Method and system for predicting students' learning behavior trends based on time series analysis

Abstract

The present invention relates to the field of learning behavior trend prediction, and specifically to a method and system for predicting the learning behavior trend of students based on time series analysis. The method extracts characteristic variables from learning logs and performs data analysis to obtain time entropy values ​​and space entropy values, and divides students into different learning groups through a clustering algorithm; obtains historical student learning behavior pattern information, and divides a learning cycle into multiple learning sub-cycles; labels students with learning behavior pattern labels according to the similarity of the learning behavior pattern information; performs data analysis on the predicted evaluation value of the student in the current learning sub-cycle and the reference evaluation value corresponding to the learning group to which the student belongs to obtain a learning effect coefficient, and when the learning effect coefficient is less than a preset effect threshold, obtains the student's learning behavior pattern label and adjusts it; solves how to more accurately analyze the student's learning effect and learning behavior, and then predicts the student's learning behavior trend, so as to provide personalized and accurate intervention measures.

Description

Technical Field

[0001] The present invention relates to the field of learning behavior trend prediction, and in particular to a method and system for predicting a learner's learning behavior trend based on time series analysis. Background Art

[0002] With the rapid development of the online education industry and the continuous growth of the number of people participating in online learning, students' learning behaviors often show extremely complex and diverse characteristics. In the analysis of students' learning behaviors, traditional methods often only focus on single-dimensional behavioral data, such as students' learning time, homework completion, etc. These methods ignore the multi-dimensional characteristics of students' learning behaviors and their dynamic characteristics that change over time and space. Students' learning time and learning location are not only affected by their individual differences, but also by the interaction of multiple factors such as the external environment and learning tasks. Therefore, the analysis of students' learning behavior data has become an important direction in the study of student behavior prediction.

[0003] However, how to efficiently analyze students' learning behaviors and predict their learning trends in order to provide personalized and precise intervention measures has become an urgent problem to be solved in the field of educational technology. Although some existing studies have achieved the prediction of students' learning behaviors to a certain extent, most of them are difficult to capture the complex interactions of students' learning behaviors in time and space dimensions; therefore, combined with advanced time series analysis methods, a student learning behavior trend prediction method based on time series analysis is proposed, which can more accurately capture students' learning behaviors and provide personalized and precise learning support services for online education. Summary of the invention

[0004] (1) Technical issues to be resolved

[0005] The purpose of the present invention is to provide a method and system for predicting student learning behavior trends based on time series analysis, so as to solve the problem of how to more accurately capture students' learning behaviors and learning effects by analyzing students' learning behaviors and learning effects, thereby adjusting students' learning behaviors in a targeted manner.

[0006] (2) Technical solution

[0007] To achieve the above object, on the one hand, the present invention provides a method for predicting the learning behavior trend of students based on time series analysis, the method comprising:

[0008] S1. Obtain the learning log of historical students, clean the learning log and extract characteristic variables, the characteristic variables include learning behavior, timestamp, and location information; perform data analysis based on the characteristic variables to obtain the time entropy value and space entropy value of historical students and record them as historical spatiotemporal entropy values, and divide the students into different learning groups through a clustering algorithm based on the historical spatiotemporal entropy values.

[0009] S2. Convert the historical students' learning behaviors into time series data and extract key features through autoencoders to obtain learning behavior pattern information, obtain the turning point of the learning behavior pattern through the time series turning point recognition algorithm, and divide the learning cycle into multiple learning sub-cycles according to the turning point; analyze the similarity of the students' learning behavior pattern information through the time series clustering method, classify the students in each learning sub-cycle according to the similarity of the learning behavior pattern information and mark the learning behavior pattern labels.

[0010] S3. Obtain the student's historical homework data and extract homework features, and obtain evaluation values ​​through data analysis of the homework features; analyze and predict the evaluation values ​​based on the time series model to obtain the student's predicted evaluation value in the current learning sub-cycle.

[0011] S4. Obtain reference evaluation values ​​of different learning groups in the current learning sub-cycle through statistical analysis methods; obtain the spatiotemporal entropy value of the student and identify the learning group to which the student belongs based on its spatiotemporal entropy value, perform data analysis on the predicted evaluation value and the reference evaluation value corresponding to the learning group to which the student belongs to obtain the student's learning effect coefficient, and when the learning effect coefficient is less than a preset effect threshold, obtain the student's learning behavior pattern label and adjust it.

[0012] Furthermore, the method of obtaining the historical homework data of the trainee and extracting the homework features, obtaining the evaluation value through data analysis of the homework features; and obtaining the predicted evaluation value of the trainee in the current learning sub-period by analyzing and predicting the evaluation value based on the time series model includes:

[0013] Obtain job evaluation standard image data as a standard data set, perform data preprocessing on the standard data set to obtain a first data set, perform denoising on the first data set using a probabilistic statistical method to obtain a second data set, and perform classification and cutting on the second data set using a cutting algorithm to obtain a third data set.

[0014] The third data set is subjected to feature extraction and then classified by SVM classifier to obtain standard features; the standard features are linked to standard evaluation scores by Delphi method to obtain feature-score relationship distribution; the student's homework features and standard features are subjected to feature matching analysis method to obtain feature similarity; when the feature similarity is greater than a preset feature similarity threshold, the standard features to which the student's homework features belong are determined and combined with the feature-score relationship distribution to obtain the student's evaluation score, and a time series analysis is performed based on the student's evaluation score to obtain a trend of evaluation score changes, and a predicted score is obtained based on the trend of evaluation score changes.

[0015] The fluctuation coefficient is obtained according to the changing trend of the assessment score, and the fluctuation coefficient is used to judge the stability of the student's assessment score; the predicted score is corrected according to the fluctuation coefficient to obtain a corrected assessment score and recorded as the predicted evaluation value.

[0016] Furthermore, the method of obtaining a fluctuation coefficient according to the evaluation score change trend, and the fluctuation coefficient is used to judge the stability of the student's evaluation score; and correcting the predicted score according to the fluctuation coefficient to obtain a corrected evaluation score and recording it as the predicted evaluation value includes:

[0017] Get the evaluation score sequence of the students in the current learning sub-cycle as ,in Indicates The evaluation score of the evaluation, is the total number of evaluations; the changing trend of the evaluation scores is obtained based on the evaluation score sequence.

[0018] The coefficient of fluctuation is the standard deviation of the evaluation score The average of the evaluation scores The ratio of; the mean of the evaluation scores The calculation formula is: ; The standard deviation of the evaluation score The calculation formula is: The fluctuation coefficient The calculation formula is:

[0019] .

[0020] The predicted evaluation score of the students obtained by the time series model is ; When the fluctuation coefficient If the fluctuation coefficient is less than the preset safety fluctuation coefficient threshold, Make corrections to get the corrected evaluation score ; The modified evaluation score The calculation formula is:

[0021] .

[0022] in, It is the correction coefficient, which indicates the influence of the volatility coefficient on the correction score and is obtained through statistical analysis of historical data.

[0023] When the fluctuation coefficient When it is greater than the preset safety fluctuation coefficient threshold, the student's learning behavior pattern label is obtained and adjusted.

[0024] Further, the method comprises extracting features from the third data set and classifying it using an SVM classifier to obtain standard features; establishing a connection between the standard features and the standard evaluation scores using the Delphi method to obtain a feature-score relationship distribution; and using a feature matching analysis method to obtain feature similarity between the student's homework features and the standard features;

[0025] The global features and local features are extracted from the third data, and the dimension reduction is performed through a dimension reduction algorithm to obtain reduced-dimensionality global features and reduced-dimensionality local features; after feature fusion of the reduced-dimensionality global features and reduced-dimensionality local features, feature training is performed through an SVM classifier and classification is performed to obtain standard features.

[0026] Before extracting features from the student's historical homework data and the third data set, low-frequency components and high-frequency components are obtained through contourlet transform; the student's homework features and the low-frequency components of the standard features are similarly calculated using the minimum Euclidean distance criterion to obtain a first similarity; when the first similarity is greater than a preset first similarity threshold, the high-frequency components of the student's homework features and the standard features are decomposed to select high-frequency feature points greater than the preset high-frequency threshold, and the high-frequency feature points are matched with the feature point matching algorithm to obtain a second similarity and recorded as feature similarity.

[0027] Furthermore, the method of performing data analysis according to the characteristic variables to obtain the time entropy value and space entropy value of the historical students and recording them as the historical time-space entropy value includes:

[0028] The time entropy value It is used to indicate the concentration of students during the learning period. The formula is:

[0029] .

[0030] in, represents the number of study periods, Indicated in The learning probability of a learning period; Indicated in The number of learning sessions in each learning session; Indicates the total number of learning times in all learning periods.

[0031] The spatial entropy value It is used to indicate the distribution of students in the learning position. The spatial entropy value The formula is:

[0032] .

[0033] in, represents the number of learned position categories, Indicated in The learning probability of a learning position; Indicated in The number of learning positions; Represents the total number of learnings at all learning positions.

[0034] Based on the same inventive concept, the present invention also provides a student learning behavior trend prediction system based on time series analysis, the system comprising: a learning group division module, a learning cycle management module, a prediction evaluation value calculation module, a learning effect analysis and behavior management module, and each module is sequentially connected in communication;

[0035] The learning group division module is used to obtain the learning logs of historical students, extract characteristic variables after data cleaning of the learning logs, and the characteristic variables include learning behavior, timestamp, and location information; perform data analysis based on the characteristic variables to obtain the time entropy value and space entropy value of the historical students and record them as historical spatiotemporal entropy values, and divide the students into different learning groups based on the historical spatiotemporal entropy values ​​through a clustering algorithm.

[0036] The learning cycle management module is used to convert the historical learning behavior of students into time series data and extract key features through the autoencoder to obtain learning behavior pattern information, obtain the turning point of the learning behavior pattern through the time series turning point recognition algorithm, and divide the learning cycle into multiple learning sub-cycles according to the turning point; analyze the similarity of the students' learning behavior pattern information through the time series clustering method, classify the students in each learning sub-cycle according to the similarity of the learning behavior pattern information, and mark the learning behavior pattern labels.

[0037] The predicted evaluation value calculation module is used to obtain the student's historical homework data and extract homework features, and obtain the evaluation value through data analysis of the homework features; the evaluation value is analyzed and predicted based on the time series model to obtain the student's predicted evaluation value in the current learning sub-cycle.

[0038] The learning effect analysis and behavior management module is used to obtain the reference evaluation values ​​of different learning groups in the current learning sub-cycle through statistical analysis methods; obtain the spatiotemporal entropy value of the students and identify the learning group to which the students belong based on their spatiotemporal entropy values, and perform data analysis on the predicted evaluation value and the reference evaluation value corresponding to the learning group to which the students belong to obtain the learning effect coefficient of the students. When the learning effect coefficient is less than the preset effect threshold, the learning behavior pattern label of the students is obtained and adjusted.

[0039] Furthermore, the prediction evaluation value calculation module includes:

[0040] The data processing module is used to obtain the job evaluation standard image data as the standard data set, perform data preprocessing on the standard data set to obtain the first data set, perform denoising on the first data set by a probability statistics method to obtain the second data set, and perform classification and cutting on the second data set by a cutting algorithm to obtain the third data set.

[0041] The prediction score analysis module is used to extract features from the third data set and then classify it using an SVM classifier to obtain standard features; establish a connection between the standard features and the standard evaluation scores through the Delphi method to obtain a feature-score relationship distribution; use a feature matching analysis method to obtain feature similarity between the student's homework features and the standard features; when the feature similarity is greater than a preset feature similarity threshold, determine the standard features to which the student's homework features belong and combine them with the feature-score relationship distribution to obtain the student's evaluation score, use time series analysis based on the student's evaluation score to obtain a trend of evaluation score changes, and obtain a predicted score based on the trend of evaluation score changes.

[0042] The predicted evaluation value analysis module is used to obtain a fluctuation coefficient according to the evaluation score change trend, and the fluctuation coefficient is used to judge the stability of the student's evaluation score; the predicted score is corrected according to the fluctuation coefficient to obtain a corrected evaluation score and recorded as the predicted evaluation value.

[0043] Furthermore, the prediction evaluation value analysis module includes the following steps:

[0044] Get the evaluation score sequence of the students in the current learning sub-cycle as ,in Indicates The evaluation score of the evaluation, is the total number of evaluations; the changing trend of the evaluation scores is obtained based on the evaluation score sequence.

[0045] The coefficient of fluctuation is the standard deviation of the evaluation score The average of the evaluation scores The ratio of; the mean of the evaluation scores The calculation formula is: ; The standard deviation of the evaluation score The calculation formula is: The fluctuation coefficient The calculation formula is:

[0046] .

[0047] The predicted evaluation score of the students obtained by the time series model is ; When the fluctuation coefficient If the fluctuation coefficient is less than the preset safety fluctuation coefficient threshold, Make corrections to get the corrected evaluation score ; The modified evaluation score The calculation formula is:

[0048] .

[0049] in, It is the correction coefficient, which indicates the influence of the volatility coefficient on the correction score and is obtained through statistical analysis of historical data.

[0050] When the fluctuation coefficient When it is greater than the preset safety fluctuation coefficient threshold, the student's learning behavior pattern label is obtained and adjusted.

[0051] Furthermore, the prediction score analysis module includes:

[0052] The standard feature management module is used to extract global features and local features from the third data, and obtain reduced-dimensionality global features and reduced-dimensionality local features after dimensionality reduction through a dimensionality reduction algorithm; after feature fusion of the reduced-dimensionality global features and the reduced-dimensionality local features, feature training and classification are performed through an SVM classifier to obtain standard features.

[0053] The feature similarity calculation module is used to obtain low-frequency components and high-frequency components through contour wave transformation before extracting features from the student's historical homework data and the third data set; the student's homework features and the low-frequency components of the standard features are similarly calculated through the minimum Euclidean distance criterion to obtain a first similarity; when the first similarity is greater than a preset first similarity threshold, the high-frequency components of the student's homework features and the standard features are decomposed to select high-frequency feature points greater than the preset high-frequency threshold, and the high-frequency feature points are matched with the feature point matching algorithm to obtain a second similarity and recorded as feature similarity.

[0054] Furthermore, the learning group division module includes the following steps:

[0055] The time entropy value It is used to indicate the concentration of students during the learning period. The formula is:

[0056] .

[0057] in, represents the number of study periods, Indicated in The learning probability of a learning period; Indicated in The number of learning sessions in each learning session; Indicates the total number of learning times in all learning periods.

[0058] The spatial entropy value It is used to indicate the distribution of students in the learning position. The spatial entropy value The formula is:

[0059] .

[0060] in, represents the number of learned position categories, Indicated in The learning probability of a learning position; Indicated in The number of learning positions; Represents the total number of learnings at all learning positions.

[0061] (3) Beneficial effects

[0062] Compared with the prior art, the present invention has the following beneficial effects:

[0063] 1. By obtaining students' historical homework data and learning logs, combined with multiple factors such as spatiotemporal entropy, learning behavior patterns, and evaluation score change trends, we can deeply explore students' learning patterns and potential trends. We can more accurately analyze students' learning status and future performance, and provide personalized intervention based on students' learning behavior.

[0064] 2. Through more accurate feature analysis of students' homework characteristics and stability analysis of assessment scores, it is possible to more accurately predict students' assessment scores and learning outcomes in future learning sub-cycles. BRIEF DESCRIPTION OF THE DRAWINGS

[0065] Figure 1 This is a flowchart of a method for predicting student learning behavior trends based on time series analysis according to Example 1 of the present invention.

[0066] Figure 2 This is a schematic diagram of the module composition of the student learning behavior trend prediction system based on time series analysis in Example 2 of the present invention. DETAILED DESCRIPTION

[0067] 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.

[0068] Before giving examples, it is necessary to explain the application scenarios of the present invention. The present invention is a method and system for predicting the learning behavior trend of students based on time series analysis. In today's online education, many students' learning behaviors are very complicated. For example, in calligraphy online education, due to their own living environment, working environment, etc., students' learning behaviors are diverse, and there is no more accurate scoring method for calligraphy evaluation. As a result, the learning behaviors of these students cannot be effectively adjusted, and they can only check the learning effect based on the usual evaluation results, which also causes the students' learning behavior to not know how to make effective adaptive adjustments. Once the learning effect is very poor and cannot be effective, most students will choose to give up. Therefore, it is necessary to analyze the learning behavior of students, and provide an accurate evaluation method for calligraphy homework to obtain evaluation results, conduct a full correlation analysis between the evaluation results and the learning behavior, and analyze different behavior patterns of historical students' learning behaviors, so that it can timely monitor whether the learning effect of students under the current learning behavior is good. If it is not good enough, the students can be advised or adjusted according to the better learning behavior pattern obtained by the analysis, so that the students can adjust their own learning behavior in a targeted manner.

[0069] Example 1: Figure 1 As shown, this embodiment provides a method for predicting student learning behavior trends based on time series analysis, the method comprising:

[0070] S1. Obtain the learning log of historical students, clean the learning log and extract characteristic variables, the characteristic variables include learning behavior, timestamp, and location information; perform data analysis based on the characteristic variables to obtain the time entropy and space entropy of historical students and record them as historical time and space entropy values, and divide students into different learning groups through clustering algorithms based on the historical time and space entropy values; data cleaning includes removing invalid data, such as duplicate records, missing value filling, etc.; format standardization, converting timestamps and location information of different formats into a standard format; removing noise, such as removing abnormal or erroneous records. Learning behavior refers to students' behaviors such as watching videos, taking tests, and submitting homework. The timestamp records the time when each learning activity occurs. The location information is the geographical location of the student when studying. These characteristic variables will reflect key information such as students' learning habits, learning time periods, and learning environment. The time entropy value reflects the distribution law of students' learning behavior in the time dimension. For example, if the student's learning time is concentrated in certain specific time periods, the time entropy value is low, indicating that their learning behavior is more regular; if the learning time is more dispersed, the time entropy value is high. The space entropy value reflects the distribution of students' learning activities in the space dimension. If the student's learning behavior occurs in multiple different places (for example, home, cafe, library, etc.), the spatial entropy value is high; if the student's learning activities are mainly concentrated in a certain place, the spatial entropy value is low. These entropy values ​​are combined to obtain the historical spatiotemporal entropy value, which is used to describe the overall learning behavior pattern of the student. The historical spatiotemporal entropy value applies clustering algorithms (such as K-means, DBSCAN, etc.) to divide the students into different learning groups. The goal of clustering is to identify groups with similar learning behaviors based on the spatiotemporal entropy values ​​of the students. Different learning groups may show different learning characteristics. For example, high temporal entropy and low spatial entropy indicate that the student's learning time is irregular, but the learning location is relatively concentrated (for example, online learning at night, not learning at work during the day). Low temporal entropy and high spatial entropy indicate that the student learns in different time periods and the learning locations are scattered (for example, studying in different places at fragmented time). Low temporal entropy and low spatial entropy indicate that the student's learning time and location are very fixed (for example, studying at home regularly). For example, there are two learning logs of students A and B. Among them, student A studies at home from 8 to 10 pm every day, which is almost unchanged. The time entropy value is low, the learning time is regular, the space entropy value is low, and the learning location is mainly at home. The historical space-time entropy value is low, which belongs to a regular learning group. Student B studies at home, office and library from 7 to 8 in the morning, 1 to 2 in the afternoon, and 10 to 11 in the evening. The learning method is more diverse, the time entropy value is high, the learning time is irregular, the space entropy value is high, the learning location is scattered, and the historical space-time entropy value is high. It belongs to an irregular but flexible learning group.

[0071] S2. Convert the learning behavior of historical students into time series data and extract key features through the autoencoder to obtain learning behavior pattern information, obtain the turning point of the learning behavior pattern through the time series turning point recognition algorithm, and divide the learning cycle into multiple learning sub-cycles according to the turning point; analyze the similarity of the learning behavior pattern information of the students through the time series clustering method, classify the students in each learning sub-cycle according to the similarity of the learning behavior pattern information and mark the learning behavior pattern label; arrange the learning behavior of the students in chronological order according to the timestamp. The structure of the autoencoder includes an encoder (compressing high-dimensional data into low-dimensional features) and a decoder (reconstructing the original data from low-dimensional features). The time series learning behavior data of the students is input, and the output is a low-dimensional feature representation extracted by the autoencoder, which can capture the key information of the learning behavior pattern of the students. Example: After the learning behavior of student A is extracted by the autoencoder, a low-dimensional representation may be obtained: low-dimensional feature A: learning time, diversity of learning content, regularity of learning, etc. Through the time series turning point recognition algorithm, for example, a method based on change point detection can be used to identify the turning point in the learning behavior pattern of the students. The turning point is the time point when the student's learning behavior changes significantly, which usually means that the student's learning mode has changed. In this embodiment, by identifying the turning week of each student, the number of people who have a mode turning week can be calculated, and the turning week with the largest number of people identified is recorded as the final turning point. Each learning sub-cycle represents a specific learning mode, and the behavior of the students in each sub-cycle is relatively consistent. The similarity of the behavior patterns of students in different learning sub-cycles is analyzed by time series clustering methods. Commonly used time series clustering methods include K-means and DBSCAN methods. By clustering the learning behavior pattern information, it is determined which students have similar learning behaviors, and then each student will be assigned a specific learning behavior pattern label in each learning sub-cycle to represent the characteristics of its behavior. For example, assuming that the behavior patterns of students A and B in the same learning cycle are clustered and analyzed, the results are as follows: Student A's learning behavior has a strong time regularity and tends to concentrate on learning. Student B's learning behavior is relatively scattered, and the distribution of learning time is irregular. Through cluster analysis, students A and B may be divided into different learning mode labels. According to the clustering results and the students' learning behavior information, the students are classified in each learning sub-cycle, and each student is labeled with a learning behavior pattern. This label reflects the characteristics of the student's learning behavior in that sub-cycle. For example, "efficient and concentrated type", "scattered and random type", etc., help understand the students' learning habits and patterns.

[0072] S3. Obtain the historical homework data of the students and extract the homework features. Obtain the evaluation value through data analysis of the said homework features; Analyze and predict the said evaluation value based on a time series model to obtain the predicted evaluation value of the students in the current learning sub-cycle; The homework in this implementation is calligraphy homework, and the key features are extracted from the calligraphy homework, including: stroke quality, such as the fluency, uniformity, and strength control of the strokes. Structural symmetry, whether the overall structure and layout of the font are coordinated. Similarity of copying, the degree of similarity between the students' calligraphy and the model copybook. For example, for a certain calligraphy homework of a student: Model copybook, the regular script character 'yong'. Stroke analysis: Horizontal strokes are straight and vertical strokes are perpendicular, and the fluency of turning is 80%. Overall structure: The font is slightly asymmetrical. Extracted features: Stroke fluency: 80%, Structural symmetry: 75%, Similarity of copying: 85%. The evaluation value of calligraphy practice can show trends and periodicity over time. Commonly used time series models include: ARIMA is applicable to the linear change of the students' evaluation value. LSTM: Suitable for capturing complex learning trends. Assume the evaluation values of student A are as follows: Week 1: 75, Week 2: 78, Week 3: 82, Week 4: 80. These evaluation values form a time series, and the evaluation value of the current learning sub-cycle can be predicted through the model. The time series model predicts the evaluation value of the current cycle by fitting historical data. For example, based on the prediction of the LSTM model, the evaluation value of student A in Week 5 may be 85.

[0073] S4. Obtain reference evaluation values ​​of different learning groups in the current learning sub-cycle by statistical analysis methods; obtain the spatiotemporal entropy value of the student and identify the learning group to which the student belongs according to its spatiotemporal entropy value, perform data analysis on the predicted evaluation value and the reference evaluation value corresponding to the learning group to which the student belongs to obtain the learning effect coefficient of the student, and when the learning effect coefficient is less than the preset effect threshold, obtain the learning behavior pattern label of the student and adjust it. The reference evaluation value is a standard value obtained from the learning group to which the student belongs, representing the average performance of the group in the current learning sub-cycle, and is obtained by statistical analysis of the historical data of all students in the group. For example: the reference evaluation value of group A: 85, the reference evaluation value of group B: 78. The learning effect coefficient is the ratio between the actual learning effect of the student and the reference evaluation value of the group. If the learning effect coefficient is less than the preset effect threshold (such as 0.8), it means that the learning effect of the student is poor and needs to be adjusted. For example, the group to which student C belongs is group A, whose predicted evaluation value is 90, while the reference evaluation value of group A is 85. The learning effect coefficient is 1.06. Since the learning effect coefficient of student C is greater than 1 (assuming the threshold is 0.8), it means that his learning effect is good and no adjustment is needed. Assume that the group to which student D belongs is group B, whose predicted evaluation value is 70, while the reference evaluation value of group B is 78, and the learning effect coefficient is 0.89. At this time, the learning effect coefficient of student D is still greater than 0.8, but lower than 1, indicating that attention and adjustment are needed. If the learning effect coefficient of student D is lower than the threshold of 0.8, it means that his learning effect is poor and needs to be adjusted, such as adjusting the learning plan, increasing the learning time or changing the learning strategy. According to the spatiotemporal entropy value and learning effect coefficient of the students, the learning behavior pattern label can be marked for the students. For example, it includes: efficient learners: continue to maintain the current learning behavior. Inefficient learners: need to adjust their learning behavior, and can make guiding adjustments based on the learning patterns of other learning groups, such as fixing the learning time and place.

[0074] Furthermore, the method of obtaining the historical homework data of the trainee and extracting the homework features, obtaining the evaluation value through data analysis of the homework features; and obtaining the predicted evaluation value of the trainee in the current learning sub-period by analyzing and predicting the evaluation value based on the time series model includes:

[0075] Obtain the standard image data of the homework evaluation criteria as the standard data set, perform data preprocessing on the standard data set to obtain the first data set, perform denoising processing on the first data set through a probability statistical method to obtain the second data set, and perform classification cutting on the second data set through a cutting algorithm to obtain the third data set; in calligraphy evaluation, the standard data set usually includes a large number of calligraphy work images as the standard data for comparison, and the standard data is different in different learning sub-cycles. Each image represents a calligraphy assignment and contains important information such as the structure, strokes, and glyphs of the calligraphy work. These image data include images of standard calligraphy fonts or standard model calligraphy copies (for example, standard regular script, running script, etc.). Data preprocessing includes: Image size standardization is to uniformly adjust all images to the same size and resolution to ensure the consistency of subsequent analysis. Grayscale conversion is to convert a color image to a grayscale image if it is a color image to reduce unnecessary complexity and facilitate the extraction of information such as glyphs and strokes. Removing background noise is to remove background noise or irrelevant elements in the image to ensure that only the main part of the calligraphy work is retained. Denoising processing aims to eliminate noise or unnecessary details in the image and retain the main calligraphy features. For example, in the image of a student's calligraphy work, there are some small stains or incomplete strokes. After denoising processing, the redundant noise in the image is removed, and the glyphs are clearer and more complete. The cutting algorithm can divide the area of the calligraphy work image to help extract the area of each calligraphy glyph or stroke, identify each character in the image through text cutting, and cut it into independent image blocks separately. Commonly used cutting algorithms include edge detection algorithms: using methods such as Canny edge detection to identify the glyph edges in the image for cutting. In this embodiment, the minimum bounding box algorithm is used for text cutting. For example, after being processed by the cutting algorithm, the image originally containing the entire calligraphy work is divided into several glyph images. For example, the calligraphy work contains the character '永', and this character is cut into an independent image block for analyzing the quality of each character separately.

[0076] The third data set is subjected to feature extraction and then classified by SVM classifier to obtain standard features; the standard features are linked to the standard evaluation scores by Delphi method to obtain feature-score relationship distribution; the student's homework features and standard features are subjected to feature matching analysis method to obtain feature similarity; when the feature similarity is greater than the preset feature similarity threshold, the standard features to which the student's homework features belong are determined and combined with the feature-score relationship distribution to obtain the student's evaluation score, and the evaluation score change trend is obtained by time series analysis according to the student's evaluation score, and the predicted score is obtained according to the evaluation score change trend; the feature vector of each calligraphy work is obtained by feature extraction of glyph structure, stroke quality, stroke fluency, etc. These features are classified by SVM classifier (support vector machine) to obtain standard features of different calligraphy works, that is, categories similar to the features of standard copybooks. For example: standard feature A: regular glyph structure and smooth strokes. Standard feature B: asymmetrical glyphs with breaks between strokes. For example: Assume that a student submits a calligraphy homework and obtains a set of feature vectors after feature extraction. Based on these features, the SVM classifier compares the student's work with the standard calligraphy features and concludes that the student's work features belong to standard feature A, that is, the characters are regular and the strokes are smooth. The Delphi method is a prediction method based on expert opinions, which is often used to obtain group judgments on complex issues. In calligraphy evaluation, the Delphi method can be used to associate each standard feature with a standard evaluation score. For example, the evaluation score of standard feature A (regular characters and smooth strokes) is 90 points. The evaluation score of standard feature B (asymmetric characters and broken strokes) is 70 points. Through the application of the Delphi method, the feature-score relationship distribution is finally obtained, that is, the relationship between the standard features and their corresponding evaluation scores. For example: With the help of the Delphi method, experts determined that the calligraphy work with "regular characters and smooth strokes" is scored as 90 points, while the calligraphy work with "asymmetric characters and broken strokes" is scored as 70 points. These evaluation scores will be used as standard references for the subsequent evaluation of the scores of students' works. After feature extraction, the student's homework obtains a set of feature vectors. Then, the similarity is calculated using the feature matching analysis method, and the similarity is used to determine which standard feature the student's homework belongs to. For example, suppose the similarity between the student's homework feature and standard feature A is 0.95, and the similarity with standard feature B is 0.60. Since the similarity is greater than the preset threshold (such as 0.8), it can be determined that the student's homework feature belongs to standard feature A. Once it is determined that the feature similarity of the student's homework is greater than the preset threshold, the student's homework score can be calculated based on the evaluation score of the standard feature through the feature-score relationship distribution. For example: Assume that the student's homework feature belongs to standard feature A, and according to the feature-score relationship distribution, the evaluation score of standard feature A is 90 points, so the student's homework evaluation score is 90 points.The student's assessment score can change over time, so time series analysis (such as ARIMA, LSTM, etc.) can be used to predict the trend of student score changes. Time series models can help analyze the student's learning progress and determine whether it is improving. For example, the student's historical homework scores are: 85 points for the first homework, 87 points for the second homework, and 90 points for the third homework. Through time series analysis, it can be found that the student's score has a steady upward trend, and the next homework score is predicted to be 92 points, indicating that the student's calligraphy level is continuously improving.

[0077] The fluctuation coefficient is obtained according to the changing trend of the assessment score, and the fluctuation coefficient is used to judge the stability of the student's assessment score; the predicted score is corrected according to the fluctuation coefficient to obtain a corrected assessment score and recorded as the predicted evaluation value.

[0078] Furthermore, the method of obtaining a fluctuation coefficient according to the evaluation score change trend, and the fluctuation coefficient is used to judge the stability of the student's evaluation score; and correcting the predicted score according to the fluctuation coefficient to obtain a corrected evaluation score and recording it as the predicted evaluation value includes:

[0079] Get the evaluation score sequence of the students in the current learning sub-cycle as ,in Indicates The evaluation score of the evaluation, is the total number of evaluations; the changing trend of the evaluation scores is obtained based on the evaluation score sequence.

[0080] The coefficient of fluctuation is the standard deviation of the evaluation score The average of the evaluation scores The ratio of; the mean of the evaluation scores The calculation formula is: ; The standard deviation of the evaluation score The calculation formula is: The fluctuation coefficient The calculation formula is:

[0081] .

[0082] The predicted evaluation score of the students obtained by the time series model is ; When the fluctuation coefficient If the fluctuation coefficient is less than the preset safety fluctuation coefficient threshold, Make corrections to get the corrected evaluation score ; The modified evaluation score The calculation formula is:

[0083] .

[0084] in, It is the correction coefficient, which indicates the influence of the volatility coefficient on the correction score and is obtained through statistical analysis of historical data.

[0085] When the fluctuation coefficient When it is greater than the preset safety fluctuation coefficient threshold, the student's learning behavior pattern label is obtained and adjusted. For example, the evaluation score of a student in the current learning sub-cycle is: , total number of evaluations Calculate the mean of the evaluation scores , calculate the standard deviation of the evaluation score , calculate the coefficient of volatility Assuming the preset safety fluctuation coefficient threshold is 0.05, 0.016<0.05, it means that the student evaluation score is relatively stable. Assuming the predicted evaluation score is , correction factor ,but = 90.72, that is, the revised forecast score is 90.72. If the volatility coefficient If it exceeds 0.05, it means that students are often in an inefficient learning state, that is, high time entropy and low space entropy, or low time entropy and high space entropy, or even high time entropy and high space entropy. Therefore, it is necessary to obtain the learning behavior pattern labels of students, adjust their learning behaviors, and make guiding adjustments according to the learning patterns of other learning groups, such as fixing the learning time or place.

[0086] Further, the method comprises extracting features from the third data set and classifying it using an SVM classifier to obtain standard features; establishing a connection between the standard features and the standard evaluation scores using the Delphi method to obtain a feature-score relationship distribution; and using a feature matching analysis method to obtain feature similarity between the student's homework features and the standard features;

[0087] Extract global features and local features from the third data, and reduce the dimension by a dimensionality reduction algorithm to obtain reduced dimension global features and reduced dimension local features; after feature fusion of the reduced dimension global features and reduced dimension local features, feature training and classification are performed by an SVM classifier to obtain standard features; global features are macro features that describe the entire image or operation, and are usually used to capture overall trends or characteristics. For example, the overall layout, stroke thickness, and stroke direction of a calligraphy work. In a calligraphy image, global features include: edge density of the image, edge distribution of the strokes; structural texture, and overall structure of the font. Local features are information that describes certain details in the image, and are usually used to capture subtle changes. For calligraphy works, local features can refer to specific details of each individual stroke or each character. For example, local features in a calligraphy image may include the starting and ending shapes of a single stroke, the stroke order and curvature of each character, and the pen tip changes of a single stroke (such as gradually thickening and thinning, etc.). Since image data usually contains high-dimensional features (especially image pixel data), dimensionality reduction algorithms can help reduce data dimensions, remove redundant features, and retain important information in the data. Common dimensionality reduction algorithms include principal component analysis (PCA) and linear discriminant analysis (LDA). Through dimensionality reduction, the information in the original feature space will be compressed to a lower dimension, but the main structural features of the data will not be lost. The global features and local features after dimensionality reduction are fused to form a comprehensive feature vector. The purpose of feature fusion is to improve the classification effect by combining global and local information. The feature fusion adopts a serial fusion method to obtain feature spaces A and B after dimensionality reduction of global features and local features, respectively. Denote the eigenvector of feature space A, The eigenvector of the feature space B is recorded as: ∈A, ∈B, then the serial fusion method obtains the fused feature vector ,Right now: ,in, The characteristic is x-dimensional, The feature is y-dimensional, then after serial fusion The feature of is (x+y)-dimensional. In this implementation, the SVM classifier will learn how to classify based on the student's calligraphy features (fused global and local features) to generate standard features for each student. After the SVM training is completed, the trained model can be used to classify the student's new homework to obtain the corresponding standard features.

[0088] Before extracting features from the student's historical homework data and the third data set, the low-frequency component and the high-frequency component are obtained by contourlet transform; the student's homework features and the low-frequency components of the standard features are similarly calculated by the minimum Euclidean distance criterion to obtain a first similarity. When the first similarity is greater than the preset first similarity threshold, the high-frequency components of the student's homework features and the standard features are decomposed to select high-frequency feature points greater than the preset high-frequency threshold, and the high-frequency feature points are matched by the feature point matching algorithm to obtain a second similarity and record it as feature similarity. Contourlet Transform is a multi-scale and multi-directional image transformation method that can more effectively capture edges and details in an image. Contourlet transform decomposes an image into two parts: the low-frequency component contains the basic information of the image, such as the overall layout, the thickness of the strokes, etc. The high-frequency component contains detailed information, such as changes in the strokes, subtle fluctuations in the ink, etc. In this way, various features in the image can be extracted more finely. The low-frequency components of the student's homework and the low-frequency components of the standard features are compared, and the minimum Euclidean distance criterion is used to calculate the similarity between them. Euclidean distance calculation formula: ;in, and Represent the low-frequency components of student assignments and standard features, and Represents the feature value of each dimension. If the first similarity is greater than the preset first similarity threshold, it means that the student's homework meets the standard features in terms of low-frequency features, and the high-frequency components are further analyzed. By decomposing the high-frequency components, the high-frequency feature points therein are selected, that is, the detail feature points in the image. The high-frequency feature points of the student's homework and the standard features are matched through a feature point matching algorithm (such as SIFT, SURF, etc.), and the similarities between the student's homework and the standard features are obtained based on information such as the position, direction and scale of the feature points. The calculated second similarity represents the similarity between the high-frequency features of the student's homework and the standard features.

[0089] Furthermore, the method of performing data analysis according to the characteristic variables to obtain the time entropy value and space entropy value of the historical students and recording them as the historical time-space entropy value includes:

[0090] The time entropy value It is used to indicate the concentration of students during the learning period. The formula is:

[0091] ; The larger it is, the more scattered the students' learning activities in different time periods are, and the more uncertain or irregular their learning behaviors are; conversely, the smaller it is, the more concentrated the students' learning time is.

[0092] in, represents the number of study periods, Indicated in The learning probability of a learning period; Indicated in The number of learning sessions in each learning session; Indicates the total number of learning times in all learning periods.

[0093] The spatial entropy value It is used to indicate the distribution of students in the learning position. The spatial entropy value The formula is:

[0094] ; The larger it is, the more dispersed the students’ learning behaviors are in different locations and the more uncertain their learning activities are. On the contrary, the smaller it is, the more concentrated the students’ learning activities are in a few locations.

[0095] in, represents the number of learned position categories, Indicated in The learning probability of a learning position; Indicated in The number of learning positions; Represents the total number of learnings at all learning positions.

[0096] Embodiment 2: Based on the same inventive concept, this embodiment also provides a student learning behavior trend prediction system based on time series analysis, the system comprising: a learning group division module, a learning cycle management module, a prediction evaluation value calculation module, a learning effect analysis and behavior management module, and each module is sequentially connected in communication;

[0097] The learning group division module is used to obtain the learning logs of historical students, extract characteristic variables after data cleaning of the learning logs, and the characteristic variables include learning behavior, timestamp, and location information; perform data analysis based on the characteristic variables to obtain the time entropy value and space entropy value of the historical students and record them as historical spatiotemporal entropy values, and divide the students into different learning groups based on the historical spatiotemporal entropy values ​​through a clustering algorithm.

[0098] The learning cycle management module is used to convert the historical learning behavior of students into time series data and extract key features through the autoencoder to obtain learning behavior pattern information, obtain the turning point of the learning behavior pattern through the time series turning point recognition algorithm, and divide the learning cycle into multiple learning sub-cycles according to the turning point; analyze the similarity of the students' learning behavior pattern information through the time series clustering method, classify the students in each learning sub-cycle according to the similarity of the learning behavior pattern information, and mark the learning behavior pattern labels.

[0099] The predicted evaluation value calculation module is used to obtain the student's historical homework data and extract homework features, and obtain the evaluation value through data analysis of the homework features; the evaluation value is analyzed and predicted based on the time series model to obtain the student's predicted evaluation value in the current learning sub-cycle.

[0100] The learning effect analysis and behavior management module is used to obtain the reference evaluation values ​​of different learning groups in the current learning sub-cycle through statistical analysis methods; obtain the spatiotemporal entropy value of the students and identify the learning group to which the students belong based on their spatiotemporal entropy values, and perform data analysis on the predicted evaluation value and the reference evaluation value corresponding to the learning group to which the students belong to obtain the learning effect coefficient of the students. When the learning effect coefficient is less than the preset effect threshold, the learning behavior pattern label of the students is obtained and adjusted.

[0101] Furthermore, the prediction evaluation value calculation module includes:

[0102] The data processing module is used to obtain the job evaluation standard image data as the standard data set, perform data preprocessing on the standard data set to obtain the first data set, perform denoising on the first data set by a probability statistics method to obtain the second data set, and perform classification and cutting on the second data set by a cutting algorithm to obtain the third data set.

[0103] The prediction score analysis module is used to extract features from the third data set and then classify it using an SVM classifier to obtain standard features; establish a connection between the standard features and the standard evaluation scores through the Delphi method to obtain a feature-score relationship distribution; use a feature matching analysis method to obtain feature similarity between the student's homework features and the standard features; when the feature similarity is greater than a preset feature similarity threshold, determine the standard features to which the student's homework features belong and combine them with the feature-score relationship distribution to obtain the student's evaluation score, use time series analysis based on the student's evaluation score to obtain a trend of evaluation score changes, and obtain a predicted score based on the trend of evaluation score changes.

[0104] The predicted evaluation value analysis module is used to obtain a fluctuation coefficient according to the evaluation score change trend, and the fluctuation coefficient is used to judge the stability of the student's evaluation score; the predicted score is corrected according to the fluctuation coefficient to obtain a corrected evaluation score and recorded as the predicted evaluation value.

[0105] Furthermore, the prediction evaluation value analysis module includes the following steps:

[0106] Get the evaluation score sequence of the students in the current learning sub-cycle as ,in Indicates The evaluation score of the evaluation, is the total number of evaluations; the changing trend of the evaluation scores is obtained based on the evaluation score sequence.

[0107] The coefficient of fluctuation is the standard deviation of the evaluation score The average of the evaluation scores The ratio of; the mean of the evaluation scores The calculation formula is: ; The standard deviation of the evaluation score The calculation formula is: The fluctuation coefficient The calculation formula is:

[0108] .

[0109] The predicted evaluation score of the students obtained by the time series model is ; When the fluctuation coefficient If the fluctuation coefficient is less than the preset safety fluctuation coefficient threshold, Make corrections to get the corrected evaluation score ; The modified evaluation score The calculation formula is:

[0110] .

[0111] in, It is the correction coefficient, which indicates the influence of the volatility coefficient on the correction score and is obtained through statistical analysis of historical data.

[0112] When the fluctuation coefficient When it is greater than the preset safety fluctuation coefficient threshold, the student's learning behavior pattern label is obtained and adjusted.

[0113] Furthermore, the prediction score analysis module includes:

[0114] A standard feature management module is used to extract global features and local features from the third data, and obtain reduced-dimensionality global features and reduced-dimensionality local features after performing dimension reduction through a dimension reduction algorithm; after performing feature fusion on the reduced-dimensionality global features and reduced-dimensionality local features, feature training is performed through an SVM classifier and classification is performed to obtain standard features;

[0115] The feature similarity calculation module is used to obtain low-frequency components and high-frequency components through contour wave transformation before extracting features from the student's historical homework data and the third data set; the student's homework features and the low-frequency components of the standard features are similarly calculated through the minimum Euclidean distance criterion to obtain a first similarity; when the first similarity is greater than a preset first similarity threshold, the high-frequency components of the student's homework features and the standard features are decomposed to select high-frequency feature points greater than the preset high-frequency threshold, and the high-frequency feature points are matched with the feature point matching algorithm to obtain a second similarity and recorded as feature similarity.

[0116] Furthermore, the learning group division module includes the following steps:

[0117] The time entropy value It is used to indicate the concentration of students during the learning period. The formula is:

[0118] .

[0119] in, represents the number of study periods, Indicated in The learning probability of a learning period; Indicated in The number of learning sessions in each learning session; Indicates the total number of learning times in all learning periods.

[0120] The spatial entropy value It is used to indicate the distribution of students in the learning position. The spatial entropy value The formula is:

[0121] .

[0122] in, represents the number of learned position categories, Indicated in The learning probability of a learning position; Indicated in The number of learning positions; Represents the total number of learnings at all learning positions.

[0123] It should be noted that, regarding the system in the above embodiment, the specific manner in which each module performs operations has been described in detail in the embodiment of the method, and will not be elaborated here.

[0124] Finally, it should be noted that: Although the present invention has been described in detail with reference to the aforementioned embodiments, it is still possible for those skilled in the art to modify the technical solutions described in the aforementioned embodiments, or to make equivalent substitutions for some of the technical features therein. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A method for predicting student learning behavior trends based on time series analysis, characterized in that: The method comprises: Obtain the learning logs of historical students, perform data cleaning on the learning logs and extract characteristic variables, wherein the characteristic variables include learning behavior, timestamp, and location information; perform data analysis based on the characteristic variables to obtain the time entropy value and space entropy value of the historical students and record them as the historical spatiotemporal entropy value; and divide the students into different learning groups through a clustering algorithm based on the historical spatiotemporal entropy value; The learning behavior of historical students is converted into time series data and key features are extracted through the autoencoder to obtain the learning behavior pattern information. The learning behavior pattern information is used through the time series turning point recognition algorithm to obtain the turning point of the learning behavior pattern. The learning cycle is divided into multiple learning sub-cycles according to the turning point. The similarity of the learning behavior pattern information of the students is analyzed through the time series clustering method. According to the similarity of the learning behavior pattern information, the students are classified in each learning sub-cycle and marked with learning behavior pattern labels; Obtain the historical homework data of the students and extract the homework features, and obtain the evaluation value through data analysis of the homework features; and analyze and predict the evaluation value based on the time series model to obtain the predicted evaluation value of the students in the current learning sub-period; Obtain reference evaluation values ​​of different learning groups in the current learning sub-cycle through statistical analysis methods; obtain the spatiotemporal entropy value of the student and identify the learning group to which the student belongs based on its spatiotemporal entropy value, perform data analysis on the predicted evaluation value and the reference evaluation value corresponding to the learning group to which the student belongs to obtain the learning effect coefficient of the student, and when the learning effect coefficient is less than a preset effect threshold, obtain the learning behavior pattern label of the student and adjust it; The method of obtaining the historical homework data of the trainee and extracting the homework features, obtaining the evaluation value through data analysis of the homework features; and obtaining the predicted evaluation value of the trainee in the current learning sub-period by analyzing and predicting the evaluation value based on the time series model includes: Obtaining homework evaluation standard image data as a standard data set, performing data preprocessing on the standard data set to obtain a first data set, performing denoising on the first data set by a probability statistics method to obtain a second data set, and performing classification and cutting on the second data set by a cutting algorithm to obtain a third data set; the standard image data includes images of standard calligraphy fonts or standard copybooks; the cutting algorithm uses a minimum bounding box algorithm to perform text cutting; Extract global features and local features from the third data set, and reduce the dimension by a dimension reduction algorithm to obtain reduced-dimensional global features and reduced-dimensional local features; perform feature fusion on the reduced-dimensional global features and reduced-dimensional local features, perform feature training and classification by a SVM classifier to obtain standard features; the feature fusion adopts a serial fusion method; Before extracting features from the student's historical homework data and the third data set, a low-frequency component and a high-frequency component are obtained by contourlet transformation; the student's homework features and the low-frequency components of the standard features are similarly calculated by the minimum Euclidean distance criterion to obtain a first similarity; when the first similarity is greater than a preset first similarity threshold, the high-frequency components of the student's homework features and the standard features are decomposed to select high-frequency feature points greater than the preset high-frequency threshold, and the high-frequency feature points are matched with the feature point matching algorithm to obtain a second similarity and recorded as feature similarity; The third data set is subjected to feature extraction and then classified by using an SVM classifier to obtain standard features; the standard features are linked to standard evaluation scores by using the Delphi method to obtain feature-score relationship distribution; the student's homework features and standard features are subjected to feature matching analysis to obtain feature similarity; when the feature similarity is greater than a preset feature similarity threshold, the standard features to which the student's homework features belong are determined and combined with the feature-score relationship distribution to obtain the student's evaluation score, and a time series analysis is performed based on the student's evaluation score to obtain an evaluation score change trend, and a predicted score is obtained based on the evaluation score change trend; A fluctuation coefficient is obtained according to the change trend of the assessment score, and the fluctuation coefficient is used to judge the stability of the assessment score of the student; the predicted score is corrected according to the fluctuation coefficient to obtain a corrected assessment score and recorded as the predicted evaluation value; The method of obtaining a fluctuation coefficient according to the evaluation score change trend, wherein the fluctuation coefficient is used to judge the stability of the student's evaluation score; and correcting the predicted score according to the fluctuation coefficient to obtain a corrected evaluation score and recording it as a predicted evaluation value includes: Get the evaluation score sequence of the students in the current learning sub-cycle as ,in Indicates The evaluation score of the evaluation, is the total number of evaluations; the changing trend of the evaluation scores is obtained according to the evaluation score sequence; The coefficient of fluctuation is the standard deviation of the evaluation score The average of the evaluation scores The ratio of; the mean of the evaluation scores The calculation formula is: ; The standard deviation of the evaluation score The calculation formula is: The fluctuation coefficient The calculation formula is: ； The predicted evaluation score of the students obtained by the time series model is ; When the fluctuation coefficient If the fluctuation coefficient is less than the preset safety fluctuation coefficient threshold, Make corrections to get the corrected evaluation score ; The modified evaluation score The calculation formula is: ； in, is the correction coefficient, which indicates the influence of the fluctuation coefficient on the correction score, obtained through statistical analysis of historical data; When the fluctuation coefficient When it is greater than the preset safety fluctuation coefficient threshold, the student’s learning behavior pattern label is obtained and adjusted; The method of performing data analysis according to the characteristic variables to obtain the time entropy value and space entropy value of the historical students and recording them as the historical time-space entropy value includes: The time entropy value It is used to indicate the concentration of students during the learning period. The formula is: ； in, represents the number of study periods, Indicated in The learning probability of a learning period; Indicated in The number of learning sessions in each learning session; Represents the total number of learning times in all learning periods; The spatial entropy value It is used to indicate the distribution of students in the learning position. The spatial entropy value The formula is: ； in, represents the number of learned position categories, Indicated in The learning probability of a learning position; Indicated in The number of learning times for each learning position; Represents the total number of learnings at all learning positions.

2. A student learning behavior trend prediction system based on time series analysis, characterized in that: The system includes: a learning group division module, a learning cycle management module, a prediction evaluation value calculation module, a learning effect analysis and behavior management module, and each module is sequentially connected in communication; The learning group division module is used to obtain the learning logs of historical students, extract characteristic variables after data cleaning of the learning logs, and the characteristic variables include learning behavior, timestamp, and location information; perform data analysis based on the characteristic variables to obtain the time entropy value and space entropy value of the historical students and record them as the historical spatiotemporal entropy value, and divide the students into different learning groups through a clustering algorithm based on the historical spatiotemporal entropy value; A learning cycle management module is used to convert the learning behavior of historical students into time series data and extract key features through an autoencoder to obtain learning behavior pattern information, obtain the turning point of the learning behavior pattern through a time series turning point recognition algorithm, and divide the learning cycle into multiple learning sub-cycles according to the turning point; analyze the similarity of the learning behavior pattern information of students through a time series clustering method, classify the students in each learning sub-cycle according to the similarity of the learning behavior pattern information, and mark the learning behavior pattern labels; The predicted evaluation value calculation module is used to obtain the historical homework data of the students and extract the homework features, and obtain the evaluation value through data analysis of the homework features; the evaluation value is analyzed and predicted based on the time series model to obtain the predicted evaluation value of the students in the current learning sub-period; The learning effect analysis and behavior management module is used to obtain the reference evaluation values ​​of different learning groups in the current learning sub-cycle through statistical analysis methods; obtain the spatiotemporal entropy value of the trainee and identify the learning group to which the trainee belongs based on its spatiotemporal entropy value, perform data analysis on the predicted evaluation value and the reference evaluation value corresponding to the learning group to which the trainee belongs to obtain the learning effect coefficient of the trainee, and when the learning effect coefficient is less than the preset effect threshold, obtain the learning behavior pattern label of the trainee and adjust it; The prediction evaluation value calculation module includes: A data processing module is used to obtain standard image data of homework evaluation as a standard data set, perform data preprocessing on the standard data set to obtain a first data set, perform denoising on the first data set by a probability statistics method to obtain a second data set, and perform classification and cutting on the second data set by a cutting algorithm to obtain a third data set; the standard image data includes images of standard calligraphy fonts or standard copybooks; the cutting algorithm uses a minimum bounding box algorithm to perform text cutting; A prediction score analysis module is used to extract features from the third data set and then classify it using an SVM classifier to obtain standard features; establish a relationship between the standard features and the standard evaluation scores using the Delphi method to obtain a feature-score relationship distribution; use a feature matching analysis method to obtain feature similarity between the student's homework features and the standard features; when the feature similarity is greater than a preset feature similarity threshold, determine the standard features to which the student's homework features belong and combine the feature-score relationship distribution to obtain the student's evaluation score, use time series analysis based on the student's evaluation score to obtain an evaluation score change trend, and obtain a predicted score based on the evaluation score change trend; A predicted evaluation value analysis module is used to obtain a fluctuation coefficient according to the evaluation score change trend, and the fluctuation coefficient is used to judge the stability of the student's evaluation score; the predicted score is corrected according to the fluctuation coefficient to obtain a corrected evaluation score and recorded as the predicted evaluation value; The prediction evaluation value analysis module comprises the following steps: Get the evaluation score sequence of the students in the current learning sub-cycle as ,in Indicates The evaluation score of the evaluation, is the total number of evaluations; the changing trend of the evaluation scores is obtained according to the evaluation score sequence; The coefficient of fluctuation is the standard deviation of the evaluation score The average of the evaluation scores The ratio of; the mean of the evaluation scores The calculation formula is: ; The standard deviation of the evaluation score The calculation formula is: The fluctuation coefficient The calculation formula is: ； The predicted evaluation score of the students obtained by the time series model is ; When the fluctuation coefficient If the fluctuation coefficient is less than the preset safety fluctuation coefficient threshold, Make corrections to get the corrected evaluation score ; The modified evaluation score The calculation formula is: ； in, is the correction coefficient, which indicates the influence of the fluctuation coefficient on the correction score, obtained through statistical analysis of historical data; When the fluctuation coefficient When it is greater than the preset safety fluctuation coefficient threshold, the student’s learning behavior pattern label is obtained and adjusted; The prediction score analysis module includes: A standard feature management module is used to extract global features and local features from the third data set, and obtain reduced-dimensionality global features and reduced-dimensionality local features after performing dimension reduction through a dimension reduction algorithm; after performing feature fusion on the reduced-dimensionality global features and reduced-dimensionality local features, feature training and classification are performed through an SVM classifier to obtain standard features; the feature fusion adopts a serial fusion method; A feature similarity calculation module is used to obtain low-frequency components and high-frequency components by contourlet transformation before extracting features from the student's historical homework data and the third data set; perform similarity calculation on the low-frequency components of the student's homework features and the standard features by using the minimum Euclidean distance criterion to obtain a first similarity; when the first similarity is greater than a preset first similarity threshold, the high-frequency components of the student's homework features and the standard features are decomposed to select high-frequency feature points greater than the preset high-frequency threshold, and the high-frequency feature points are matched with the feature point matching algorithm to obtain a second similarity and recorded as feature similarity; The learning group division module includes the following steps: The time entropy value It is used to indicate the concentration of students during the learning period. The formula is: ； in, represents the number of study periods, Indicated in The learning probability of a learning period; Indicated in The number of learning sessions in each learning session; Represents the total number of learning times in all learning periods; The spatial entropy value It is used to indicate the distribution of students in the learning position. The spatial entropy value The formula is: ； in, represents the number of learned position categories, Indicated in The learning probability of a learning position; Indicated in The number of learning times for each learning position; Represents the total number of learnings at all learning positions.

Images