A quantitative evaluation method for learning behavior
Through the timing evolution model based on attention mechanism and the artificial fish school algorithm with adaptive step size, the problem of massive learning data evaluation in the field of education is solved, accurate quantitative evaluation of learning behavior and intelligent decision-making support are realized, and educational and teaching efficiency is improved.
Patent Information
- Application Number
- CN202411060035.4
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2024-08-05
- Publication Date
- 2024-11-05
- Estimated Expiration
- 2044-08-05
AI Technical Summary
The field of education faces the problem of how to effectively evaluate students' learning behavior and effects from massive learning data. Traditional methods are difficult to accurately capture the correlations and differences between knowledge points, and the optimization efficiency is low.
Using a time-series evolution model based on attention mechanism and an adaptive step size artificial fish school algorithm, through the knowledge point division and quantitative evaluation of learning behavior data, a quantitative evaluation model of learning behavior is constructed, combined with singular value decomposition and fully connected network, and optimized the model to improve the accuracy and efficiency of the assessment.
It realizes accurate quantitative evaluation of students' learning behavior, dynamically adjusts the search step size, improves model optimization efficiency and convergence speed, provides intelligent auxiliary decision-making support, and improves educational and teaching efficiency and level.
Smart Images

Figure CN118586780B_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of learning behavior assessment, and in particular to a learning behavior quantitative assessment method. Background Art
[0002] With the rapid development of information technology and the Internet, the field of education has gradually ushered in an era of informatization and digitization. Traditional paper textbooks and classroom teaching methods have gradually been replaced by digital teaching resources and online learning platforms. This has resulted in a large amount of digital data in the students' learning process, including learning time, learning content, answering status, etc. At the same time, education administrators, teachers, and parents are also paying more and more attention to students' learning status and academic performance. How to obtain useful information from these massive learning data and evaluate students' learning behavior and results has become an important issue facing the current education field. Summary of the invention
[0003] To address this problem, this patent proposes a method for quantitatively evaluating learning behavior, which uses artificial intelligence technology to analyze students' learning behavior data, provide educators with intelligent decision-making support, and further improve the efficiency and level of education and teaching.
[0004] To achieve the above object, the present invention provides a method for quantitatively evaluating learning behavior, comprising the following steps:
[0005] S1: Collecting learning behavior data, the learning behavior data includes exercise knowledge point association data and exercise interaction data; and dividing the exercise interaction data according to the knowledge points to obtain exercise interaction sub-data, wherein data segmentation based on the exercise knowledge point strong association matrix is an implementation method for dividing the exercise interaction data;
[0006] S2: Construct a learning behavior quantitative evaluation model, which takes the exercise interaction sub-data sequence as input and the learning behavior quantitative evaluation results of the corresponding knowledge points as output, wherein the time series evolution model based on the attention mechanism is the implementation method of the quantitative evaluation model;
[0007] S3: Optimizing and solving the constructed learning behavior quantitative evaluation model to obtain an optimal model instance, and using the optimal model instance obtained by the solution to perform quantitative evaluation of learning behavior on all knowledge points, and integrating the quantitative evaluation results to obtain a learning behavior quantitative evaluation vector, wherein the artificial fish swarm algorithm with adaptive step size is the implementation method of the optimization solution;
[0008] S4: Perform comprehensive quantitative evaluation on the calculated learning behavior quantitative evaluation vector to obtain the learning behavior quantitative evaluation result.
[0009] As a further improvement method of the present invention:
[0010] Optionally, in step S1, learning behavior data is collected, and exercise interaction data is divided according to knowledge points to obtain exercise interaction sub-data, including:
[0011] S11: Collect learning behavior data, which includes exercise knowledge point association data and exercise interaction data, wherein the exercise knowledge point association data records the knowledge points involved in each exercise and is represented by an exercise knowledge point association matrix A, where the rows of the matrix correspond to exercises, the columns correspond to knowledge points, and the elements Indicates whether exercise e involves knowledge point k, with a value of 1 for involved and a value of 0 for not involved; the exercise interaction data records the students' answers to the exercises and is represented by the exercise interaction matrix B. The rows of the matrix correspond to students, the columns correspond to exercises, and the elements Indicates whether the answer of student m to exercise e is correct, with a value of 1 for correct and 0 for wrong;
[0012] S12: Calculate the knowledge point coverage matrix C based on the exercise knowledge point association matrix A and the exercise interaction matrix B, where each element in the knowledge point coverage matrix C represents each student's mastery of each knowledge point. The calculation formula is:
[0013] ;
[0014] in:
[0015] Indicates the mastery of the kth knowledge point by the mth student;
[0016] Indicates the answer of the mth student to the eth exercise;
[0017] Indicates the knowledge points involved in the e-th exercise, and e represents the exercise index;
[0018] S13: Perform singular value decomposition on the calculated knowledge point coverage matrix. The calculation formula is: ,in represents the singular value matrix, represents the left singular vector matrix, represents the right singular vector matrix. The elements whose singular values in the decomposed singular value matrix are less than the specified threshold are set to 0, that is, the non-significant correlation items between the exercises and the knowledge points are eliminated, and the singular value matrix with strong correlation between the exercises and the knowledge points is obtained. ;
[0019] S14: Reconstruct the knowledge point coverage matrix according to the calculated strongly correlated singular value matrix, and the calculation formula is:
[0020] ;
[0021] in:
[0022] represents the reconstructed knowledge point coverage matrix,
[0023] The strong correlation matrix of the exercise knowledge points is calculated based on the reconstructed knowledge point coverage matrix. The calculation formula is:
[0024] ;
[0025] in:
[0026] Represents the strong correlation matrix of the knowledge points in the exercise;
[0027] represents the inverse of the exercise interaction matrix,
[0028] The elements in the strong correlation matrix of exercise knowledge points whose element values exceed the preset threshold are set to 1, indicating that there is a strong correlation between the exercise and the knowledge point, and the rest are set to 0. The exercise interaction data of students on different knowledge points are divided, and the exercises with element values of 1 in the strong correlation matrix of exercise knowledge points are used as the exercise interaction sub-data of students on the corresponding knowledge points.
[0029] Optionally, constructing a learning behavior quantitative evaluation model in step S2 includes:
[0030] The model includes: input layer, embedding layer, LSTM layer, attention layer and output layer. The specific process of quantitative evaluation of learning behavior includes:
[0031] S21: The input layer receives the student's exercise interaction sub-data sequence ,in, represents the student's exercise interaction result at time step t, and T represents the length of the exercise interaction sub-data sequence;
[0032] S22: The embedding layer converts the discrete exercise interaction results into semantic result vectors;
[0033] S23: LSTM layer semantic result vector according to time step t and the hidden state at the previous time step and cell status Output the hidden state of the current time step and cell status ;
[0034] S24: Attention layer for all hidden states of LSTM layer Perform attention weighting processing to obtain weighted context vector and attention weight;
[0035] S25: The output layer converts the context vector and hidden state The combination of is used to obtain the quantitative evaluation results of students' learning behavior through a fully connected network. The calculation formula is:
[0036] ;
[0037] in:
[0038] Represents the weight matrix parameters of the fully connected network;
[0039] Represents the bias vector parameters of the fully connected network.
[0040] Optionally, the step S22 converts the discrete exercise interaction results into a semantic result vector, including:
[0041] S221: constructing a knowledge point association matrix, wherein the elements of the knowledge point association matrix represent associations between knowledge points, a value of 1 indicates that an association exists, and a value of 0 indicates that an association does not exist;
[0042] S222: Aggregate the adjacent node information of the knowledge point to generate a compression vector of the corresponding knowledge point. The calculation formula is: ;
[0043] in:
[0044] Represents the compressed vector of knowledge point i;
[0045] Represents the row vector of knowledge point j processed according to one-hot encoding;
[0046] Represents the association relationship between knowledge point i and knowledge point j in the knowledge point association matrix;
[0047] represents the encoding weight matrix;
[0048] f represents the activation function of the encoding part;
[0049] N represents the number of adjacent knowledge points of knowledge point i;
[0050] S223: Remap the encoded compressed vector to the reconstructed knowledge point row vector. The calculation formula is: ;
[0051] in:
[0052] Represents the compressed vector of knowledge point j;
[0053] represents the decoding weight matrix;
[0054] Represents the activation function of the decoding part;
[0055] Represents the reconstructed knowledge point row vector;
[0056] S224: Optimizing the encoding part and the decoding part parameter matrix by minimizing the reconstruction loss, and using the optimized encoding part to encode the knowledge point to obtain the compressed vector of the knowledge point, wherein the reconstruction loss function calculation formula is:
[0057] ;
[0058] in;
[0059] represents the reconstruction loss;
[0060] Represents the row vector of knowledge point i;
[0061] S225: Perform weighted fusion encoding on the exercises according to the strong association matrix of the exercise knowledge points and the knowledge point encoding vector to obtain the exercise semantic encoding vector, and then concatenate the exercise interaction results to obtain the final semantic result vector. The calculation formula is:
[0062] ;
[0063] in:
[0064] Represents the compressed vector weight of the mth knowledge point;
[0065] Represents the semantic encoding vector of exercise e;
[0066] Represents the compressed vector of knowledge point m;
[0067] represents the set of associated knowledge point indexes of exercise e. Optionally, the attention weighting processing in step S24 includes:
[0068] S241: Calculate the energy score of the hidden state at any time in all hidden states of the LSTM layer. The calculation formula is:
[0069] ;
[0070] in:
[0071] Respectively represent the hidden state at time t;
[0072] and denote the query vector weight matrix and the key vector weight matrix respectively;
[0073] represents the parameter vector;
[0074] Represents the hidden state energy score at time t;
[0075] S242: Calculate the attention weight according to the hidden state energy score. The calculation formula is:
[0076] ;
[0077] in:
[0078] represents the attention weight at time t;
[0079] S243: The hidden state is weighted in combination with the attention weight to obtain a context vector, and the calculation formula is:
[0080] ;
[0081] in:
[0082] They represent the hidden states at time t respectively.
[0083] Optionally, in step S3, optimizing and solving the constructed learning behavior quantitative evaluation model to obtain an optimal model instance includes:
[0084] S31: Randomly generate the initial population, including the location of the artificial fish and the initial step length , and set the number of iterations R, the step size adjustment rate ;
[0085] S32: Calculate the fitness value of each artificial fish according to the objective function of the quantitative evaluation model;
[0086] S33: Iteratively calculate the position change of each artificial fish, and update the position of each artificial fish according to the position change. The calculation formula is:
[0087] ;
[0088] in:
[0089] and Respectively represent the position of the i-th artificial fish at time t and time t+1;
[0090] represents the position change of the i-th artificial fish at time t;
[0091] The position change is adjusted based on the distance between the i-th artificial fish and the preset number of nearest neighboring fish. The calculation formula is:
[0092] ;
[0093] in:
[0094] M represents the number of preset neighboring fish;
[0095] represents the position of the jth artificial fish at time t;
[0096] represents the step length of the i-th artificial fish at time t;
[0097] represents the Euclidean distance between the i-th artificial fish and the j-th artificial fish at time t;
[0098] S34: Update the step length of each artificial fish, the calculation formula is:
[0099] ;
[0100] in:
[0101] represents the step length of the i-th artificial fish at time t+1;
[0102] represents the fitness value of the i-th fish at time t;
[0103] represents the average fitness value of all fish;
[0104] S35: Iteratively execute steps S33 and S34 until the maximum number of times is reached. The artificial fish position with the best fitness is the optimal model parameter, and the model is instantiated according to the optimal model parameter to obtain the optimal model instance.
[0105] Optionally, in step S4, performing comprehensive quantitative evaluation on the calculated learning behavior quantitative evaluation vector to obtain a learning behavior quantitative evaluation result includes:
[0106] S41: summing the knowledge point association matrix by columns to obtain the association degree of each knowledge point, and normalizing it to obtain the weight parameter of each knowledge point;
[0107] S42: performing an inner product operation on the weight parameter obtained by solving and the learning behavior quantitative evaluation vector to obtain the final learning system behavior quantitative evaluation result.
[0108] Compared with the prior art, the present invention proposes a learning behavior quantitative evaluation method, which has the following advantages:
[0109] First, this scheme proposes a temporal evolution method based on the attention mechanism. This scheme divides the learning behavior data according to knowledge points, which can better capture the correlation and differences between different knowledge points, and is conducive to the accuracy of subsequent quantitative evaluation of learning behaviors. It also adopts a temporal evolution model based on the attention mechanism to quantitatively evaluate the exercise interaction sub-data. By considering the importance of each time step, the model can pay more attention to the learning behavior characteristics of the corresponding knowledge points, thereby more effectively quantifying and evaluating the learning behavior.
[0110] At the same time, this scheme proposes an artificial fish swarm algorithm with adaptive step size, which is used to optimize and solve the quantitative evaluation model of learning behavior. It can dynamically adjust the search step size, thereby improving the optimization efficiency and convergence speed of the model.
[0111] In addition, this solution comprehensively uses methods such as data partitioning based on knowledge points, the temporal evolution model of the attention mechanism, the artificial fish swarm algorithm with adaptive step size, and comprehensive quantitative evaluation to achieve quantitative evaluation of learning behavior and provide auxiliary decision support. BRIEF DESCRIPTION OF THE DRAWINGS
[0112] Figure 1 A flowchart of a method for quantitatively evaluating learning behavior provided in one embodiment of the present invention.
[0113] The realization of the purpose, functional features and advantages of the present invention will be further explained in conjunction with embodiments and with reference to the accompanying drawings. DETAILED DESCRIPTION
[0114] It should be understood that the specific embodiments described herein are only used to explain the present invention, and are not used to limit the present invention.
[0115] The embodiment of the present application provides a method for quantitatively evaluating learning behavior. The execution subject of the method for quantitatively evaluating learning behavior includes but is not limited to at least one of the electronic devices such as a server and a terminal that can be configured to execute the method provided by the embodiment of the present application. In other words, the method for quantitatively evaluating learning behavior can be executed by software or hardware installed on a terminal device or a server device, and the software can be a blockchain platform. The server includes but is not limited to: a single server, a server cluster, a cloud server or a cloud server cluster, etc.
[0116] Example 1
[0117] A learning behavior quantitative evaluation method comprises the following steps:
[0118] S1: Collecting learning behavior data, the learning behavior data includes exercise knowledge point association data and exercise interaction data; and dividing the exercise interaction data according to the knowledge points to obtain exercise interaction sub-data, wherein data segmentation based on the exercise knowledge point strong association matrix is an implementation method for dividing the exercise interaction data;
[0119] In the step S1, learning behavior data is collected, and exercise interaction data is divided according to knowledge points to obtain exercise interaction sub-data, including:
[0120] S11: Collect learning behavior data, which includes exercise knowledge point association data and exercise interaction data, wherein the exercise knowledge point association data records the knowledge points involved in each exercise and is represented by an exercise knowledge point association matrix A, where the rows of the matrix correspond to exercises, the columns correspond to knowledge points, and the elements Indicates whether exercise e involves knowledge point k, with a value of 1 for involved and a value of 0 for not involved; the exercise interaction data records the students' answers to the exercises and is represented by the exercise interaction matrix B. The rows of the matrix correspond to students, the columns correspond to exercises, and the elements Indicates whether the answer of student m to exercise e is correct, with a value of 1 for correct and 0 for wrong;
[0121] S12: Calculate the knowledge point coverage matrix C based on the exercise knowledge point association matrix A and the exercise interaction matrix B, where each element in the knowledge point coverage matrix C represents each student's mastery of each knowledge point. The calculation formula is:
[0122] ;
[0123] in:
[0124] Indicates the mastery of the kth knowledge point by the mth student;
[0125] Indicates the answer of the mth student to the eth exercise;
[0126] Indicates the knowledge points involved in the e-th exercise, and e represents the exercise index;
[0127] S13: Perform singular value decomposition on the calculated knowledge point coverage matrix. The calculation formula is: ,in represents the singular value matrix, represents the left singular vector matrix, represents the right singular vector matrix. The elements whose singular values in the decomposed singular value matrix are less than the specified threshold are set to 0, that is, the non-significant correlation items between the exercises and the knowledge points are eliminated, and the singular value matrix with strong correlation between the exercises and the knowledge points is obtained. ;
[0128] S14: Reconstruct the knowledge point coverage matrix according to the calculated strongly correlated singular value matrix, and the calculation formula is:
[0129] ;
[0130] in:
[0131] represents the reconstructed knowledge point coverage matrix,
[0132] The strong correlation matrix of the exercise knowledge points is calculated based on the reconstructed knowledge point coverage matrix. The calculation formula is:
[0133] ;
[0134] in:
[0135] Represents the strong correlation matrix of the knowledge points in the exercise;
[0136] represents the inverse of the exercise interaction matrix,
[0137] The elements in the strong correlation matrix of exercise knowledge points whose element values exceed the preset threshold are set to 1, indicating that there is a strong correlation between the exercise and the knowledge point, and the rest are set to 0. The exercise interaction data of students on different knowledge points are divided, and the exercises with element values of 1 in the strong correlation matrix of exercise knowledge points are used as the exercise interaction sub-data of students on the corresponding knowledge points.
[0138] S2: Construct a quantitative evaluation model for learning behavior, which takes the exercise interaction sub-data sequence as input and the quantitative evaluation results of the learning behavior of the corresponding knowledge points as output, wherein the time series evolution model based on the attention mechanism is the implementation method of the quantitative evaluation model.
[0139] The step S2 constructs a learning behavior quantitative evaluation model, including:
[0140] The model includes: input layer, embedding layer, LSTM layer, attention layer and output layer. The specific process of quantitative evaluation of learning behavior includes:
[0141] S21: The input layer receives the student's exercise interaction sub-data sequence ,in, represents the student's exercise interaction result at time step t, and T represents the length of the exercise interaction sub-data sequence;
[0142] S22: The embedding layer converts the discrete exercise interaction results into semantic result vectors;
[0143] S23: LSTM layer semantic result vector according to time step t and the hidden state at the previous time step and cell status Output the hidden state of the current time step and cell status ;
[0144] S24: Attention layer for all hidden states of LSTM layer Perform attention weighting processing to obtain weighted context vector and attention weight;
[0145] S25: The output layer converts the context vector and hidden state The combination of is used to obtain the quantitative evaluation results of students' learning behavior through a fully connected network. The calculation formula is:
[0146] ;
[0147] in:
[0148] Represents the weight matrix parameters of the fully connected network;
[0149] Represents the bias vector parameters of the fully connected network.
[0150] The step S22 converts the discrete exercise interaction results into a semantic result vector, including:
[0151] S221: constructing a knowledge point association matrix, wherein the elements of the knowledge point association matrix represent associations between knowledge points, a value of 1 indicates that an association exists, and a value of 0 indicates that an association does not exist;
[0152] S222: Aggregate the adjacent node information of the knowledge point to generate a compression vector of the corresponding knowledge point. The calculation formula is:
[0153] ;
[0154] in:
[0155] Represents the compressed vector of knowledge point i;
[0156] Represents the row vector of knowledge point j processed according to one-hot encoding;
[0157] Represents the association relationship between knowledge point i and knowledge point j in the knowledge point association matrix;
[0158] represents the encoding weight matrix;
[0159] f represents the activation function of the encoding part;
[0160] N represents the number of adjacent knowledge points of knowledge point i;
[0161] S223: Remap the encoded compressed vector to the reconstructed knowledge point row vector. The calculation formula is:
[0162] ;
[0163] in:
[0164] Represents the compressed vector of knowledge point j;
[0165] represents the decoding weight matrix;
[0166] Represents the activation function of the decoding part;
[0167] Represents the reconstructed knowledge point row vector;
[0168] S224: Optimizing the encoding part and the decoding part parameter matrix by minimizing the reconstruction loss, and using the optimized encoding part to encode the knowledge point to obtain the compressed vector of the knowledge point, wherein the reconstruction loss function calculation formula is:
[0169] ;
[0170] in;
[0171] represents the reconstruction loss;
[0172] Represents the row vector of knowledge point i;
[0173] S225: Perform weighted fusion encoding on the exercises according to the strong association matrix of the exercise knowledge points and the knowledge point encoding vector to obtain the exercise semantic encoding vector, and then concatenate the exercise interaction results to obtain the final semantic result vector. The calculation formula is:
[0174] ;
[0175] in:
[0176] Represents the compressed vector weight of the mth knowledge point;
[0177] Represents the semantic encoding vector of exercise e;
[0178] Represents the compressed vector of knowledge point m;
[0179] Represents the set of associated knowledge point indexes of exercise e.
[0180] The attention weighting process in step S24 includes:
[0181] S241: Calculate the energy score of the hidden state at any time in all hidden states of the LSTM layer. The calculation formula is:
[0182] ;
[0183] in:
[0184] Respectively represent the hidden state at time t;
[0185] and denote the query vector weight matrix and the key vector weight matrix respectively;
[0186] represents the parameter vector;
[0187] Represents the hidden state energy score at time t;
[0188] S242: Calculate the attention weight according to the hidden state energy score. The calculation formula is:
[0189] ;
[0190] in:
[0191] represents the attention weight at time t;
[0192] S243: The hidden state is weighted in combination with the attention weight to obtain a context vector, and the calculation formula is:
[0193] ;
[0194] in:
[0195] represents the attention weight at time t;
[0196] They represent the hidden states at time t respectively.
[0197] S3: Optimize and solve the constructed learning behavior quantitative evaluation model to obtain the optimal model instance, and use the optimal model instance obtained to perform quantitative evaluation of the learning behavior of all knowledge points, and integrate the quantitative evaluation results to obtain the learning behavior quantitative evaluation vector, wherein the artificial fish swarm algorithm with adaptive step size is the implementation method of the optimization solution.
[0198] In the step S3, the constructed learning behavior quantitative evaluation model is optimized and solved to obtain the optimal model instance, including:
[0199] S31: Randomly generate the initial population, including the location of the artificial fish and the initial step length , and set the number of iterations R, the step size adjustment rate ;
[0200] S32: Calculate the fitness value of each artificial fish according to the objective function of the quantitative evaluation model;
[0201] S33: Iteratively calculate the position change of each artificial fish, and update the position of each artificial fish according to the position change. The calculation formula is:
[0202] ;
[0203] in:
[0204] and Respectively represent the position of the i-th artificial fish at time t and time t+1;
[0205] represents the position change of the i-th artificial fish at time t;
[0206] The position change is adjusted based on the distance between the i-th artificial fish and the preset number of nearest neighboring fish. The calculation formula is:
[0207] ;
[0208] in:
[0209] M represents the number of preset neighboring fish;
[0210] represents the position of the jth artificial fish at time t;
[0211] represents the step length of the i-th artificial fish at time t;
[0212] represents the Euclidean distance between the i-th artificial fish and the j-th artificial fish at time t;
[0213] S34: Update the step length of each artificial fish, the calculation formula is:
[0214] ;
[0215] in:
[0216] represents the step length of the i-th artificial fish at time t+1;
[0217] Indicates the step size adjustment rate;
[0218] represents the fitness value of the i-th fish at time t;
[0219] represents the average fitness value of all fish;
[0220] S35: Iteratively execute steps S33 and S34 until the maximum number of times is reached. The artificial fish position with the best fitness is the optimal model parameter, and the model is instantiated according to the optimal model parameter to obtain the optimal model instance.
[0221] S4: Perform comprehensive quantitative evaluation on the calculated learning behavior quantitative evaluation vector to obtain the learning behavior quantitative evaluation result.
[0222] In step S4, the calculated learning behavior quantitative evaluation vector is subjected to comprehensive quantitative evaluation to obtain a learning behavior quantitative evaluation result, including:
[0223] S41: summing the knowledge point association matrix by columns to obtain the association degree of each knowledge point, and normalizing it to obtain the weight parameter of each knowledge point;
[0224] S42: performing an inner product operation on the weight parameter obtained by solving and the learning behavior quantitative evaluation vector to obtain the final learning system behavior quantitative evaluation result.
[0225] It should be noted that the serial numbers of the above embodiments of the present invention are only for description and do not represent the advantages and disadvantages of the embodiments. And the terms "including", "comprising" or any other variants thereof in this article are intended to cover non-exclusive inclusion, so that a process, device, article or method including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, device, article or method. In the absence of further restrictions, an element defined by the sentence "including a ..." does not exclude the presence of other identical elements in the process, device, article or method including the element.
[0226] Through the description of the above implementation methods, those skilled in the art can clearly understand that the above-mentioned embodiment methods can be implemented by means of software plus a necessary general hardware platform, and of course by hardware, but in many cases the former is a better implementation method. Based on such an understanding, the technical solution of the present invention is essentially or the part that contributes to the prior art can be embodied in the form of a software product, which is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk) as described above, and includes a number of instructions for a terminal device (which can be a mobile phone, a computer, a server, or a network device, etc.) to execute the methods described in each embodiment of the present invention.
[0227] The above are only preferred embodiments of the present invention, and are not intended to limit the patent scope of the present invention. Any equivalent structure or equivalent process transformation made using the contents of the present invention specification and drawings, or directly or indirectly applied in other related technical fields, are also included in the patent protection scope of the present invention.
Claims
1. A learning behavior quantitative evaluation method, characterized in that: The method comprises: S1: Collect learning behavior data, the learning behavior data includes exercise knowledge point association data and exercise interaction data; and divide the exercise interaction data according to the knowledge points to obtain exercise interaction sub-data, and divide the exercise interaction data based on the data segmentation of the exercise knowledge point strong association matrix; In the step S1, learning behavior data is collected, and exercise interaction data is divided according to knowledge points to obtain exercise interaction sub-data, including: S11: Collect learning behavior data, which includes exercise knowledge point association data and exercise interaction data, wherein the exercise knowledge point association data records the knowledge points involved in each exercise and is represented by an exercise knowledge point association matrix A, where the rows of the matrix correspond to exercises, the columns correspond to knowledge points, and the elements Indicates whether exercise e involves knowledge point k, with a value of 1 for involved and a value of 0 for not involved; the exercise interaction data records the students' answers to the exercises and is represented by the exercise interaction matrix B. The rows of the matrix correspond to students, the columns correspond to exercises, and the elements Indicates whether the answer of student m to exercise e is correct, with a value of 1 for correct and 0 for wrong; S12: Calculate the knowledge point coverage matrix C based on the exercise knowledge point association matrix A and the exercise interaction matrix B, where each element in the knowledge point coverage matrix C represents each student's mastery of each knowledge point. The calculation formula is: ; in: Indicates the mastery of the kth knowledge point by the mth student; Indicates the answer of the mth student to the eth exercise; Indicates the knowledge points involved in the e-th exercise, and e represents the exercise index; S13: Perform singular value decomposition on the calculated knowledge point coverage matrix. The calculation formula is: ,in represents the singular value matrix, represents the left singular vector matrix, represents the right singular vector matrix. The elements whose singular values in the decomposed singular value matrix are less than the specified threshold are set to 0, that is, the non-significant correlation items between the exercises and the knowledge points are eliminated, and the singular value matrix with strong correlation between the exercises and the knowledge points is obtained. ; S14: Reconstruct the knowledge point coverage matrix according to the calculated strongly correlated singular value matrix, and the calculation formula is: ; in: represents the reconstructed knowledge point coverage matrix, The strong correlation matrix of the exercise knowledge points is calculated based on the reconstructed knowledge point coverage matrix. The calculation formula is: ; in: Represents the strong correlation matrix of the knowledge points in the exercise; represents the inverse of the exercise interaction matrix, Set the elements in the strong correlation matrix of exercise knowledge points whose values exceed the preset threshold to 1, indicating that the exercise and the knowledge point are strongly correlated, and set the others to 0, divide the students' exercise interaction data on different knowledge points, and use the exercises with element values of 1 in the strong correlation matrix of exercise knowledge points as the students' exercise interaction sub-data on the corresponding knowledge points; S2: Construct a learning behavior quantitative evaluation model, which takes the exercise interaction sub-data sequence as input and the learning behavior quantitative evaluation results of the corresponding knowledge points as output, wherein the time series evolution model based on the attention mechanism is the implementation method of the quantitative evaluation model; Step S2 includes: constructing a knowledge point association matrix, wherein the elements of the knowledge point association matrix represent the association relationship between knowledge points, a value of 1 indicates the existence of an association relationship, and a value of 0 indicates the absence of an association relationship; S3: Optimizing and solving the constructed learning behavior quantitative evaluation model to obtain an optimal model instance, and using the optimal model instance obtained by the solution to perform quantitative evaluation of learning behavior on all knowledge points, and integrating the quantitative evaluation results to obtain a learning behavior quantitative evaluation vector, wherein the artificial fish swarm algorithm with adaptive step size is the implementation method of the optimization solution; S4: Performing a comprehensive quantitative evaluation on the calculated learning behavior quantitative evaluation vector to obtain a learning behavior quantitative evaluation result, including: S41: Sum the knowledge point association matrix by columns to obtain the association degree of each knowledge point, and normalize it to obtain the weight parameter of each knowledge point; S42: performing an inner product operation on the weight parameter obtained by solving and the learning behavior quantitative evaluation vector to obtain the final learning system behavior quantitative evaluation result.
2. A learning behavior quantitative evaluation method as claimed in claim 1, characterized in that: The step S2 constructs a learning behavior quantitative evaluation model, including: The model includes: input layer, embedding layer, LSTM layer, attention layer and output layer. The specific process of quantitative evaluation of learning behavior includes: S21: The input layer receives the student's exercise interaction sub-data sequence ,in, represents the student's exercise interaction result at time step t, and T represents the length of the exercise interaction sub-data sequence; S22: The embedding layer converts the discrete exercise interaction results into semantic result vectors; S23: LSTM layer semantic result vector according to time step t and the hidden state at the previous time step and cell status Output the hidden state of the current time step and cell status ; S24: Attention layer for all hidden states of LSTM layer Perform attention weighting processing to obtain weighted context vector and attention weight; S25: The output layer converts the context vector and hidden state The combination of is used to obtain the quantitative evaluation results of students' learning behavior through a fully connected network. The calculation formula is: ; in: Represents the weight matrix parameters of the fully connected network; Represents the bias vector parameters of the fully connected network.
3. A learning behavior quantitative evaluation method as claimed in claim 2, characterized in that: The step S22 converts the discrete exercise interaction results into a semantic result vector, including: S221: constructing a knowledge point association matrix, wherein the elements of the knowledge point association matrix represent associations between knowledge points, a value of 1 indicates that an association exists, and a value of 0 indicates that an association does not exist; S222: Aggregate the adjacent node information of the knowledge point to generate a compression vector of the corresponding knowledge point. The calculation formula is: ; in: Represents the compressed vector of knowledge point i; Represents the row vector of knowledge point j processed according to one-hot encoding; Represents the association relationship between knowledge point i and knowledge point j in the knowledge point association matrix; represents the encoding weight matrix; f represents the activation function of the encoding part; N represents the number of adjacent knowledge points of knowledge point i; S223: Remap the encoded compressed vector to the reconstructed knowledge point row vector. The calculation formula is: ; in: Represents the compressed vector of knowledge point j; represents the decoding weight matrix; Represents the activation function of the decoding part; Represents the reconstructed knowledge point row vector; S224: Optimizing the encoding part and the decoding part parameter matrix by minimizing the reconstruction loss, and using the optimized encoding part to encode the knowledge point to obtain the compressed vector of the knowledge point, wherein the reconstruction loss function calculation formula is: ; in; represents the reconstruction loss; Represents the row vector of knowledge point i; S225: Perform weighted fusion encoding on the exercises according to the strong association matrix of the exercise knowledge points and the knowledge point encoding vector to obtain the exercise semantic encoding vector, and then concatenate the exercise interaction results to obtain the final semantic result vector. The calculation formula is: ; in: Represents the compressed vector weight of the mth knowledge point; Represents the semantic encoding vector of exercise e; Represents the compressed vector of knowledge point m; Represents the set of associated knowledge point indexes of exercise e.
4. A learning behavior quantitative evaluation method as claimed in claim 2, characterized in that: The attention weighting process in step S24 includes: S241: Calculate the energy score of the hidden state at any time in all hidden states of the LSTM layer. The calculation formula is: ; in: Respectively represent the hidden state at time t; and denote the query vector weight matrix and the key vector weight matrix respectively; represents the parameter vector; Represents the hidden state energy score at time t; S242: Calculate the attention weight according to the hidden state energy score. The calculation formula is: ; in: represents the attention weight at time t; S243: The hidden state is weighted in combination with the attention weight to obtain a context vector, and the calculation formula is: ; in: They represent the hidden states at time t respectively.
5. A learning behavior quantitative evaluation method as claimed in claim 1, characterized in that: In the step S3, the constructed learning behavior quantitative evaluation model is optimized and solved to obtain the optimal model instance, including: S31: Randomly generate the initial population, including the location of the artificial fish and initial step length , and set the number of iterations R, the step size adjustment rate ; S32: Calculate the fitness value of each artificial fish according to the objective function of the quantitative evaluation model; S33: Iteratively calculate the position change of each artificial fish, and update the position of each artificial fish according to the position change. The calculation formula is: ; in: and Respectively represent the position of the i-th artificial fish at time t and time t+1; represents the position change of the i-th artificial fish at time t; The position change is adjusted based on the distance between the i-th artificial fish and the preset number of nearest neighboring fish. The calculation formula is: ; in: M represents the number of preset neighboring fish; represents the position of the jth artificial fish at time t; represents the step length of the i-th artificial fish at time t; represents the Euclidean distance between the i-th artificial fish and the j-th artificial fish at time t; S34: Update the step length of each artificial fish, the calculation formula is: ; in: represents the step length of the i-th artificial fish at time t+1; represents the fitness value of the i-th fish at time t; represents the average fitness value of all fish; S35: Iteratively execute steps S33 and S34 until the maximum number of times is reached. The artificial fish position with the best fitness is the optimal model parameter, and the model is instantiated according to the optimal model parameter to obtain the optimal model instance.
