Method for predicting future learning performance of user and related equipment

Abstract

The invention provides a prediction method for future learning performance of a user and related equipment. The prediction of the future learning performance of the user can be more accurate. The method comprises the following steps: determining a historical learning data matrix corresponding to a target user; determining a direct relation tensor and an indirect relation tensor corresponding to the historical learning data matrix based on a tensor self-attention mechanism; determining a fusion matrix according to the direct relation tensor and the indirect relation tensor; determining the attention weight of each feature dimension in a plurality of feature dimensions according to the fusion matrix; determining a multi-dimensional tensor attention matrix according to the attention weight of each feature dimension; and predicting the future learning performance of the target user based on the multi-dimensional tensor attention matrix.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of knowledge tracing, and in particular to a method for predicting future learning performance of a user and related equipment. BACKGROUND

[0002] Knowledge Tracing (KT) is a comprehensive technology involving both educational technology and computer technology, aiming to model the transformation process of learners' mastery of knowledge concepts in the process of knowledge transfer through their interaction records in historical learning data, and to predict their future learning performance on this basis. This technology can help educators and learners better understand learning and provide targeted support and guidance.

[0003] Knowledge concepts play a crucial role in the process of knowledge tracing. Knowledge concepts are an important basis for teaching effectiveness evaluation, personalized learning recommendation, and teaching strategy formulation, because the effectiveness of teaching often depends on the learners' mastery of knowledge concepts, and different learners may have different mastery and interest in different knowledge concepts. How to accurately capture students' learning state and predict their future learning performance based on the relationship between knowledge concepts is a problem that needs to be solved urgently. SUMMARY

[0004] The embodiments of the present application provide a method for predicting future learning performance of a user and related equipment, which can predict the future learning performance of a target user more accurately.

[0005] The first aspect of the present application provides a method for predicting future learning performance of a user, comprising: determining a historical learning data matrix corresponding to a target user; determining a direct relationship tensor and an indirect relationship tensor corresponding to the historical learning data matrix based on a tensorized self-attention mechanism; determining a fusion matrix according to the direct relationship tensor and the indirect relationship tensor; determining an attention weight of each feature dimension in a plurality of feature dimensions according to the fusion matrix; determining a multi-dimensional tensorized attention matrix according to the attention weight of each feature dimension; predicting the future learning performance of the target user based on the multi-dimensional tensorized attention matrix.

[0006] The second aspect of the present application provides a device for predicting future learning performance of a user, comprising: determining a historical learning data matrix corresponding to a target user; determine a direct relationship tensor and an indirect relationship tensor corresponding to the historical learning data matrix based on a tensorized self-attention mechanism; determine a fusion matrix according to the direct relationship tensor and the indirect relationship tensor; determine an attention weight of each feature dimension in a plurality of feature dimensions according to the fusion matrix; determine a multi-dimensional tensorized attention matrix according to the attention weight of each feature dimension; predict the future learning performance of the target user based on the multi-dimensional tensorized attention matrix.

[0007] The second aspect of the present application provides a device for predicting future learning performance of a user, comprising: a first determining module configured to determine a historical learning data matrix corresponding to a target user; a second determining module configured to determine a direct relationship tensor and an indirect relationship tensor corresponding to the historical learning data matrix based on a tensorized self-attention mechanism; a third determining module configured to determine a fusion matrix according to the direct relationship tensor and the indirect relationship tensor; a fourth determining module configured to determine an attention weight of each feature dimension in a plurality of feature dimensions according to the fusion matrix; a fifth determining module configured to determine a multi-dimensional tensorized attention matrix according to the attention weight of each feature dimension; a predicting module configured to predict the future learning performance of the target user based on the multi-dimensional tensorized attention matrix.

[0008] In a possible design, the first determining module is specifically configured to: determine a historical learning data sequence corresponding to the target user, the historical learning data sequence being represented in the form of a binary tuple; project the historical learning data sequence to an embedding space to convert each binary tuple in the historical learning data sequence into a vector; determine the historical learning data matrix according to the conversion of each binary tuple in the historical learning data sequence into a vector.

[0009] In a possible design, the second determining module is specifically configured to: process the historical learning data matrix by different types of projection matrices to determine a query and a key-value pair in each subspace in a plurality of subspaces; extend the query and the key-value pair to different feature dimensions in the each subspace by an extension matrix; The direct relationship matrix between knowledge concepts corresponding to the history learning matrix in each of the plurality of feature dimensions is calculated based on a scaled dot-product self-attention mechanism; The direct relationship matrix corresponding to a target feature dimension is subjected to a tensorization operation to obtain the direct relationship tensor, the target feature dimension being any one of the plurality of feature dimensions; An indirect relationship matrix between knowledge concepts corresponding to the history learning matrix is calculated according to the query and the key-value pair and the dimension expansion matrix; The indirect relationship matrix is copied and subjected to a tensorization operation to obtain the indirect relationship tensor, the size of the direct relationship tensor and the size of the indirect relationship tensor having a correlation relationship.

[0010] In a possible design, the third determining module is specifically configured to: The first slice is extracted from the direct relationship tensor and the second slice is extracted from the indirect relationship tensor in sequence with the feature dimension as an index; The first slice and the second slice are subjected to additive fusion to obtain the fusion matrix.

[0011] In a possible design, the third determining module is specifically configured to: The first slice and the second slice are determined by the following formula: wherein, is the first slice, f is the feature dimension, is the second slice; The fusion matrix is determined by the following formula: ; wherein, is the fusion matrix.

[0012] In a possible design, the fourth determining module is specifically configured to: The attention weight of each of the plurality of feature dimensions is determined by the following formula: wherein, is the attention weight of each of the plurality of feature dimensions, is the fusion matrix, is a position mask matrix corresponding to the history learning data matrix, is a key-value pair.

[0013] In a possible design, the fifth determining module is specifically configured to: The multi-dimensional tensorized attention matrix is determined by the following formula: ; Wherein, S is the multi-dimensional tensorized attention matrix, is the attention weight of the first feature dimension in the plurality of feature dimensions, .

[0014] In a possible design, the prediction module is specifically configured to: The future learning performance vector of the target user is determined by the following formula: Wherein, is the future learning performance vector, The length of is equal to the total number of knowledge concepts corresponding to the historical learning data matrix, , , S is the multi-dimensional tensorized attention matrix; The future learning performance of the target user is predicted according to the future learning performance vector.

[0015] The third aspect of the present application provides an electronic device, comprising a memory and a processor, wherein the processor is configured to execute the computer management program stored in the memory to realize the steps of the prediction method of the future learning performance of the user according to the first aspect.

[0016] The fourth aspect of the present application provides a computer readable storage medium, which stores a computer management program, wherein the computer management program is executed by a processor to realize the steps of the prediction method of the future learning performance of the user according to the first aspect.

[0017] As can be seen from the above, in the embodiments of the present application, the tensorization self-attention mechanism is used to simultaneously focus on the direct relationship and indirect relationship between knowledge concepts, and the influence of the knowledge concepts on the knowledge state of the target user in the learning process of the target user is better analyzed based on this. The natural advantages of multi-dimensional data are shown by the tensor, the correlation between the attention weights on different feature dimensions in the subspace is completely preserved by tensorizing the attention weights, and the influence degree between knowledge concepts is better analyzed based on the tensorized attention weights, so that the prediction result is more accurate when predicting the future learning performance of the target user. As can be seen from the above, in the embodiments of the present application, the tensorization self-attention mechanism is used to simultaneously focus on the direct relationship and indirect relationship between knowledge concepts, and the influence of the knowledge concepts on the knowledge state of the target user in the learning process of the target user is better analyzed based on this. The natural advantages of multi-dimensional data are shown by the tensor, the correlation between the attention weights on different feature dimensions in the subspace is completely preserved by tensorizing the attention weights, and the influence degree between knowledge concepts is better analyzed based on the tensorized attention weights, so that the prediction result is more accurate when predicting the future learning performance of the target user. As can be seen from the above, in the embodiments of the present application, the tensorization self-attention mechanism is used to simultaneously focus on the direct relationship and indirect relationship between knowledge concepts, and the influence of the knowledge concepts on the knowledge state of the target user in the learning process of the target user is better analyzed based on this. The natural advantages of multi-dimensional data are shown by the tensor, the correlation between the attention weights on different feature dimensions in the subspace is completely preserved by tensorizing the attention weights, and the influence degree between knowledge concepts is better analyzed based on the tensorized attention weights, so that the prediction result is more accurate when predicting the future learning performance of the target user. BRIEF DESCRIPTION OF DRAWINGS

[0018] Figure 1 A schematic diagram of different relationships between knowledge concepts provided by the present application; Figure 2A flowchart illustrating a method for predicting a user's future learning performance provided in an embodiment of the present invention; Figure 3 Another flowchart illustrating the method for predicting a user's future learning performance provided in an embodiment of the present invention; Figure 4 A schematic diagram of the virtual structure of a device for predicting a user's future learning performance provided in an embodiment of the present invention; Figure 5 A schematic diagram of the hardware structure of a device for predicting a user's future learning performance provided in an embodiment of the present invention; Figure 6 A schematic diagram of an embodiment of the electronic device provided in this invention; Figure 7 A schematic diagram illustrating an embodiment of a computer-readable storage medium provided in this invention. Detailed Implementation

[0019] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, those skilled in the art can make their own judgments without creative effort. All other embodiments obtained are within the scope of protection of this invention.

[0020] In the following description, specific embodiments of the invention will be described with reference to steps and symbols performed by one or more computers, unless otherwise stated. Therefore, these steps and operations will be referred to several times as being performed by a computer, and computer execution as referred to herein includes the execution by electronic signals representing data in a structured format. The operation of the computer processing unit. This operation transforms the data or maintains it at a location in the computer's memory system, which can be reconfigured or otherwise alter the operation of the computer in a manner well known to those skilled in the art. The data structure maintained by the data is the physical location of the memory, which has the data format... The specific characteristics defined herein. However, the principles of the present invention have been described above, and this is not intended to be a limitation. Those skilled in the art will understand that many of the steps and operations described below can also be implemented in hardware.

[0021] The principles of this invention are applied using many other general-purpose or purpose-specific computing, communication environments, or configurations. Examples of well-known computing systems, environments, and configurations suitable for use with this invention include (but are not limited to) handheld phones, personal computers, servers, multiprocessor systems, microcomputer-based systems, and mainframe computers. Computers and distributed computing environments, including any of the aforementioned systems or devices.

[0022] The terms "first", "second", and "third" and the like in the present invention are used to distinguish different objects, rather than to describe a particular order. In addition, the terms "include" and "have" and any variations thereof are intended to cover non-exclusive inclusion.

[0023] The relationship between knowledge concepts mainly includes direct relationship and indirect relationship, which play different roles in cognitive structure and knowledge system. The direct relationship refers to the mutual dependence and mutual restriction between knowledge concepts that are closer in the learning process. The indirect relationship refers to the mutual dependence and mutual restriction between knowledge concepts through more intermediaries and intermediate links. Most of the current knowledge tracking technologies focus on the direct relationship between knowledge concepts and conduct research on this basis. For example, BKT (Bayesian Knowledge Tracing) tracks the change of a learner's knowledge concept and can know the change of the learner's knowledge concept mastery. This method establishes an HMM model for each knowledge concept to predict the learner's mastery of a specific knowledge concept, but does not consider the correlation between knowledge concepts. DKT (Deep Knowledge Tracing) is a knowledge tracking method based on deep learning, which uses a single hidden vector to summarize the knowledge state, making it difficult to track the learner's mastery of a specific concept. The DKVMN (Dynamic Key-Value Memory Networks) method draws on the idea of MANN (memory-augmented neural networks) and uses a static key matrix and a dynamic value matrix to automatically discover the underlying concepts of practice, but has limitations in capturing the evolution of the learner's knowledge state. SAKT (Self-Attentive Knowledge Tracing) is a knowledge tracking method based on self-attention mechanism, which uses the encoder part of the Transformer to do sequence tasks, aiming to more accurately simulate and predict the learner's mastery of knowledge concepts in a series of learning activities, but only performs better when dealing with a small amount of learning data of learners. AKT (Attentive Knowledge Tracing) is a knowledge tracking method that combines attention mechanism, cognitive science theory and psychometric model, which uses self-attention mechanism to encode problem embedding and output the context representation of the problem, but also ignores a more comprehensive analysis of the relationship between knowledge concepts. ​​​​​​

[0024] Please see Figure 1 , Figure 1 This is a schematic diagram illustrating the different relationships between knowledge concepts provided by the present invention. Although indirect relationships are not as direct as direct relationships, they still play an important role in the knowledge system, helping to build a complete knowledge framework and deepen the understanding of knowledge.

[0025] This invention proposes a method and related equipment for predicting a user's future learning performance, specifically as follows: In analyzing the direct relationships between various knowledge concepts, we first define the projection matrix of each subspace: , These weight matrices are parameters learned during model training. Then, the input matrix X, containing the learner's historical learning data, knowledge concepts, and answer responses, is linearly... Mapping to different subspaces allows us to obtain queries and key-value pairs for each subspace: The specific calculation formula is as follows: , , in, This represents the h-th subspace, where H represents the total number of subspaces. And through the extended matrix , , naturally and Extending to different feature dimensions in each subspace, the specific calculation formula is as follows: Among them, the extended matrices of different feature dimensions are also learnable. Let f be the f-th feature dimension, and F be the total number of feature dimensions in the subspace; Finally, the direct relationships between knowledge concepts in each feature dimension are calculated based on the scaled dot product self-attention mechanism. The specific calculation formula is as follows: = f ( , )= ( ); in, A matrix representing the direct relationships between knowledge concepts. This represents an activation function. To better preserve the relationships between direct relations existing in different subspaces, the direct relation matrix of each feature dimension is treated as a slice, and tensor quantization is performed: ; Form a direct relation tensor between knowledge concepts .

[0026] In analyzing the indirect relationships between knowledge concepts, based on the self-attention mechanism, the importance of each element in the input sequence to the entire input sequence is calculated, and an improvement using multidimensional tensor quantization is introduced into the analysis of indirect relationships between knowledge concepts. The specific calculation formula is as follows: in, A matrix representing the indirect relationships between knowledge concepts. This represents an activation function. , , This represents the dimension expansion matrix, where the column dimensions represent the number of feature dimensions, and... The matrix is ​​copied row by row n times and tensorized to obtain the tensor with direct relation. Indirect relation tensors of uniform size Based on the direct and indirect relation tensors of knowledge concepts, a complete and coherent knowledge framework can be constructed by simultaneously focusing on these two types of relations during the knowledge tracing process, resulting in accurate and comprehensive knowledge tracking. It can efficiently capture learners' knowledge status and predict their future learning performance.

[0027] The following description focuses on the prediction method for user future learning performance provided by this invention from the perspective of a device for predicting user future learning performance. This device can be a server or a service unit within a server, and is not specifically limited.

[0028] See also Figure 2 and Figure 3 Please see Figure 2 , Figure 2 This is a flowchart illustrating a method for predicting a user's future learning performance provided in an embodiment of the present invention. Figure 3 This is another flowchart illustrating the method for predicting a user's future learning performance provided by the present invention. The method includes: 301. Determine the historical learning data matrix corresponding to the target user.

[0029] In this embodiment, the device for predicting a user's future learning performance can acquire high-dimensional, discrete historical learning data corresponding to the target user and convert this historical learning data into low-dimensional continuous embedding vectors. The trained continuous embedding vectors can capture the features and relationships of the input data. This conversion not only reduces... The input dimension improves the calculation efficiency and can more effectively process and learn the data. For example, in the problem-solving records of the target user, the embedding layer can learn the correlations between different knowledge points and the difficulty of the questions and other characteristics, and these characteristic representations are passed to the subsequent neural network layers for further processing and learning. Specifically, the user future learning performance prediction apparatus can determine a historical learning data sequence corresponding to the target user, the historical learning data sequence being represented in the form of a binary tuple; and project the historical learning data sequence into an embedding space to convert each binary tuple in the historical learning data sequence into a vector. Then, a historical learning data matrix is determined according to the conversion of each binary tuple in the historical learning data sequence into a vector.

[0030] That is, the user future learning performance prediction apparatus can present the historical learning data corresponding to the target user in the form of a sequence (i.e., a historical learning data sequence), and each element in the sequence is represented in the form of a binary tuple. The historical learning data sequence is projected into an embedding space by a one-hot encoding algorithm to convert each binary tuple in the historical learning data sequence into a vector. wherein, represents the serial number of each historical learning data, T represents the total amount of historical learning data, represents the knowledge concept index corresponding to the tth question, represents the total number of all knowledge concepts in the historical learning data corresponding to the target user, represents the answer condition of the tth question, and thus a historical learning data matrix of the target user in the form of a matrix is obtained. wherein, n is a pre-set maximum sequence length.As the output of the embedding layer, the features and relationships of the historical learning data of the target user are included. 302. Determine the direct relationship tensor and the indirect relationship tensor corresponding to the historical learning data matrix based on the tensorized self-attention mechanism.

[0031] In this embodiment, the user future learning performance prediction apparatus processes the historical learning data matrix by different types of projection matrices to determine the query and key-value pairs in each subspace in the plurality of subspaces; that is, the user future learning performance prediction apparatus determines the query and key-value pairs in each subspace in the plurality of subspaces by projecting the output result of the embedding layer into the plurality of subspaces through the tensorized self-attention layer.

[0032] ​​​​(i.e., historical learning data matrix), using different types of projection matrices: , obtaining the query and key-value pairs in each subspace: , represents the hth subspace, H represents the total number of subspaces, and d represents the latent feature dimension in the subspace; After that, by extending the matrix , the and are naturally extended to different feature dimensions in the subspace; And based on the scaled dot-product self-attention mechanism, the direct relationship matrix between knowledge concepts in each feature dimension is calculated: , where represents the fth feature dimension, and F represents the total number of feature dimensions in the subspace; At the same time, in order to better preserve the association between direct relationships existing in different subspaces, the direct relationship matrix of each feature dimension is regarded as a slice, and the slice is tensorized to form a direct relationship tensor between knowledge concepts: .

[0033] According to the query and key-value pairs and the dimension expansion matrix, the indirect relationship matrix between knowledge concepts corresponding to the historical learning matrix is calculated, and the indirect relationship matrix is copied and tensorized to obtain the indirect relationship tensor, i.e., by in the query and key-value pairs and the dimension expansion matrix: , the indirect relationship matrix between knowledge concepts is calculated, and then the matrix is copied n times by row to obtain the indirect relationship tensor with the same size as the direct relationship tensor .

[0034] 303. Determine the fusion matrix according to the direct relationship tensor and the indirect relationship tensor.

[0035] In this embodiment, the prediction device of user future learning performance extracts a first slice from the direct relationship tensor and a second slice from the indirect relationship tensor in turn by indexing the feature dimension f through the tensorized self-attention layer, and the specific formula is as follows: ; After that, the first slice and the second slice are additively fused to obtain the fusion matrix, and the specific formula is as follows: ; wherein is the fusion matrix.

[0036] 304. Determine the attention weight of each feature dimension in the plurality of feature dimensions according to the fusion matrix.

[0037] In this embodiment, in order to detect and extract the structural features similar to the fusion matrix and the mask, so as to identify and analyze the specific knowledge concept in the historical learning data of the target user, the user future learning performance prediction device sets the elements in the tensor position mask matrix as follows: According to the following rules: When x is greater than y, 0, and negative infinity in other cases.

[0038] Then, the Softmax function is used to calculate , , The attention weight of each feature dimension in the plurality of feature dimensions is obtained. Specifically, the user future learning performance prediction device can determine the attention weight of each feature dimension in the plurality of feature dimensions according to the following formula: It can be understood that the knowledge concept attention weight matrix of the current subspace obtained after global average pooling is: as the output result of the tensor self-attention layer.

[0039] 305. Determine the multi-dimensional tensor attention matrix according to the attention weight of each feature dimension.

[0040] In this embodiment, after determining the attention weight of each feature dimension, the user future learning performance prediction device can determine the multi-dimensional tensor attention matrix according to the attention weight of each feature dimension, that is, the user future learning performance prediction device realizes parallelization in calculation through the multi-head parallel processing mode, from accelerate the training and inference process of the model, which is particularly important for processing the historical learning data corresponding to the target user. Specifically, the user future learning performance prediction device connects the multi-dimensional tensor self-attention matrix in each subspace through a multi-head connection layer to obtain a multi-dimensional tensor attention matrix , the formula is as follows: Among them, , as the output result of the multi-head connection layer, contains the knowledge concept weight information in each subspace.

[0041] ​​306、predicting the future learning performance of the target user based on the multi-dimensional tensorized attention matrix.

[0042] In this embodiment, the prediction device of the future learning performance of the user can first determine the future learning performance vector of the target user by the following formula: wherein, is the future learning performance vector, the length of is equal to the total number of knowledge concepts corresponding to the historical learning data matrix, , S is the multi-dimensional tensorized attention matrix; and predict the future learning performance of the target user according to the future learning performance vector. For the convenience of understanding, how to predict the future learning performance of the target user is described in detail as follows: This step mainly includes two parts: a feedforward neural network part and a prediction fully connected network part. First, the prediction device of the future learning performance of the user utilizes the excellent processing capability of the feedforward neural network in complex nonlinear mapping to capture the interaction between knowledge concepts in different feature dimensions. The feedforward neural network used in the feedforward neural network part is as follows: wherein, These parameter matrices can be obtained by model training, and then the prediction device of the future learning performance of the user inputs each row in the output matrix of the feedforward neural network to a fully connected network with a Sigmoid activation function to predict the future learning performance of the target user. The specific performance formula is as follows: wherein, is a vector with a length equal to the total number of knowledge concepts (i.e., the future learning performance vector), and each element in the future learning performance vector is in the range of 0 to 1, representing the mastery degree of the corresponding knowledge concept. The closer the value is to 0, the lower the mastery degree, and vice versa, the closer the value is to 1, the higher the mastery degree. At the same time, it also represents the probability of the learner being able to correctly answer when facing exercises containing the corresponding knowledge concept, e represents the total number of knowledge concepts, and plays a role in limiting shape.

[0043] It should be noted that the prediction device of the future learning performance of the user minimizes and by setting the following loss function: The cross entropy loss between the two is calculated, and the learnable parameters are trained to minimize the negative log-likelihood of the target user's response to the question, thereby improving the prediction effect.

[0044] In summary, it can be seen that, in the embodiments provided by the present application, the tensorized self-attention mechanism is used to simultaneously focus on the direct relationship and indirect relationship between knowledge concepts, and the influence of the knowledge concepts on the knowledge state of the target user in the learning process is better analyzed based on this. The natural advantages of multi-dimensional data are used to save the correlation between the attention weights in different feature dimensions in the subspace through tensorization of the attention weights, and the influence degree between the knowledge concepts is better analyzed based on the tensorized attention weights, so that the prediction result is more accurate when predicting the future learning performance of the target user.

[0045] The above describes the embodiments of the present application from the perspective of predicting the future learning performance of the user, and the following describes the embodiments of the present application from the perspective of a device for predicting the future learning performance of the user.

[0046] Referring to Figure 4 , the virtual structure schematic diagram of the device for predicting the future learning performance of the user in the embodiments of the present application, the device for predicting the future learning performance of the user 400 includes: A first determination module 401 is configured to determine a historical learning data matrix corresponding to a target user. A second determination module 402 is configured to determine a direct relationship tensor and an indirect relationship tensor corresponding to the historical learning data matrix based on a tensorized self-attention mechanism. A third determination module 403 is configured to determine a fusion matrix according to the direct relationship tensor and the indirect relationship tensor. A fourth determination module 404 is configured to determine an attention weight of each feature dimension in a plurality of feature dimensions according to the fusion matrix. A fifth determination module 405 is configured to determine a multi-dimensional tensorized attention matrix according to the attention weight of each feature dimension. A prediction module 406 is configured to predict the future learning performance of the target user based on the multi-dimensional tensorized attention matrix.

[0047] In one possible design, the first determination module 401 is specifically configured to: determine a historical learning data sequence corresponding to the target user, the historical learning data sequence being represented in the form of a binary tuple; project the historical learning data sequence to an embedding space to convert each binary tuple in the historical learning data sequence into a vector.​​ determine the historical learning data matrix according to conversion of each binary tuple in the historical learning data sequence into a vector.

[0048] In a possible design, the second determining module 402 is specifically configured to: process the historical learning data matrix through different types of projection matrices to determine a query and a key-value pair of each subspace in a plurality of subspaces; extend the query and the key-value pair to different feature dimensions in the each subspace through an extension matrix; calculate a direct relationship matrix between knowledge concepts corresponding to the historical learning matrix in each feature dimension of the plurality of feature dimensions based on a scaled dot-product self-attention mechanism; perform a tensorization operation on a direct relationship matrix corresponding to a target feature dimension to obtain a direct relationship tensor, the target feature dimension being any one of the plurality of feature dimensions; calculate an indirect relationship matrix between knowledge concepts corresponding to the historical learning matrix according to the query and the key-value pair and a dimension extension matrix; copy the indirect relationship matrix and perform a tensorization operation to obtain an indirect relationship tensor, the direct relationship tensor and the indirect relationship tensor having a correlation relationship in size.

[0049] In a possible design, the third determining module 403 is specifically configured to: extract a first slice from the direct relationship tensor and a second slice from the indirect relationship tensor in sequence with a feature dimension as an index; perform additive fusion on the first slice and the second slice to obtain the fusion matrix.

[0050] In a possible design, the third determining module 403 is specifically configured to: determine the first slice and the second slice through the following formula: wherein, is the first slice, f is the feature dimension, is the second slice; determine the fusion matrix through the following formula: ; wherein, is the fusion matrix.

[0051] In a possible design, the fourth determining module 404 is specifically configured to: The attention weight of each feature dimension in the plurality of feature dimensions is determined by the following formula: wherein, is the attention weight of each feature dimension, is the fusion matrix, is a position mask matrix corresponding to the historical learning data matrix, is a key-value pair.

[0052] In a possible design, the fifth determination module 405 is specifically configured to: The multi-dimensional tensorized attention matrix is determined by the following formula: wherein, S is the multi-dimensional tensorized attention matrix, is the attention weight of the first feature dimension in the plurality of feature dimensions,

[0053] In a possible design, the prediction module 406 is specifically configured to: The future learning performance vector of the target user is determined by the following formula: wherein, is the future learning performance vector, the length of is equal to the total number of knowledge concepts corresponding to the historical learning data matrix, S is the multi-dimensional tensorized attention matrix; The future learning performance of the target user is predicted according to the future learning performance vector.

[0054] The above Figure 4 The prediction device for the future learning performance of a user in the embodiment of the present application is described from the perspective of a modular functional entity, and the prediction device for the future learning performance of a user in the embodiment of the present application is described in detail from the perspective of hardware processing. Please refer to Figure 5 FIG. 5, which is an embodiment schematic diagram of the prediction device 500 for the future learning performance of a user in the embodiment of the present application, and the prediction device 500 for the future learning performance of a user includes: an input device 501, an output device 502, a processor 503, and a memory 504 (wherein the number of processors 503 can be one or more, Figure 5 and the input device 501, the output device 502, the processor 503, and the memory 504 can be connected through a communication bus or other means, wherein, Figure 5 ​​​Take the communication bus connection as an example.

[0055] The processor 503 is configured to execute the following steps by invoking operation instructions stored in the memory 504: determine a historical learning data matrix corresponding to a target user; determine a direct relationship tensor and an indirect relationship tensor corresponding to the historical learning data matrix based on a tensorized self-attention mechanism; determine a fusion matrix according to the direct relationship tensor and the indirect relationship tensor; determine an attention weight of each feature dimension in a plurality of feature dimensions according to the fusion matrix; determine a multi-dimensional tensorized attention matrix according to the attention weight of each feature dimension; predict a future learning performance of the target user based on the multi-dimensional tensorized attention matrix.

[0056] The processor 503 is further configured to execute the following steps by invoking operation instructions stored in the memory 504: Figure 3 any of the corresponding embodiments.

[0057] Please refer to Figure 6 , Figure 6 an embodiment of an electronic device provided by the present embodiment.

[0058] As Figure 6 shown, the present embodiment provides an electronic device, which includes a memory 610, a processor 620, and a computer program 611 stored in the memory 610 and executable on the processor 620. When the processor 620 executes the computer program 611, the following steps are implemented: determine a historical learning data matrix corresponding to a target user; determine a direct relationship tensor and an indirect relationship tensor corresponding to the historical learning data matrix based on a tensorized self-attention mechanism; determine a fusion matrix according to the direct relationship tensor and the indirect relationship tensor; determine an attention weight of each feature dimension in a plurality of feature dimensions according to the fusion matrix; determine a multi-dimensional tensorized attention matrix according to the attention weight of each feature dimension; predict a future learning performance of the target user based on the multi-dimensional tensorized attention matrix.

[0059] In the specific implementation process, when the processor 620 executes the computer program 611, any of the following steps can be implemented: Figure 3 any of the corresponding embodiments.

[0060] Since the electronic device introduced in the embodiment is the device used by the calculation device for exciting the frequency point unit in one array antenna in the embodiment of the present application, based on the method introduced in the embodiment of the present application, a person skilled in the art can understand the specific implementation of the electronic device in the embodiment and various changes thereof Therefore, how the electronic device implements the method in the embodiment of the present application is not described in detail here, as long as the device used by a person skilled in the art to implement the method in the embodiment of the present application belongs to the scope of protection of the present application.

[0061] Please refer to Figure 7 , Figure 7 An embodiment of a computer readable storage medium provided in the embodiment of the present application is shown in the figure.

[0062] As Figure 7 shown, the embodiment of the present application further provides a computer readable storage medium 700, which stores a computer program 711, and the computer program 711 is executed by a processor to implement the following steps: determining a historical learning data matrix corresponding to a target user; determining a direct relationship tensor and an indirect relationship tensor corresponding to the historical learning data matrix based on a tensorization self-attention mechanism; determining a fusion matrix according to the direct relationship tensor and the indirect relationship tensor; determining an attention weight of each feature dimension in a plurality of feature dimensions according to the fusion matrix; determining a multi-dimensional tensorization attention matrix according to the attention weight of each feature dimension; predicting a future learning performance of the target user based on the multi-dimensional tensorization attention matrix.

[0063] In the specific implementation process, the computer program 711 is executed by the processor to implement Figure 3 any implementation in the corresponding embodiment.

[0064] It should be noted that in the above embodiments, the description of each embodiment has its own emphasis, and the parts not described in detail in a certain embodiment can be referred to the related description of other embodiments.

[0065] A person skilled in the art should understand that the embodiments of the present application can be provided as a method, a system, or a computer program product. Therefore, the present application can adopt a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Moreover, the present application can adopt a form of one or more computer programs containing computer program instructions embodied in computer readable media. A computer program product implemented on a computer-usable storage medium (including but not limited to disk storage, CD-ROM, optical storage, etc.) with machine-usable program code.

[0066] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions are provided to a processor of a general-purpose computer, a special-purpose computer, an embedded computer, or other programmable data processing device to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing device, produce instructions for implementing the process. Figure 1 One or more processes and / or box Figure 1 A device that provides the functions specified in one or more boxes.

[0067] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or methods box Figure 1 The function specified in one or more boxes.

[0068] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One process or multiple Process and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0069] This invention also provides a computer program product comprising computer software instructions that, when executed on a processing device, cause the processing device to perform actions such as... Figure 3 The process in the corresponding embodiment.

[0070] The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, all or part of the processes or functions described in the embodiments of the present invention are generated. The computer may be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device.

[0071] The computer instructions can be stored in or transferred from one computer-readable medium to another computer-readable medium, such as from one website, computer, server, or data center to another website, computer, server, or data center, through a wired (such as coaxial cable, optical fiber, digital subscriber line (DSL), etc.) or wireless (such as infrared, wireless, microwave, etc.) medium. The computer-readable medium can be any available medium or a set of media including, but not limited to, a magnetic medium, such as a floppy diskette or a hard disk drive, an optical medium, such as a DVD, or a semiconductor medium, such as a solid state disk (SSD), etc., which is integrally formed with or is included in a server, data center, etc. data storage device.

[0072] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the above-described system, device and unit can refer to the corresponding processes in the foregoing method embodiments, which will not be repeated here.

[0073] In several embodiments provided in the present application, it should be understood that the disclosed system, device and method can be implemented in other ways. For example, the above-described device embodiments are only schematic, for example, the division of the units is only a logical function division, and actual implementation can be in another division manner, for example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. In addition, the coupling or direct coupling or communication connection between the units shown or discussed can be indirect coupling or communication connection through some interface, device or unit, and can be electrical, mechanical or other forms.

[0074] The units described as separate components can or can not be physically separated, and the components shown as units can or can not be physical units, that is, they can be located in one place, or can be distributed on multiple network units. According to actual needs, some or all of the units can be selected to realize the purpose of the embodiment scheme.

[0075] ​​​​​Furthermore, the functional units in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0076] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can... The software product is manifested in the form of a software product, stored in a storage medium, and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium package... This includes various media that can store program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0077] The above description and embodiments are merely illustrative of the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or some of the technical features can be altered. Equivalent substitutions may be made; however, these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the various embodiments of the present invention.

Claims

1. A method for predicting a user's future learning performance, characterized in that, The method comprises the following steps: determining a historical learning data matrix corresponding to a target user; determining a direct relationship tensor and an indirect relationship tensor corresponding to the historical learning data matrix based on a tensorized self-attention mechanism; determining a fusion matrix according to the direct relationship tensor and the indirect relationship tensor; determining an attention weight of each feature dimension in a plurality of feature dimensions according to the fusion matrix; determining a multi-dimensional tensorized attention matrix according to the attention weight of each feature dimension; predicting a future learning performance of the target user based on the multi-dimensional tensorized attention matrix.

2. The method of claim 1, wherein, The method comprises the following steps: determining a historical learning data sequence corresponding to the target user, the historical learning data sequence being represented in the form of a binary tuple; projecting the historical learning data sequence to an embedding space to convert each binary tuple in the historical learning data sequence into a vector; determining the historical learning data matrix according to the conversion of each binary tuple in the historical learning data sequence into a vector.

3. The method of claim 1, wherein, The method comprises the following steps: processing the historical learning data matrix through different types of projection matrices to determine a query and a key-value pair of each subspace in a plurality of subspaces; extending the query and the key-value pair to different feature dimensions in the each subspace through an expansion matrix; calculating a direct relationship matrix between knowledge concepts corresponding to the historical learning matrix in each feature dimension in the plurality of feature dimensions based on a scaled dot-product self-attention mechanism; tensorizing a target feature dimension corresponding to the direct relationship matrix to obtain the direct relationship tensor, the target feature dimension being any one of the plurality of feature dimensions; calculating an indirect relationship matrix between the knowledge concepts corresponding to the historical learning matrix according to the query and the key-value pair and a dimension expansion matrix; copying and tensorizing the indirect relationship matrix to obtain the indirect relationship tensor, the direct relationship tensor and the indirect relationship tensor having a correlation in size.

4. The method of claim 1, wherein, The method comprises the following steps: indexing the feature dimensions, and sequentially extracting a first slice from the direct relationship tensor and a second slice from the indirect relationship tensor; additively fusing the first slice and the second slice to obtain the fusion matrix.

5. The method of claim 4, wherein, The method comprises the following steps: The first slice and the second slice are determined by the following formula: wherein, is the first slice, f is the feature dimension, is the second slice; The method comprises the following steps: The fusion matrix is determined by the following formula: ； wherein is the fusion matrix.

6. The method according to any one of claims 1 to 5, characterized in that, The method comprises the following steps: The attention weight of each feature dimension in the plurality of feature dimensions is determined by the following formula: wherein, is the attention weight for the each feature dimension, is the fusion matrix, is the position mask matrix corresponding to the historical learning data matrix, is the key-value pair.

7. The method according to any one of claims 1 to 5, characterized in that, The determining the multi-dimensional tensorized attention matrix according to the attention weight of each feature dimension comprises: The multi-dimensional tensorized attention matrix is determined by the following formula: ； wherein S is the multi-dimensional tensorized attention matrix, is an attention weight for a 1st feature dimension among the plurality of feature dimensions, .

8. The method according to any one of claims 1 to 5, characterized in that, The predicting the future learning performance of the target user based on the multi-dimensional tensorized attention matrix comprises: The future learning performance vector of the target user is determined by the following formula: wherein, is a future learning performance vector, has a length equal to a total number of knowledge concepts corresponding to the historical learning data matrix, , S is the multi-dimensional tensorized attention matrix; The future learning performance of the target user is predicted according to the future learning performance vector.

9. A device for predicting future learning performance of a user, characterized by, Comprise: A first determining module is configured to determine a historical learning data matrix corresponding to a target user; A second determining module is configured to determine a direct relationship tensor and an indirect relationship tensor corresponding to the historical learning data matrix based on a tensorized self-attention mechanism; A third determining module is configured to determine a fusion matrix according to the direct relationship tensor and the indirect relationship tensor; A fourth determining module is configured to determine an attention weight of each feature dimension in a plurality of feature dimensions according to the fusion matrix; A fifth determining module is configured to determine a multi-dimensional tensorized attention matrix according to the attention weight of each feature dimension; A predicting module is configured to predict the future learning performance of the target user based on the multi-dimensional tensorized attention matrix.

10. An electronic device, comprising: Comprise: A memory and a processor, wherein the processor is configured to execute a computer management program stored in the memory to implement the method for predicting the future learning performance of a user according to any one of claims 1 to 8.

Images