based on the training data, the validation set is constructed based on the validation data, and the test set is constructed based on the test data. During training, the sample data in the training set is shuffled, and then is sent into the model in batches, and the model parameters of the cognitive diagnosis model are updated based on the back propagation of the first loss and the second loss. After each round of training, the current cognitive diagnosis model is tested on the validation set and the test set, the hyperparameters of the model are adjusted according to the results on the validation set to improve the performance of the model, and the best model on the test set is saved.

[0158] In some embodiments, the server can train the cognitive diagnosis model in a joint training manner, take the weighted sum of the first loss and the second loss as the final loss, and perform model training based on the final loss. The calculation method of the final loss is shown in formula (16):

[0159] loss = a x loss1 + (1-a) x loss2 (16);

[0160] Wherein, loss represents the final loss of joint training; a represents a weight parameter; loss1 represents the first loss; loss2 represents the first loss.

[0161] In some embodiments, the server can train the cognitive diagnosis model in an alternating training manner, that is, for different batches and different rounds of training data, the first loss and the second loss are alternately used for model training.

[0162] For example, for the current 10 batches of training data, the first loss is used for training, and the next 10 batches of training data are trained using the second loss. Alternatively, the i-th round is trained using the first loss, and the i+1-th round is trained using the second loss.

[0163] It should be noted that the number of layers, activation functions and the number of neurons in the fully connected layer involved in the embodiments of the present application can be replaced by other reasonable hyperparameters, which are not limited by the embodiments of the present application.

[0164] It should be noted that in order to make the cognitive diagnosis model provided by the embodiments of the present application easier to understand, refer to Figure 6 Figure 6 is a model framework diagram of a cognitive diagnosis model according to the embodiments of the present application. Refer to Figure 6 ​As shown, the cognitive diagnosis model includes an embedding layer, a fusion layer, a diagnosis layer, and a completion layer. The embedding layer is used to perform embedding processing on the sample object, the knowledge graph, and the sample question, and outputs object features, a node feature matrix, question knowledge features, difficulty features, and discrimination features, where the node feature matrix includes node features of multiple nodes. The fusion layer is used to concatenate the object features and the node features respectively to obtain a concatenation matrix, which includes concatenation features of multiple nodes. The concatenation matrix is fully connected based on a fully connected layer to obtain knowledge state features. Based on the above question knowledge features, difficulty features, discrimination features, and knowledge state features, object answering features are obtained. The diagnosis layer is used to predict the above object answering features based on a multi-layer perceptron to obtain prediction result information. The input of the completion layer is an edge matrix obtained based on the node feature matrix, and the output is an adjacency matrix after completion of the knowledge graph, based on which edges are added in the knowledge graph.

[0165] The scheme provided by the embodiments of the present application can make the dependent relationship between knowledge be added during training, so as to determine the mastery of knowledge by the object based on the answering situation of the object, and then train the cognitive diagnosis model, so that the cognitive diagnosis model obtained by training has high accuracy.

[0166] The above knowledge diagnosis model can be applied to various scenarios, such as question recommendation and learning situation report, as shown in Figure 7 As shown, Figure 7 is an architecture schematic diagram of an application scenario provided by an embodiment of the present application. First, the data of the questions answered by the student is collected, which is derived from online adaptive evaluation, daily homework, examination data, and can also be derived from offline examination, homework, etc., and can be used after being input through a unified examination marking platform. The question has been labeled with labels such as involved knowledge and difficulty. Then the data of the questions answered by the student is input into the cognitive diagnosis model trained based on the method provided by the embodiments of the present application, to obtain the mastery of each knowledge by the student, that is, the learning situation portrait. In some embodiments, the mastery of knowledge by the student can also be discretized by setting different thresholds. For example, it is discretized into three categories of not mastered, partially mastered, and mastered, which can improve the visualization effect and facilitate subsequent test question recommendation. Figure 8 is a schematic diagram of a learning situation report provided by an embodiment of the present application. The learning situation report is generated based on the student portrait and the historical answering data of the student, as shown in Figure 8 As shown, the student has mastered 17 pieces of knowledge, has not mastered 19 pieces of knowledge, and has partially mastered 7 pieces of knowledge, Figure 8 It also includes the knowledge details of the student's examination. In addition toFigure 8 In addition to the content shown, the learning situation report can also show the student's learning duration, answer duration, and horizontal and vertical comparisons, and the present application will not be listed one by one.

[0167] The following will be described taking the recommendation of questions according to the student's mastery of knowledge as an example. Figure 9 is a flowchart of a question recommendation method provided by an embodiment of the present application, referring to Figure 9 As shown in the figure, in the embodiment of the present application, it is described taking the terminal execution as an example. The question recommendation method includes the following steps:

[0168] 901, the terminal performs cognitive diagnosis on the second object based on the target cognitive diagnosis model, and obtains the target knowledge state feature of the second object.

[0169] In the embodiment of the present application, the target cognitive diagnosis model is trained based on the node features of the multiple nodes in the knowledge graph, the nodes in the knowledge graph represent knowledge, the edges in the knowledge graph represent the dependency relationship between knowledge, the node features include the features of the corresponding nodes and the features of the nodes connected to the corresponding nodes, and the target knowledge state feature is used to represent the mastery degree of the knowledge represented by the multiple nodes by the second object.

[0170] 902, the terminal determines at least one question title to be recommended based on the target knowledge state feature.

[0171] In the embodiment of the present application, the terminal can determine at least one question title to be recommended according to the student's mastery of knowledge, such as recommending a question containing weakly mastered knowledge to the student.

[0172] Referring to Figure 10 As shown in the figure, Figure 10 is a schematic diagram of a question recommendation interface provided by an embodiment of the present application. Figure 10 Exemplarily shows the knowledge mastery of the student account currently logged in by the terminal and the question recommendation control. By triggering the question recommendation control, at least one question title to be recommended can be determined.

[0173] 903, the terminal recommends at least one question title to the second object.

[0174] In the embodiment of the present application, the terminal can display the recommended at least one question title online in real time, and can also format the at least one question title into the form of a test paper for printing.

[0175] Figure 11 is a block diagram of a training device of a cognitive diagnosis model provided by an embodiment of the present application. The device is used to execute the training method of the cognitive diagnosis model, referring to Figure 11As shown, the apparatus comprises: an embedding module 1101, a fusion module 1102, and a training module 1103.

[0176] The embedding module 1101 is configured to perform embedding processing on a first object, a plurality of nodes in a knowledge graph, and a first question based on a cognitive diagnosis model of an i th iteration, to obtain an object feature of the first object, a plurality of node features of the plurality of nodes, and a question feature of the first question, the object feature being used to represent a mastery of knowledge of the first object in at least one second question, the nodes in the knowledge graph representing knowledge, and edges in the knowledge graph representing dependency relationships between knowledge, and the node features comprising features of corresponding nodes and features of nodes connected to the corresponding nodes.

[0177] The fusion module 1102 is configured to fuse the object feature and the plurality of node features based on the cognitive diagnosis model of the i th iteration, to obtain a knowledge state feature, the knowledge state feature being used to represent a mastery degree of the first object to knowledge represented by the plurality of nodes.

[0178] The training module 1103 is configured to train the cognitive diagnosis model of the i th iteration based on the question feature and the knowledge state feature, with a result of an answer of the first object to the first question as supervision information, the cognitive diagnosis model being used to make a prediction based on input objects and questions to obtain a probability of the object correctly answering the question.

[0179] In some embodiments, the fusion module 1102 is configured to, for any node, splice the object feature, a node feature of the node, and a feature of a predecessor node of the node based on the cognitive diagnosis model of the i th iteration to obtain a spliced feature of the node, the predecessor node representing predecessor knowledge of knowledge represented by the node; and perform full connection on the spliced features of the plurality of nodes based on the cognitive diagnosis model of the i th iteration to obtain the knowledge state feature.

[0180] In some embodiments, the embedding module 1101 is configured to determine an object vector of the first object based on an object identifier of the first object; obtain an object embedding matrix of the first object based on the object identifier; and multiply the object vector and the object embedding matrix to obtain the object feature of the first object.

[0181] In some embodiments, the embedding module 1101 is configured to, for any node in the knowledge graph, obtain at least one neighbor node connected to the node; and perform weighted summation on features of the at least one neighbor node and the feature of the node based on node weights of the at least one neighbor node to obtain the node feature of the node, the node weights being used to represent importance of the corresponding neighbor node to the node.

[0182] In some embodiments, the question characteristics of the first question include a discrimination characteristic, a question knowledge characteristic, and a difficulty characteristic, and the question knowledge characteristic is used to represent knowledge contained in the first question.

[0183] The embedding module 1101 is configured to determine a question vector of the first question based on a question identifier of the first question, obtain a discrimination embedding matrix, a question knowledge embedding matrix, and a difficulty embedding matrix of the first question based on the question identifier, and determine the discrimination characteristic, the question knowledge characteristic, and the difficulty characteristic based on the question vector, the discrimination embedding matrix, the question knowledge embedding matrix, and the difficulty embedding matrix.

[0184] In some embodiments, the embedding module 1101 is configured to multiply the question vector and the discrimination embedding matrix to obtain the discrimination characteristic, multiply the question vector and the question knowledge embedding matrix to obtain the question knowledge characteristic, and multiply the question vector and the difficulty embedding matrix to obtain the difficulty characteristic.

[0185] In some embodiments, the training module 1103 is configured to obtain a feature difference by subtracting the difficulty characteristic from the knowledge state characteristic, multiply the question knowledge characteristic and the feature difference bit by bit to obtain a Hadamard product, multiply the Hadamard product and the discrimination characteristic to obtain an object answering characteristic, perform prediction on the object answering characteristic based on the cognitive diagnosis model of the i th iteration to obtain prediction result information, and adjust the model parameters of the cognitive diagnosis model of the i th iteration based on a difference between the prediction result information and an answer result of the first object to the first question.

[0186] In some embodiments, Figure 12 is a block diagram of another training device of a cognitive diagnosis model provided by an embodiment of the present application, as shown in Figure 12 The device further includes:

[0187] The obtaining module 1104 is configured to obtain an edge matrix of the knowledge graph based on the cognitive diagnosis model of the i th iteration, and an element in the edge matrix is used to represent an edge between the plurality of nodes.

[0188] The prediction module 1105 is configured to perform prediction on the edge matrix based on the cognitive diagnosis model of the i th iteration to obtain edge probability information, and the edge probability information is used to represent a probability that an edge exists between any two nodes in the plurality of nodes.

[0189] The graph processing module 1106 is configured to add an edge in the knowledge graph based on the edge probability information.

[0190] In some embodiments, as shown in Figure 12 The device further includes:

[0191] The parameter adjustment module 1107 is configured to adjust the model parameter of the cognitive diagnosis model in the i th iteration based on the difference between the edge probability information and the existing edge in the knowledge graph.

[0192] The training device for the cognitive diagnosis model provided in the embodiments of the present application can determine the mastery of the knowledge by the subject based on the answering condition of the subject, and then train the cognitive diagnosis model, so that the trained cognitive diagnosis model has high accuracy.

[0193] It should be noted that the training device for the cognitive diagnosis model provided in the above embodiments is only used as an example for the division of the above functional modules during model training. In actual applications, the above functions can be completed by different functional modules according to needs, that is, the internal structure of the device is divided into different functional modules to complete all or part of the above described functions. In addition, the training device for the cognitive diagnosis model and the training method for the cognitive diagnosis model provided in the above embodiments belong to the same concept, and the specific implementation process is described in the method embodiments, which will not be repeated here.

[0194] Figure 13 is a block diagram of a question recommendation device according to an embodiment of the present application. The device is used to execute the above question recommendation method, as shown in Figure 13 The device includes a subject processing module 1301, a determination module 1302, and a recommendation module 1303.

[0195] The subject processing module 1301 is configured to perform cognitive diagnosis on a second subject based on a target cognitive diagnosis model to obtain a target knowledge state feature of the second subject. The target cognitive diagnosis model is trained based on node features of multiple nodes in a knowledge graph. The nodes in the knowledge graph represent knowledge, and the edges in the knowledge graph represent the dependency relationship between the knowledge. The node features include the features of the corresponding nodes and the features of the nodes connected to the corresponding nodes. The target knowledge state feature is used to represent the mastery of the knowledge represented by the multiple nodes by the second subject.

[0196] The determination module 1302 is configured to determine at least one target question to be recommended based on the target knowledge state feature.

[0197] The recommendation module 1303 is configured to recommend the at least one target question to the second subject.

[0198] It should be noted that the question recommendation device provided in the above embodiments only takes the above division of functional modules as an example when recommending questions, and in actual application, the above functions can be completed by different functional modules according to needs, that is, the internal structure of the device is divided into different functional modules to complete all or part of the functions described above. In addition, the question recommendation device and the question recommendation method provided in the above embodiments belong to the same concept, and the specific implementation process is described in the method embodiments, which will not be repeated here.

[0199] In the embodiments of the present application, the computer device can be configured as a terminal or a server. When the computer device is configured as a terminal, the technical solutions provided by the embodiments of the present application can be implemented by the terminal as an execution subject. When the computer device is configured as a server, the technical solutions provided by the embodiments of the present application can be implemented by the server as an execution subject. The technical solutions provided by the present application can also be implemented through the interaction between the terminal and the server, and the embodiments of the present application do not limit this.

[0200] When the computer device is configured as a terminal, Figure 14 is a structural block diagram of a terminal 1400 according to an embodiment of the present application. The terminal 1400 can be a portable mobile terminal, such as a smartphone, a tablet computer, an MP3 player (Moving Picture Experts Group Audio Layer III), an MP4 player (Moving Picture Experts Group Audio Layer IV), a notebook computer or a desktop computer. The terminal 1400 can also be referred to as a user equipment, a portable terminal, a laptop terminal, a desktop terminal, or other names.

[0201] Generally, the terminal 1400 includes a processor 1401 and a memory 1402.

[0202] The processor 1401 can include one or more processing cores, such as a 4-core processor, an 8-core processor, and the like. The processor 1401 can be implemented in at least one of a hardware form of a DSP (Digital Signal Processing), an FPGA (Field-Programmable Gate Array), a PLA (Programmable Logic Array). The processor 1401 can also include a main processor and a coprocessor, the main processor being a processor for processing data in an awake state, also known as a CPU (Central Processing Unit), and the coprocessor being a low-power processor for processing data in a standby state. In some embodiments, the processor 1401 can be integrated with a GPU (Graphics Processing Unit) for rendering and drawing content required to be displayed by the display screen. In some embodiments, the processor 1401 can further include an AI (Artificial Intelligence) processor for processing computing operations related to machine learning.

[0203] The memory 1402 can include one or more computer-readable storage media that can be non-transitory. The memory 1402 can also include high-speed random access memory and nonvolatile, computer-readable storage media such as one or more magnetic disk storage devices, flash memory devices. In some embodiments, the non-transitory computer-readable storage medium in the memory 1402 is used to store at least one computer program for being executed by the processor 1401 to implement the topic recommendation method provided by the method embodiments in the present application.

[0204] In some embodiments, the terminal 1400 can also optionally include a peripheral device interface 1403 and at least one peripheral device. The processor 1401, the memory 1402, and the peripheral device interface 1403 can be connected through a bus or a signal line. Each peripheral device can be connected to the peripheral device interface 1403 through a bus, a signal line, or a circuit board. Specifically, the peripheral device includes at least one of a radio frequency circuit 1404, a display screen 1405, a camera assembly 1406, an audio circuit 1407, a positioning assembly 1408, and a power supply 1409.

[0205] The peripheral interface 1403 can be used to connect at least one I / O (Input / Output) related peripheral device to the processor 1401 and the memory 1402. In some embodiments, the processor 1401, the memory 1402 and the peripheral interface 1403 are integrated on the same chip or circuit board; in some other embodiments, any one or two of the processor 1401, the memory 1402 and the peripheral interface 1403 can be implemented on a separate chip or circuit board, and the present embodiments are not limited in this regard.

[0206] The radio frequency circuit 1404 is used to receive and send RF (Radio Frequency) signals, also known as electromagnetic signals. The radio frequency circuit 1404 communicates with the communication network and other communication devices through electromagnetic signals. The radio frequency circuit 1404 converts electrical signals into electromagnetic signals for transmission, or converts received electromagnetic signals into electrical signals. In some embodiments, the radio frequency circuit 1404 includes an antenna system, an RF transceiver, one or more amplifiers, a tuner, an oscillator, a digital signal processor, a codec chipset, a subscriber identity module card, and the like. The radio frequency circuit 1404 can communicate with other terminals through at least one wireless communication protocol. The wireless communication protocol includes but is not limited to: the World Wide Web, a metropolitan area network, an intranet, various generations of mobile communication networks (2G, 3G, 4G and 5G), a wireless local area network and / or a WiFi (Wireless Fidelity) network. In some embodiments, the radio frequency circuit 1404 can also include NFC (Near Field Communication) related circuitry, and the present application is not limited in this regard.

[0207] The display screen 1405 is configured to display a UI (User Interface). The UI can include graphics, text, icons, video, and any combination thereof. When the display screen 1405 is a touch display screen, the display screen 1405 is further configured to capture touch signals on or above the surface of the display screen 1405. The touch signals can be input to the processor 1401 as control signals for processing. In this case, the display screen 1405 can also be configured to provide virtual buttons and / or virtual keyboard, also known as soft buttons and / or soft keyboard. In some embodiments, the display screen 1405 can be one, disposed on the front panel of the terminal 1400; in other embodiments, the display screen 1405 can be at least two, respectively disposed on different surfaces of the terminal 1400 or in a folding design; in other embodiments, the display screen 1405 can be a flexible display screen, disposed on a curved surface or a folding surface of the terminal 1400. Even, the display screen 1405 can also be disposed in an irregular shape, i.e., a special-shaped screen. The display screen 1405 can be made of LCD (Liquid Crystal Display), OLED (Organic Light-Emitting Diode), etc.

[0208] The camera assembly 1406 is configured to capture images or videos. In some embodiments, the camera assembly 1406 includes a front camera and a rear camera. Typically, the front camera is disposed on the front panel of the terminal, and the rear camera is disposed on the back of the terminal. In some embodiments, the rear camera is at least two, which is any one of a main camera, a depth-of-field camera, a wide-angle camera, and a telephoto camera, to realize the background blur function by fusing the main camera and the depth-of-field camera, the panoramic shooting and VR (Virtual Reality) shooting function by fusing the main camera and the wide-angle camera, or other fusion shooting functions. In some embodiments, the camera assembly 1406 can further include a flash. The flash can be a single-color temperature flash or a dual-color temperature flash. The dual-color temperature flash refers to the combination of a warm light flash and a cold light flash, which can be used for light compensation under different color temperatures.

[0209] The audio circuit 1407 can include a microphone and a speaker. The microphone is used to collect sound waves of a user and an environment, and convert the sound waves into an electrical signal input to the processor 1401 for processing, or input to the radio frequency circuit 1404 to realize voice communication. For the purpose of stereo sound collection or noise reduction, the microphone can be multiple, respectively arranged at different parts of the terminal 1400. The microphone can also be an array microphone or an omnidirectional collection type microphone. The speaker is used to convert the electrical signal from the processor 1401 or the radio frequency circuit 1404 into sound waves. The speaker can be a traditional diaphragm speaker, or a piezoelectric ceramic speaker. When the speaker is a piezoelectric ceramic speaker, not only can the electrical signal be converted into a sound wave that a human can hear, but also can be converted into a sound wave that a human cannot hear for the purpose of ranging, etc. In some embodiments, the audio circuit 1407 can also include a headphone jack.

[0210] The positioning component 1408 is used to position the current geographic position of the terminal 1400 to realize navigation or LBS (Location Based Service). The positioning component 1408 can be a positioning component based on the GPS (Global Positioning System) of the United States, the Beidou system of China or the Galileo system of Russia.

[0211] The power supply 1409 is used to supply power to each component in the terminal 1400. The power supply 1409 can be alternating current, direct current, disposable battery or rechargeable battery. When the power supply 1409 includes a rechargeable battery, the rechargeable battery can be a wired charging battery or a wireless charging battery. The wired charging battery is a battery charged through a wired line, and the wireless charging battery is a battery charged through a wireless coil. The rechargeable battery can also be used to support fast charging technology.

[0212] In some embodiments, the terminal 1400 further includes one or more sensors 1410. The one or more sensors 1410 include but are not limited to: an acceleration sensor 1411, a gyroscope sensor 1412, a pressure sensor 1413, a fingerprint sensor 1414, an optical sensor 1415 and a proximity sensor 1416.

[0213] The acceleration sensor 1411 can detect the acceleration size in three coordinate axes of the coordinate system established by the terminal 1400. For example, the acceleration sensor 1411 can be used to detect the components of the gravitational acceleration in three coordinate axes. The processor 1401 can control the display screen 1405 to display the user interface in a landscape view or a portrait view according to the gravitational acceleration signal collected by the acceleration sensor 1411. The acceleration sensor 1411 can also be used for game or user motion data collection.

[0214] The gyroscope sensor 1412 can detect the body direction and rotation angle of the terminal 1400, and can collect 3D motions of a user on the terminal 1400 in cooperation with the acceleration sensor 1411. The processor 1401 can implement the following functions according to the data collected by the gyroscope sensor 1412: motion sensing (e.g., changing a UI according to a tilt operation of the user), image stabilization during shooting, game control, and inertial navigation.

[0215] The pressure sensor 1413 can be disposed on a side frame of the terminal 1400 and / or under the display screen 1405. When the pressure sensor 1413 is disposed on the side frame of the terminal 1400, a grip signal of a user on the terminal 1400 can be detected, and left / right hand recognition or a shortcut operation can be performed by the processor 1401 according to the grip signal collected by the pressure sensor 1413. When the pressure sensor 1413 is disposed under the display screen 1405, a pressure operation of a user on the display screen 1405 can be detected by the processor 1401, and an operable control on a UI can be controlled according to the pressure operation. The operable control includes at least one of a button control, a scroll bar control, an icon control, and a menu control.

[0216] The fingerprint sensor 1414 is used to collect a fingerprint of a user, and the identity of the user can be recognized by the processor 1401 according to the fingerprint collected by the fingerprint sensor 1414 or by the fingerprint sensor 1414 according to the collected fingerprint. When the identity of the user is recognized as a trusted identity, the processor 1401 authorizes the user to perform a related sensitive operation, which includes unlocking a screen, viewing encrypted information, downloading software, payment, and changing a setting, etc. The fingerprint sensor 1414 can be disposed on the front, back, or side of the terminal 1400. When a physical button or a manufacturer's logo is disposed on the terminal 1400, the fingerprint sensor 1414 can be integrated with the physical button or the manufacturer's logo.

[0217] The optical sensor 1415 is used to collect ambient light intensity. In an embodiment, the processor 1401 can control the display brightness of the display screen 1405 according to the ambient light intensity collected by the optical sensor 1415. Specifically, when the ambient light intensity is high, the display brightness of the display screen 1405 is increased, and when the ambient light intensity is low, the display brightness of the display screen 1405 is decreased. In another embodiment, the processor 1401 can also dynamically adjust the shooting parameters of the camera assembly 1406 according to the ambient light intensity collected by the optical sensor 1415.

[0218] The proximity sensor 1416, also known as a distance sensor, is typically located on the front panel of the terminal 1400. The proximity sensor 1416 is used to detect the distance between the user and the front of the terminal 1400. In one embodiment, when the proximity sensor 1416 detects that the distance between the user and the front of the terminal 1400 is gradually decreasing, the processor 1401 controls the display screen 1405 to switch from a screen-on state to a screen-off state; when the proximity sensor 1416 detects that the distance between the user and the front of the terminal 1400 is gradually increasing, the processor 1401 controls the display screen 1405 to switch from a screen-off state to a screen-on state.

[0219] Those skilled in the art will understand that Figure 14 The structure shown does not constitute a limitation on terminal 1400 and may include more or fewer components than shown, or combine certain components, or use different component arrangements.

[0220] When a computer device is configured as a server, Figure 15 This is a schematic diagram of a server structure according to an embodiment of this application. The server 1500 can vary significantly due to different configurations or performance. It may include one or more Central Processing Units (CPUs) 1501 and one or more memories 1502. The memory 1502 stores at least one computer program, which is loaded and executed by the processor 1501 to implement the training method of the cognitive diagnostic model provided in the above-described method embodiments. Of course, the server may also have wired or wireless network interfaces, a keyboard, and input / output interfaces for input and output. The server may also include other components for implementing device functions, which will not be elaborated here.

[0221] This application also provides a computer-readable storage medium storing at least one computer program. This computer program is loaded and executed by a processor of a computer device to implement the operations performed by the computer device in the training method and question recommendation method of the cognitive diagnostic model described above. For example, the computer-readable storage medium may be a read-only memory (ROM), a random access memory (RAM), a compact disc read-only memory (CD-ROM), magnetic tape, floppy disk, or optical data storage device, etc.

[0222] In some embodiments, the computer program related to the embodiments of the present application can be deployed on a computer device to execute, or on multiple computer devices located in one place to execute, or on multiple computer devices distributed in multiple places and interconnected through a communication network to execute. The multiple computer devices distributed in multiple places and interconnected through a communication network can constitute a blockchain system.

[0223] The embodiments of the present application further provide a computer program product, which comprises computer program code stored in a computer readable storage medium. A processor of a computer device reads the computer program code from the computer readable storage medium. The processor executes the computer program code, so that the computer device executes the training method of the cognitive diagnosis model provided in various optional implementation manners described above, or so that the computer device executes the question recommendation method provided in various optional implementation manners described above.

[0224] Those of ordinary skill in the art can understand that all or part of the steps of the above-mentioned embodiments can be completed by hardware, or by a program instructing relevant hardware, and the program can be stored in a computer readable storage medium. The storage medium mentioned above can be a read-only memory, a magnetic disk or an optical disk, etc.

[0225] The above only describes optional embodiments of the present application and is not intended to limit the present application. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application shall be included in the protection scope of the present application.

Claims

1. A training method for a cognitive diagnostic model, characterized in that, The method comprises: based on the cognitive diagnosis model of the i th round of iteration, embedding processing is performed on the first object, a plurality of nodes in the knowledge graph and the first question respectively to obtain object features of the first object, a plurality of node features of the plurality of nodes and question features of the first question, the object features are obtained by multiplying the object vector of the first object and the object embedding matrix, the object embedding matrix includes the mastery of knowledge of the first object in at least one second question, the nodes in the knowledge graph represent knowledge, the edges in the knowledge graph represent the dependency relationship between knowledge, the node features include the features of the corresponding node and the features of the nodes connected to the corresponding node, and i is a positive integer; for any node, based on the cognitive diagnosis model of the i th round of iteration, the object features, the node features of the node and the features of the predecessor node of the node are spliced to obtain the splicing features of the node, and the knowledge represented by the predecessor node is the predecessor knowledge of the knowledge represented by the node; based on the cognitive diagnosis model of the i th round of iteration, full connection is performed on the splicing features of the plurality of nodes to obtain knowledge state features, and the knowledge state features are used to represent the mastery degree of the first object to the knowledge represented by the plurality of nodes; taking the answer result of the first object to the first question as supervision information, based on the question features and the knowledge state features, the cognitive diagnosis model of the i th round of iteration is trained, and the cognitive diagnosis model is used for prediction based on the input object and question to obtain the probability that the object correctly answers the question.

2. The method of claim 1, wherein, The process of embedding processing on the plurality of nodes in the knowledge graph to obtain a plurality of node features comprises: for any node in the knowledge graph, at least one neighbor node connected to the node is obtained; based on the node weight of the at least one neighbor node, the features of the at least one neighbor node and the features of the node are weighted and summed to obtain the node features, and the node weight is used to represent the importance of the corresponding neighbor node to the node.

3. The method of claim 1, wherein, The question features of the first question include discrimination features, question knowledge features and difficulty features, and the question knowledge features are used to represent the knowledge contained in the first question; The process of embedding processing on the first question to obtain the question features of the first question comprises: based on the question identification of the first question, the question vector of the first question is determined; based on the question identification, the discrimination embedding matrix, the question knowledge embedding matrix and the difficulty embedding matrix of the first question are obtained; based on the question vector, the discrimination embedding matrix, the question knowledge embedding matrix and the difficulty embedding matrix, the discrimination features, the question knowledge features and the difficulty features are determined.

4. The method of claim 3, wherein, The determination of the discrimination features, the question knowledge features and the difficulty features based on the question vector, the discrimination embedding matrix, the question knowledge embedding matrix and the difficulty embedding matrix comprises: multiplying the question vector and the discrimination embedding matrix to obtain the discrimination feature; multiplying the question vector and the question knowledge embedding matrix to obtain the question knowledge feature; multiplying the question vector and the difficulty embedding matrix to obtain the difficulty feature.

5. The method of claim 3, wherein, The training of the cognitive diagnosis model of the i th iteration based on the question feature and the knowledge state feature includes: differencing the knowledge state feature and the difficulty feature to obtain a feature difference value; multiplying the question knowledge feature and the feature difference value bit by bit to obtain a Hadamard product; multiplying the Hadamard product and the discrimination feature to obtain an object answering feature; based on the cognitive diagnosis model of the i th iteration, predicting the object answering feature to obtain prediction result information, the prediction result information being used to indicate the probability of the first object correctly answering the first question; based on the difference between the prediction result information and the answer result of the first object to the first question, adjusting the model parameters of the cognitive diagnosis model of the i th iteration.

6. The method of claim 1, wherein, The method further includes: based on the cognitive diagnosis model of the i th iteration, obtaining an edge matrix of the knowledge graph, elements in the edge matrix being used to represent edges between the plurality of nodes; based on the cognitive diagnosis model of the i th iteration, predicting the edge matrix to obtain edge probability information, the edge probability information being used to represent the probability of edges existing between any two nodes in the plurality of nodes; based on the edge probability information, adding edges in the knowledge graph.

7. The method of claim 6, wherein, The method further includes: based on the difference between the edge probability information and the existing edges in the knowledge graph, adjusting the model parameters of the cognitive diagnosis model of the i th iteration.

8. A subject recommendation method characterized by comprising: The method includes: based on the target cognitive diagnosis model, performing cognitive diagnosis on a second object to obtain a target knowledge state feature of the second object, the target cognitive diagnosis model being trained based on the training method of the cognitive diagnosis model according to any one of claims 1-7; based on the target knowledge state feature, determining at least one target question to be recommended; recommending the at least one target question to the second object.

9. A training device for a cognitive diagnostic model, characterized in that, The device includes: an embedding module configured to perform embedding processing on a first object, a plurality of nodes in a knowledge graph, and a first question based on a cognitive diagnosis model of the i th iteration to obtain an object feature of the first object, a plurality of node features of the plurality of nodes, and a question feature of the first question, the object feature being obtained by multiplying an object vector of the first object and an object embedding matrix, the object embedding matrix including a mastery of knowledge of the first object on at least one second question, the nodes in the knowledge graph representing knowledge, the edges in the knowledge graph representing dependency relationships between knowledge, and the node features including features of corresponding nodes and features of nodes connected to the corresponding nodes. a fusion module configured to, for any node, concatenate the object feature, a node feature of the node, and a feature of a predecessor node of the node based on the cognitive diagnosis model of the i-th iteration, the predecessor node representing a predecessor knowledge of a knowledge represented by the node, and perform full connection on the concatenated features of the plurality of nodes based on the cognitive diagnosis model of the i-th iteration to obtain a knowledge state feature, the knowledge state feature being used to represent a mastery level of the first object on the knowledge represented by the plurality of nodes; a training module configured to, based on the answer result of the first object on the first question as supervision information, train the cognitive diagnosis model of the i-th iteration based on the question feature and the knowledge state feature, the cognitive diagnosis model being used to make a prediction based on an input object and question to obtain a probability of the object correctly answering the question.

10. A title recommendation device characterized by comprising: The device comprises: an object processing module configured to perform cognitive diagnosis on a second object based on a target cognitive diagnosis model to obtain a target knowledge state feature of the second object, the target cognitive diagnosis model being trained based on the training method of the cognitive diagnosis model according to any one of claims 1-7; a determination module configured to determine at least one target question to be recommended based on the target knowledge state feature; a recommendation module configured to recommend the at least one target question to the second object.

11. A computer device, comprising: The computer device comprises a processor and a memory, the memory being used to store at least one piece of computer program, the at least one piece of computer program being loaded and executed by the processor to implement the training method of the cognitive diagnosis model according to any one of claims 1-7, or the at least one piece of computer program being loaded and executed by the processor to implement the question recommendation method according to claim 8.

12. A computer-readable storage medium, characterized in that, The computer readable storage medium is used to store at least one piece of computer program, the at least one piece of computer program being used to implement the training method of the cognitive diagnosis model according to any one of claims 1-7, or the at least one piece of computer program being used to implement the question recommendation method according to claim 8.

13. A computer program product comprising a computer program, characterized in that, The computer program is executed by the processor to implement the training method of the cognitive diagnosis model according to any one of claims 1-7, or the computer program is executed by the processor to implement the question recommendation method according to claim 8.

Images