Metacognitive competence prediction method and device, equipment and storage medium

Abstract

The invention discloses a meta-cognitive competence prediction method, device and equipment and a storage medium, and relates to the technical field of data processing, and the method comprises the following steps: obtaining meta-cognitive competence data and learning behavior data of a learner; performing correlation analysis and clustering on the meta-cognitive competence data and the learning behavior data to obtain key behavior characterization indexes of the meta-cognitive competence; and converting the key behavior characterization indexes into score values. According to the method, a plurality of different behavior representation frameworks do not need to be constructed, the key behavior representation indexes are directly obtained through a correlation analysis and clustering method, the index obtaining time is greatly shortened, and the obtaining efficiency is improved. On the basis, the key behavior characterization indexes are converted into score values, the problem that the neural network is prone to local optimum due to the fact that existing behavior characterization indexes have large differences is solved, and the accuracy of the score values is greatly improved.

Description

Technical Field

[0001] The present invention relates to the field of data processing technology, and in particular to a metacognitive ability prediction method, device, equipment and storage medium. Background Art

[0002] Metacognitive ability refers to the learner's ability to plan, monitor, evaluate and regulate his or her own cognition and learning process (Flavell, 1979). As a core quality that represents students' autonomy and initiative, metacognitive ability is a key factor in making students autonomous learners. At the same time, metacognitive ability is an important factor affecting students' learning quality, especially in online learning environments that require higher autonomy and initiative. Improving learners' metacognitive ability is a powerful guarantee for improving learning quality and building a high-quality education system.

[0003] At present, when characterizing metacognitive ability, it is necessary to first obtain behavioral representation indicators, then convert the behavioral representation indicators into corresponding score values, and characterize metacognitive ability according to the score values. When obtaining behavioral representation indicators, it is usually necessary to construct a variety of different behavioral representation frameworks. For example, in online learning situations, Cerezo et al. applied inductive mining algorithms to classify learning behaviors into five categories: planning, learning, practice, discussion, and review, and constructed a behavioral representation framework for metacognitive ability. There are also studies that use deep learning technology to construct a behavioral representation framework for metacognitive ability by collecting behavioral data such as learners viewing learning goals and watching learning videos (Cheng et al., 2021). Wang Hongjiang et al. constructed a behavioral framework for representing metacognitive ability based on a systematic literature review method, and divided a total of 39 learning behavior indicators to characterize learners' metacognitive ability. Then multiple behavioral data are input into a variety of different behavioral representation frameworks to obtain corresponding behavioral representation indicators. Finally, the behavioral representation indicators are mapped to corresponding score values ​​through the existing neural network model.

[0004] In summary, the existing behavior characterization index acquisition lacks a unified behavior characterization framework, which results in a long time and low efficiency in index acquisition. When calculating the score value, due to the use of different behavior characterization frameworks, there are large differences in the behavior characterization indicators obtained, which makes the neural network prone to fall into the local optimum, resulting in inaccurate final calculated score values. Summary of the invention

[0005] The present invention provides a metacognitive ability prediction method, device, equipment and storage medium, which solves the problem that some behavioral representation indicators lack a unified behavioral representation framework, resulting in a long indicator acquisition time, low efficiency, and inaccurate calculated score values.

[0006] In a first aspect, the present invention provides a method for predicting metacognitive ability, comprising the following steps:

[0007] Obtain learners' metacognitive ability data and learning behavior data;

[0008] Conduct correlation analysis and clustering on metacognitive ability data and learning behavior data to obtain key behavioral representation indicators of metacognitive ability;

[0009] Convert key behavioral representation indicators into score values, and use the score values ​​to predict metacognitive ability;

[0010] The key behavior characterization indicators are converted into score values, including:

[0011] Obtain the correlation coefficients of key behavior representation indicators and construct a correlation coefficient matrix;

[0012] Obtain multiple eigenvalues ​​and multiple eigenvectors of the correlation coefficient matrix, and generate principal components according to the multiple eigenvectors;

[0013] Obtain multiple information contribution rates of each eigenvalue to the principal component, multiply and sum the multiple information contribution rates with the corresponding principal component to obtain a score value.

[0014] Preferably, correlation analysis and clustering are performed on the metacognitive ability data and the learning behavior data to obtain key behavioral representation indicators of metacognitive ability, including the following steps:

[0015] Perform correlation analysis on metacognitive ability data and learning behavior data, and obtain the corresponding correlation coefficient; the correlation coefficient is obtained by the following formula:

[0016]

[0017] In the formula, r s represents the correlation coefficient, n represents the number of samples of learning behavior data obtained, and d i Indicates the level difference between learning behavior data and metacognitive ability data;

[0018] Compare the correlation coefficient with the threshold, and use the corresponding learning behavior data with a correlation coefficient greater than the threshold as the relevant behavior representation index of the learner's metacognitive ability;

[0019] The relevant behavioral representation indicators are clustered using the R-type clustering algorithm to obtain the key behavioral representation indicators of metacognitive ability.

[0020] Preferably, after converting the key behavior characterization indicators into score values, the following is further included:

[0021] Mining learning behavior data to obtain learning behavior sequences, and performing difference analysis on learning behavior sequences to obtain key behavior sequences;

[0022] The key behavior sequences are converted into prediction values, and the development trend of meta-capabilities is characterized by the prediction values.

[0023] Preferably, mining the learning behavior data to obtain the learning behavior sequence includes the following steps:

[0024] Constructing state sequence variables and observation sequence variables based on Markov, wherein the state sequence variables are hidden variables and the observation sequence variables are learning behavior sequences;

[0025] Based on the dependency relationship between the state sequence variables and the observation value sequence variables, the joint probability between the state sequence variables and the observation value sequence variables is obtained;

[0026] Based on the joint probability, the state transition probability between the state sequence variable and the observation value sequence variable is obtained;

[0027] The learning behavior data is mined based on the state transition probability to obtain the learning behavior sequence.

[0028] Preferably, the learning behavior sequence is differentially analyzed by using the differential sequence mining technology DSM and the piecewise linear segmentation algorithm PLS to obtain the key behavior sequence, which specifically includes the following steps:

[0029] The PLS algorithm is used to divide the learner's learning behavior sequence into the first behavior data set and the second behavior data set;

[0030] Use the DSM algorithm to mine multiple frequent sequence patterns in the first behavior data set and the second behavior data set respectively;

[0031] Obtain the frequencies iSupportleft and iSupportright of a specific subsequence appearing in a certain behavior sequence in the first behavior data set and the second behavior data set respectively;

[0032] For each frequent sequence pattern, a t-test is used to test the average values ​​of iSupportleft and iSupportright of the pattern in the first and second behavior data sets and the size of the set threshold;

[0033] If the average value is greater than the set threshold, it is considered that the frequent sequence pattern has a difference between the first behavior data set and the second behavior data set, and the key behavior sequence is obtained.

[0034] Preferably, converting the key behavior sequence into a prediction value includes:

[0035] Input the key behavior sequence into the Holt two-parameter exponential smoothing model to obtain the predicted value;

[0036] The Holt two-parameter exponential smoothing model is shown as follows:

[0037] F t+m =S t +b t m.

[0038] in,

[0039] S t =αx t +(1-α)(S t-1 +b t-1 ).

[0040] b t =γ(S t -S t-1 )+(1-γ)b t-1 .

[0041] In the formula, F represents the predicted value, S represents the development trend of the behavior sequence, b represents the change in the development trend of the behavior sequence, m represents the interval number of the development trend, x represents the key behavior sequence, α and γ represent the model parameters, and t represents the period;

[0042] The parameters of the Holt two-parameter exponential smoothing model are adjusted by adaptive smoothing parameters, as shown below:

[0043]

[0044] In the formula, α′ t and γ′ t represents the adaptive smoothing parameter.

[0045] In a second aspect, the present invention further provides a metacognitive ability prediction device, comprising:

[0046] An acquisition module is used to acquire learners’ metacognitive ability data and learning behavior data;

[0047] Clustering module, used to perform correlation analysis and clustering on metacognitive ability data and learning behavior data, and obtain key behavioral representation indicators of metacognitive ability;

[0048] A conversion module, used to convert key behavioral representation indicators into score values;

[0049] The conversion module comprises:

[0050] A construction module is used to obtain the correlation coefficients of key behavior representation indicators and construct a correlation coefficient matrix;

[0051] A generation module, used for obtaining multiple eigenvalues ​​and multiple eigenvectors of the correlation coefficient matrix, and generating principal components according to the multiple eigenvectors;

[0052] The calculation module is used to obtain multiple information contribution rates of each eigenvalue to the principal component, multiply the multiple information contribution rates with the corresponding principal component and sum them to obtain a score value.

[0053] In a third aspect, the present invention further provides a computer device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the above-mentioned metacognitive ability prediction method when executing the program.

[0054] In a fourth aspect, the present invention further provides a computer-readable storage medium, wherein the storage medium stores a computer program, and when the computer program is executed by a processor, the above-mentioned metacognitive ability prediction method is implemented.

[0055] Compared with the prior art, the present invention has the following beneficial effects:

[0056] The present invention first obtains the learner's metacognitive ability data and learning behavior data, and then performs correlation analysis and clustering on the metacognitive ability data and learning behavior data to obtain key behavioral characterization indicators of metacognitive ability. The present invention does not need to construct a variety of different behavioral characterization frameworks, but directly obtains key behavioral characterization indicators through correlation analysis and clustering methods, which greatly shortens the indicator acquisition time and improves the acquisition efficiency. On this basis, the key behavioral characterization indicators are converted into score values. The differences between the indicators obtained by clustering in the present invention are very small. Furthermore, in the conversion process, principal component analysis is used to calculate the eigenvalues ​​and principal components of the key behavioral characterization indicators, and the score values ​​of the behavioral indicators are expressed by calculating the contribution rate of the eigenvalues ​​to the principal components. This avoids the problem that the existing behavioral characterization indicators have large differences, which makes the neural network prone to fall into local optimality, and greatly improves the accuracy of the score values. BRIEF DESCRIPTION OF THE DRAWINGS

[0057] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0058] Figure 1 A flow chart of a method for predicting metacognitive ability of the present invention;

[0059] Figure 2 It is a schematic diagram of the structure of the hidden Markov model of the present invention;

[0060] Figure 3 It is a schematic diagram of the hierarchical structure model of key behavior characterization indicators of metacognitive ability of the present invention;

[0061] Figure 4 It is a schematic diagram of the multi-layer feedforward neural network structure of the present invention. DETAILED DESCRIPTION

[0062] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0063] Aiming at the needs and bottlenecks of dynamic representation and intelligent diagnosis of metacognitive ability, and targeting the multidimensional, implicit and dynamic development characteristics of metacognitive ability, the present invention realizes implicit metacognitive ability based on explicit behavioral data, and provides a metacognitive ability prediction method. Figure 1 , including the following steps:

[0064] Step 1: Obtain learners’ metacognitive ability data and explicit learning behavior data.

[0065] In view of the multidimensional and implicit characteristics of metacognitive ability, based on metacognitive theory and related research on metacognitive ability models, a "metacognitive ability framework for online learners" is constructed, including metacognitive knowledge, metacognitive monitoring and metacognitive regulation. Among them, metacognitive knowledge is a metacognitive state before the start of a learning task, including cognitive strategies, learning motivation, and self-efficacy. Metacognitive regulation is reflection in action, including setting goals, making plans, and implementing strategies. Metacognitive monitoring is reflection on action, including learners' evaluation of factual knowledge (tasks and content), procedural knowledge (exploration process), and conditional knowledge (strategies). Self-report tools such as questionnaires are used to investigate learners' metacognitive ability levels.

[0066] Based on this framework, the "Online Learner Metacognitive Ability Questionnaire" was developed to further improve and verify the "Online Learner Metacognitive Ability Framework" on the one hand, and to capture the implicit metacognitive ability data of online learners on the other hand. Metacognitive ability data specifically refers to cognitive data that describes cognition, that is, data about individuals' monitoring, evaluation, and regulation of their own cognitive processes and results. This includes the digitized records of the three main components of individual cognitive knowledge, experience, and monitoring.

[0067] In view of the dynamic development characteristics of metacognitive ability, based on self-directed learning theory and related research on online learning behavior mining, an "online learning behavior framework" is constructed. It includes three dimensions: planning, execution and regulation, and evaluation and reflection. Among them, planning includes goal setting, learning plan and resource acquisition, execution and regulation includes task execution, task observation, task adjustment and interpersonal communication, and evaluation and reflection includes learning evaluation and learning reflection.

[0068] In online learning situations, self-directed learning is the main way to support learners’ efficient learning. Therefore, the present invention describes the online learning behavior labeling system from the perspective of the self-directed learning process of “planning-execution and regulation-evaluation and reflection”. Learning behavior is an important external manifestation of learners’ intrinsic metacognitive characteristics. With the development of online learning and learning analysis technology, technical tools can be used to track and record learners’ learning behaviors in real time, thereby capturing students’ learning behavior data. For example, online learning behaviors mainly include seven categories: login, browsing, resource sharing, testing, works, mutual evaluation and online discussion.

[0069] Based on the "Online Learner Metacognitive Ability Framework" and the "Online Learning Behavior Framework", learners' implicit metacognitive ability data and explicit learning behavior data in online learning situations are captured respectively, and used as sample data for subsequent operations.

[0070] Step 2: Conduct correlation analysis and clustering on metacognitive ability data and learning behavior data to obtain key behavioral representation indicators of metacognitive ability.

[0071] The present invention explores the correlation between explicit learning behavior and implicit metacognitive ability through correlation analysis. Since Spearman correlation analysis has strong robustness, the present invention performs correlation analysis by calculating the Spearman correlation coefficient and obtains multiple coefficient values. Finally, the learning behavior significantly correlated with metacognitive ability (learning behavior with a coefficient value greater than a set threshold) is determined as a metacognitive ability-related behavior characterization indicator.

[0072] The Spearman correlation coefficient was calculated using the following formula:

[0073]

[0074] Among them, n is the number of samples, and d is the level difference between learning behavior data and metacognitive ability data. This step constructs a key behavior representation index system for metacognitive ability, aiming to achieve a static mapping between explicit learning behavior indicators and implicit metacognitive ability.

[0075] Determining metacognitive ability-related behavioral representation indicators based on correlation analysis is prone to generate redundant information. In order to improve the efficiency of calculation and analysis, the present invention uses an R-type clustering algorithm to cluster metacognitive ability-related behavioral representation indicators, aiming to reduce the dimensionality of the relevant behavioral representation indicators, mine typical behavioral representation indicators, and thus construct a key behavioral representation indicator system for metacognitive ability.

[0076] Based on correlation analysis, we construct a set of metacognitive ability related behavior representation indicators B = {b1, b2, ..., b n}. Calculate the similarity coefficient r ij , perform R-type clustering based on indicator set B. Select an appropriate confidence level to determine the optimal number of clusters K, and use the Kruskal-Wallis test to determine the rationality of the optimal number of clusters K. Output clustering result C = {C1, C2, ..., C K}, cluster the related behavior representation index set B into K types of metacognitive ability key behavior representation indicators. At the same time, the Calinski-Harabasz (CH) index, the square error within the group (SSE) and the sum of squares between groups (SSB) are used to comprehensively evaluate the performance of the R-type clustering algorithm. The similarity coefficient is calculated by the following formula:

[0077]

[0078] Another solution is to combine the conditional generalized variance minimization method and the minimum maximum deviation method to calculate the discrimination between behavioral indicators, so as to screen the key behavioral representation indicators of metacognitive ability. Or combine factor analysis and recursive elimination method to construct a key behavioral representation indicator system of metacognitive ability by performing multiple recursive calculations on the behavioral indicator factors.

[0079] Step 3: Convert key behavioral indicators into score values.

[0080] First, calculate the correlation coefficient r between key behavioral indicators ij , thus constructing the correlation coefficient matrix R = (r ij )m×m. Secondly, calculate the eigenvalues ​​{λ1,λ2,…,λ m} and the eigenvectors {u1,u2,…,u m}, thereby generating m principal components {y1,y2,…,y m}.

[0081]

[0082] Then, the key behavior indicator characteristic values ​​{λ1,λ2,…,λ m} for the principal components {y1,y2,…,y m}'s information contribution rate b j , to indicate the size of the characteristic value of the key behavior characterization indicator:

[0083]

[0084] Finally, multiply multiple information contribution rates with the corresponding principal components and sum them to get the score value:

[0085]

[0086] Another possible solution is to use regression analysis, such as linear regression or multivariate regression, or machine learning algorithms, such as decision trees, support vector machines, etc. Then, a score is calculated based on the predictive power of the key behavioral indicators.

[0087] On this basis, further judging the level of learners' metacognitive ability according to certain evaluation criteria is an important research content of metacognitive ability diagnosis. The present invention combines the Delphi method (Delphi), the analytic hierarchy process (AHP) and the random forest algorithm to calculate the evaluation benchmark of metacognitive ability to determine the intensity threshold of metacognitive ability. Among them, Delphi and AHP are used to determine the initial intensity threshold of metacognitive ability, and the random forest algorithm is used to dynamically adjust and optimize the initial intensity threshold. First, a hierarchical structure model of key behavioral representation indicators of metacognitive ability is constructed ( Figure 3 ), using the Delphi method to recruit experts in the field of metacognition and behavioral science to judge the strength threshold of key behavioral representation indicators of metacognitive ability. Then, AHP was used to calculate the initial strength threshold of metacognitive ability. The specific steps are:

[0088] S1: Construct the behavior indicator strength threshold judgment matrix A = (a ij )n×m, where a ij It represents the intensity threshold assigned by the i-th expert to the j-th behavior indicator, n represents the number of behavior indicators, and m represents the number of experts.

[0089] S2: Check the consistency of the judgment matrix and calculate the initial intensity threshold ω of the behavior indicator i (Formula 7).

[0090] S3: Using the weighted superposition method, the initial intensity threshold ω of the behavior indicator is i The initial intensity threshold of metacognitive ability was obtained by stratified summation.

[0091]

[0092] Among them, W i Represents the feature vector composed of the intensity threshold of the i-th behavior indicator.

[0093] Finally, the random forest algorithm is used to construct a metacognitive ability intensity threshold training model. In the training model, the characteristic variables are composed of key behavioral representation indicators of metacognitive ability, and the response variables are composed of metacognitive ability. The random forest model dynamically adjusts the intensity threshold of key behavioral representation indicators by calculating the importance of key behavioral representation indicators to metacognitive ability, thereby achieving dynamic adjustment and optimization of the initial intensity threshold of metacognitive ability.

[0094] Step 4: Mine the learners’ learning behavior data to obtain the learning behavior sequence, and perform difference analysis on the learning behavior sequence to obtain the key behavior sequence.

[0095] Learners' learning behaviors are explicit, easy to observe, and have time-series characteristics. Especially in online learning situations, log technology can automatically track learners' learning behaviors in real time and record the order in which the behaviors occur (time-series). The recorded online learning behavior data usually contains fields such as user ID, timestamp, and specific learning behaviors. Although learners' metacognitive abilities are invisible, the Hidden Markov Model (HMM) can effectively mine the overall learning behavior pattern based on the explicit behavior sequence that occurs during the learning process.

[0096] HMM adds hidden variables that cannot be directly observed. By constructing a Markov chain and calculating the state transition probability, it can realize learning behavior sequence mining. HMM contains two sets of variables, namely state sequence variables (Y) and observation sequence variables (X). Figure 2 As shown. State sequence variable Y = {y1, y2, ..., y n} is a hidden variable, where y i Represents the system state at the i-th moment. Observation sequence variable X={x1,x2,...,x n} represents an explicit learning behavior sequence, where x i Represents the specific learning behavior of the learner at the i-th moment.

[0097] Figure 2 The arrows in the HMM represent the explicit learning behavior sequence variables X = {x1, x2, ..., x n} and latent variables Y = {y1,y2,...,y n}. Among them, the explicit learning behavior sequence variable X={x1,x2,...,x n} depends on the corresponding latent variable Y = {y1,y2,...,y n Based on the dependency relationship between variables, we first calculate the joint probability of variables P(x1y1,…,X n y n), and then calculate the state transition probability A = [a ij ] N×N , where a ij =P(y t+1 =s j |y t =s i ). The joint probability of variables is calculated as follows:

[0098]

[0099] The state transition probability A indicates that if the hidden variable state is s at time t i , then the hidden variable state at time t+1 is s j By calculating the state transition probability, HMM realizes the probability representation of the learning behavior sequence that occurs during the learning process, thereby mining the learning behavior sequence of the learner over time. Another possible solution is to use sequential pattern mining algorithms (Sequential Pattern Mining), such as AprioriAll, GSP, FreeSpan, and PrefixSpan, to identify frequent learning behavior patterns and mine learning behavior sequences.

[0100] On the basis of behavior sequence mining, the present invention further combines differential sequence mining (DSM) and piecewise linear segmentation (PLS) to mine the differences between behavior sequences, and realizes the dynamic mapping between explicit behavior sequences and implicit metacognitive abilities. DSM can extract common behavior sequences from two sets of time series data sets, and mine the differences between common behavior sequences in the two sets of data sets through significance analysis.

[0101] Prepare two sets of time series data sets. The present invention captures the implicit metacognitive ability data of learners in online learning situations based on the "Online Metacognitive Ability Questionnaire", and divides learners into left group (high metacognitive ability group) and right group (low metacognitive ability group) according to the questionnaire scores using the PLS algorithm. Among them, the left group contains the learning behavior data of learners with high metacognitive ability (learner questionnaire score ≧ the average score of all learners), and the right group contains the learning behavior data of learners with low metacognitive ability (learner questionnaire score < the average score of all learners).

[0102] The DSM algorithm is used to mine the frequent sequence patterns (s-frequent) in the left group and right group behavior data sets respectively, and the behavior sequence patterns sFreqleft and sFreqright shared by the left group and the right group are identified by calculating s-support (sequencesupport). s-support indicates the frequency of behavior sequences containing a specific subsequence in the data set. According to existing research, the threshold of s-support is set to 0.5, that is, if the frequency of a subsequence appearing in all behavior sequences exceeds 50%, the sequence is determined to be s-frequent.

[0103] Finally, for all s-frequent frequent sequence patterns mined in the left group and right group data sets, the frequency iSupportleft and iSupportright of a specific subsequence in the two data sets appearing in a certain behavior sequence are calculated respectively through i-support (instance-support). For each s-frequent pattern, the t-test is used to test whether the i-support average values ​​of the pattern in the leftgroup and rightgroup data sets are significantly different. If the significance level p-value is lower than the threshold (usually 0.05 is considered statistically significant), the s-frequent pattern is considered to be significantly different between the left group and right group data sets, and the sequence corresponding to the s-frequent pattern is the key behavior sequence. The key behavior sequence is X={x t ,t=1,2,…,n}.

[0104] Step 5: Convert key behavior sequences into predictive values.

[0105] Constructing metacognitive ability representation based on HMM key behavior sequence X={x t ,t=1,2,…,n}, which is used as the basis for the analysis of the development trend of metacognitive ability. Then, the Holt two-parameter exponential smoothing model is constructed to mine the short-term, medium-term and long-term development trends of metacognitive ability based on the key behavior sequence X. Specifically, the behavior sequence trend smoothing model S is constructed t =αx t +(1-α)(S t-1 +b t-1 ), based on the actual behavior observation value x in period t t Analyze the development trend of behavior sequence t . Constructing a smoothing model of the trend variation of the behavior sequence b t =γ(St -S t-1 )+(1-γ)b t-1 , analyze the development trend of behavior sequence S t The change in b t The predicted value is calculated by the following formula:

[0106] F t+m =S t +b t m (8)

[0107] Based on the behavior sequence development trend S t and its variation b t Analytical prediction value F t+m Among them, m represents the interval period of the development trend. The value of m will be determined based on expert experience to determine whether it represents the short-term, medium-term or long-term development trend of metacognitive ability.

[0108] The two parameters α and γ in the Holt two-parameter exponential smoothing model play a key role and directly affect the accuracy of the predicted value. In order to improve the flexibility of parameters α and γ, the present invention introduces an adaptive smoothing parameter adjustment mechanism. The adaptive smoothing parameter α′ of the Holt two-parameter exponential smoothing model in different t periods is t and γ′ t The Holt two-parameter exponential smoothing model with adaptive smoothing parameter adjustment mechanism adaptively adjusts the parameter α′ according to the changes of the behavior sequence in different periods. t and γ′ t , thereby improving the accuracy of the meta-prediction value.

[0109]

[0110] Another possible approach is to characterize the key behavior sequence X = {x t ,t=1,2,…,n}, use the Autoregressive Integrated Moving Average Model to perform differential operations, or build a residual autoregressive model to fit the behavior sequence to analyze and explore the development trends of learners' metacognitive ability in the short, medium and long term.

[0111] Step 6: Input the key behavioral representation indicators and key behavioral sequences of metacognitive ability into the prediction model to predict learners' metacognitive ability and the development trend of metacognitive ability.

[0112] The purpose of realizing metacognitive ability modeling based on behavior is to predict the dynamically developing metacognitive ability based on the dynamically changing explicit behavior. The present invention constructs a multi-layer feedforward neural network (HLFF) and realizes intelligent prediction of metacognitive ability by combining static learning behavior indicators and dynamic learning behavior sequences. HLFF consists of an input layer, a hidden layer and an output layer. Each layer contains multiple neurons, such as Figure 4 shown.

[0113] The input layer in HLFF represents the key behavioral representation indicators of metacognitive ability, which specifically includes three types of neurons: behavioral indicator neurons, behavioral sequence neurons, and bias neurons. Among them, the number of behavioral indicator neurons m is the number of key behavioral representation indicators of metacognitive ability, and each behavioral indicator neuron b i =F bi / F bs , where F bi represents the frequency of occurrence of the i-th key behavior, F bs represents the frequency of occurrence of all key behaviors. The number of behavior sequence neurons n is the number of key behavior sequences with significant differences. Each behavior sequence neuron l i =F li / F ls , where F li represents the frequency of occurrence of the i-th key behavior sequence, F ls Indicates the frequency of occurrence of all key behavior sequences. Bias neurons increase the parameters of the neural network by setting the bias term, thereby enhancing the fitting ability of the neural network.

[0114] The hidden layer in HLFF is used to convert key behavior characterization indicators into score values ​​and key behavior sequences into predicted values. The number of neurons determines the prediction effect of the entire neural network. The present invention uses a genetic algorithm to determine the optimal number of neurons in the hidden layer, and trains the constructed HLFF neural network on this basis. The output layer in HLFF is the predicted metacognitive ability level of the learner, that is, the intelligent prediction of metacognitive ability is achieved by combining behavioral indicators and behavioral sequences.

[0115] Based on the same concept, the present invention also provides a metacognitive ability prediction device, including an acquisition module, a clustering module and a conversion module.

[0116] The acquisition module is used to obtain learners' metacognitive ability data and learning behavior data.

[0117] The clustering module is used to perform correlation analysis and clustering on metacognitive ability data and learning behavior data to obtain key behavioral representation indicators of metacognitive ability.

[0118] The conversion module is used to convert key behavioral representation indicators into score values ​​and predict metacognitive ability through the score values.

[0119] The transformation module includes a construction module, a generation module and a calculation module.

[0120] The construction module is used to obtain the correlation coefficients of key behavior characterization indicators and construct a correlation coefficient matrix.

[0121] The generation module is used to obtain multiple eigenvalues ​​and multiple eigenvectors of the correlation coefficient matrix, and generate principal components according to the multiple eigenvectors.

[0122] The calculation module is used to obtain multiple information contribution rates of each eigenvalue to the principal component, multiply and sum the multiple information contribution rates with the corresponding principal component, and obtain a score value.

[0123] The present invention also provides a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the above-mentioned metacognitive ability prediction method is implemented when the processor executes the program.

[0124] The present invention also provides a computer-readable storage medium, which stores a computer program. When the computer program is executed by a processor, the above-mentioned metacognitive ability prediction method is implemented.

[0125] Although the preferred embodiments of the present invention have been described, those skilled in the art may make other changes and modifications to these embodiments once they have learned the basic creative concept. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments and all changes and modifications that fall within the scope of the present invention.

[0126] Obviously, those skilled in the art may make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if these modifications and variations of the present invention fall within the scope of the claims of the present invention and their equivalents, the present invention is also intended to include these modifications and variations.

Claims

1. A method for predicting metacognitive ability, characterized in that: The following steps are involved: Obtain learners' metacognitive ability data and learning behavior data; Conduct correlation analysis and clustering on metacognitive ability data and learning behavior data to obtain key behavioral representation indicators of metacognitive ability; Convert key behavioral representation indicators into score values, and use the score values ​​to predict metacognitive ability; The key behavior characterization indicators are converted into score values, including: Obtain the correlation coefficients of key behavior representation indicators and construct a correlation coefficient matrix; Obtain multiple eigenvalues ​​and multiple eigenvectors of the correlation coefficient matrix, and generate principal components according to the multiple eigenvectors; Obtain multiple information contribution rates of each eigenvalue to the principal component, multiply and sum the multiple information contribution rates with the corresponding principal component to obtain a score value.

2. A method for predicting metacognitive ability according to claim 1, characterized in that: Correlation analysis and clustering of metacognitive ability data and learning behavior data are performed to obtain key behavioral representation indicators of metacognitive ability, including the following steps: Perform correlation analysis on metacognitive ability data and learning behavior data, and obtain the corresponding correlation coefficient; the correlation coefficient is obtained by the following formula: In the formula, r s represents the correlation coefficient, n represents the number of samples of learning behavior data obtained, and d i Indicates the level difference between learning behavior data and metacognitive ability data; Compare the correlation coefficient with the threshold, and use the corresponding learning behavior data with a correlation coefficient greater than the threshold as the relevant behavior representation index of the learner's metacognitive ability; The relevant behavioral representation indicators are clustered using the R-type clustering algorithm to obtain the key behavioral representation indicators of metacognitive ability.

3. A method for predicting metacognitive ability according to claim 1, characterized in that: After converting the key behavioral indicators into score values, it also includes: Mining learning behavior data to obtain learning behavior sequences, and performing difference analysis on learning behavior sequences to obtain key behavior sequences; The key behavior sequences are converted into prediction values, and the development trend of meta-capabilities is characterized by the prediction values.

4. A method for predicting metacognitive ability as claimed in claim 3, characterized in that: Mining the learning behavior data to obtain the learning behavior sequence includes the following steps: Constructing state sequence variables and observation sequence variables based on Markov, wherein the state sequence variables are hidden variables and the observation sequence variables are learning behavior sequences; Based on the dependency relationship between the state sequence variables and the observation value sequence variables, the joint probability between the state sequence variables and the observation value sequence variables is obtained; Based on the joint probability, the state transition probability between the state sequence variable and the observation value sequence variable is obtained; The learning behavior data is mined based on the state transition probability to obtain the learning behavior sequence.

5. A method for predicting metacognitive ability as claimed in claim 3, characterized in that: The difference sequence mining technology DSM and the piecewise linear segmentation algorithm PLS are used to analyze the learning behavior sequence and obtain the key behavior sequence, which specifically includes the following steps: The PLS algorithm is used to divide the learner's learning behavior sequence into the first behavior data set and the second behavior data set; Use the DSM algorithm to mine multiple frequent sequence patterns in the first behavior data set and the second behavior data set respectively; Obtain the frequencies iSupportleft and iSupportright of a specific subsequence appearing in a certain behavior sequence in the first behavior data set and the second behavior data set respectively; For each frequent sequence pattern, a t-test is used to test the average values ​​of iSupportleft and iSupportright of the pattern in the first and second behavior data sets and the size of the set threshold; If the average value is greater than the set threshold, it is considered that the frequent sequence pattern has a difference between the first behavior data set and the second behavior data set, and the key behavior sequence is obtained.

6. A method for predicting metacognitive ability as claimed in claim 3, characterized in that: Transform key behavioral sequences into predictive values, including: Input the key behavior sequence into the Holt two-parameter exponential smoothing model to obtain the predicted value; The Holt two-parameter exponential smoothing model is shown as follows: F t+m =S t +b t m; in, S t =αx t +(1-α)(S t-1 +b t-1 ), b t =γ(S t -S t-1 )+(1-γ)b t-1 ; In the formula, F represents the predicted value, S represents the development trend of the behavior sequence, b represents the change in the development trend of the behavior sequence, m represents the interval number of the development trend, x represents the key behavior sequence, α and γ represent the model parameters, and t represents the period; The parameters of the Holt two-parameter exponential smoothing model are adjusted by adaptive smoothing parameters, as shown below: In the formula, α t ′ and γ t ′ represents the adaptive smoothing parameter.

7. A metacognitive ability prediction device, characterized in that: include: An acquisition module is used to acquire learners’ metacognitive ability data and learning behavior data; Clustering module, used to perform correlation analysis and clustering on metacognitive ability data and learning behavior data, and obtain key behavioral representation indicators of metacognitive ability; The conversion module is used to convert key behavioral representation indicators into score values, and predict metacognitive ability through the score values; The conversion module comprises: A construction module is used to obtain the correlation coefficients of key behavior representation indicators and construct a correlation coefficient matrix; A generation module, used for obtaining multiple eigenvalues ​​and multiple eigenvectors of the correlation coefficient matrix, and generating principal components according to the multiple eigenvectors; The calculation module is used to obtain multiple information contribution rates of each eigenvalue to the principal component, multiply the multiple information contribution rates with the corresponding principal component and sum them to obtain a score value.

8. A computer device, characterized in that: The method comprises a memory, a processor and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, the method for predicting metacognitive ability described in any one of claims 1 to 7 is implemented.

9. A computer-readable storage medium, characterized in that: The storage medium stores a computer program, and when the computer program is executed by a processor, the method for predicting metacognitive ability according to any one of claims 1 to 7 is implemented.

Images