A mistake question classification analysis and improvement method based on big data
By classifying incorrect questions using big data analysis and the Naive Bayes classification algorithm, and combining factor analysis to calculate the error rate and score improvement rate, the problem of insufficient error classification analysis was solved, thus improving the effect of improving students' academic performance.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- ZHONGNAN XUNZHI TECH CO LTD
- Filing Date
- 2024-01-15
- Publication Date
- 2026-08-04
AI Technical Summary
The lack of effective methods for classifying and analyzing incorrect answers in current technologies has resulted in insufficient improvement in students' academic performance.
The Naive Bayes classification algorithm based on big data is used to classify and summarize the reasons for incorrect questions, and factor analysis is combined to calculate the error rate and score improvement rate. Targeted learning guidance is then provided through the learning information server.
It enables precise categorization of incorrect answers and targeted training, thereby improving the efficiency of students' academic performance.
Smart Images
Figure SMS_1 
Figure SMS_2 
Figure SMS_3
Abstract
Description
Technical Field
[0001] This invention relates to the field of student learning data processing, specifically to a method for classifying, analyzing, and improving incorrect answers based on big data. Background Technology
[0002] Error classification analysis involves analyzing students' learning process and academic performance. It analyzes students' learning outcomes within a specific unit or activity to help them improve their grades and progress.
[0003] If a method for classifying and analyzing incorrect answers could be provided, enabling the classification and summarization of the reasons for students' mistakes, the calculation of the weight of each student's error, and the application of targeted training, it would help improve students' academic performance. Summary of the Invention
[0004] The technical problem to be solved by this invention is to provide a method for classifying and analyzing incorrect questions based on big data, which is beneficial to improving students' academic performance.
[0005] The present invention achieves its objective by employing the following technical solution:
[0006] A method for classifying and improving incorrect answers based on big data, characterized by the following steps:
[0007] Step 1: Using the learning server, the Naive Bayes classification algorithm is used to classify and summarize the reasons for incorrect answers, including knowledge-based errors and non-knowledge-based errors;
[0008] The implementation process of the Naive Bayes classification algorithm consists of the following steps:
[0009] 1) For each of the n attributes A1, A2...An in the sample set, use an n-dimensional feature vector X = {x1, x2, ..., xn} to represent the data;
[0010] 2) Assume there are m classes C1, C2, ..., Cm, and a sample X of unknown class;
[0011] P(Ci|X) represents the probability that Ci will occur under the condition of X;
[0012] Under the condition of X, the hypothesis that maximizes the probability of Ci occurring, i.e., when P(Ci|X) is maximized, is considered that X can be assigned to the category Ci. This is called the maximum a posteriori hypothesis, where P(Ci|X) can be calculated using the formula mentioned earlier:
[0013]
[0014] For P(X|Ci), due to the assumption of class conditional independence, it is assumed that:
[0015]
[0016] 3) Based on the analysis in step 2), when classifying X, it is necessary to calculate P(X|Ci)P(Ci) for each class Ci. Then, among classes C1, C2, ... Cm, the class Ci that maximizes P(X|Ci)P(Ci) is the class to which the sample X is finally assigned.
[0017] Step 2: Using the learning data server, we will analyze and summarize the weight of the same error in the overall error reasons, the weight of the error type in the score, and the weight of the error type across all students using factor analysis.
[0018] ① The incorrect question bank was classified according to the Naive Bayes classification algorithm, and the total number of each category was obtained as m1, m2, ... m8. The proportion of each type of error among all errors was calculated:
[0019] P1 = m1 / (m1 + m2 + m3 + ... + m8)
[0020] Based on the above formula, the proportion of each error cause in the total number of errors can be calculated;
[0021] ② Based on the Naive Bayes classification algorithm, calculate the proportion of each incorrectly answered question's score in the overall score:
[0022] R-errors = Error score / Overall score
[0023] ③ Based on the Naive Bayes classification algorithm, the total number and score of all incorrect questions can be used to calculate the proportion of each question among all students:
[0024] R_student_errors = Number of student errors / (Number of student errors + Number of student correct answers);
[0025] Step 3: Students take multiple exams or practice sessions through the student terminal. The learning server calculates the percentage of students who improve their scores for different types of mistakes, and the improvement effect after targeted learning for different mistakes.
[0026] (1) There are 8 types of incorrect questions. For each exam, the scores Ga of all incorrect questions are calculated based on the first algorithm, and the scores Ga1, Ga2, ... Ga8 of each type of incorrect question are calculated.
[0027] (2) The statistical results of the next exam are the total score of all wrong questions Gb, and the score of each type of wrong question Gb1, Gb2, ... Gb8;
[0028] (3) Calculate the percentage increase in student scores for each type of incorrect question:
[0029] R = (Ga - Gb) / Ga
[0030] Based on the above formula, by substituting the scores corresponding to different errors in different exams, we can calculate the student's score improvement rates R1, R2, ... R8 for different errors, as well as the total error improvement rate R.
[0031] (4) By repeating steps (2) and (3), the improvement of students' scores in different or even multiple exams can be obtained, so that different improvement methods can be used for different improvement rates.
[0032] As a further limitation of this technical solution, step four is also included: The learning server calculates the speed at which students improve after learning in different directions, and then, based on the achieved scores and parallel analysis with other students, judges the efficiency of different targeted learning methods in improving scores, and calculates the optimal time allocation for targeted learning.
[0033] As a further limitation of this technical solution, in order to provide targeted practice for different errors and to consider the time and effort invested in learning, it is necessary to quantify the rate of improvement for different errors:
[0034] (i) There are a total of 8 types of incorrect questions. In each exam, the scores Ga of all incorrect questions are calculated based on the first algorithm, and the scores Ga1, Ga2, ... Ga8 of each type of incorrect question.
[0035] (ii) Based on the actual situation of students, the tolerance level for each type of error, that is, the standard score for each type of error, is determined. The tolerance levels for different errors are denoted as t1, t2, ... t8, which represent the standard for allowing points to be lost for different errors. In other words, if the points lost for different errors are below the t value, it is considered normal and indicates that this part has met the standard.
[0036] (III) Calculate the improvement rate v for different errors. The first time, each type of error scores the highest because it did not undergo the targeted practice and improvement of this invention. The first time is denoted as Ga1, Ga2, ... Ga8, and the final passing score is denoted as t1, t2, ... t8. The number of attempts n1, n2, ... n8 are used to achieve the passing score.
[0037] v=(Ga-t) / n
[0038] Based on the above formula, the boost rates v1, v2, ... v8 for different errors can be calculated;
[0039] (iv) By comparing the improvement speed v of different errors and ranking them, we can determine the time, effort and resources required to improve different errors. Errors that improve quickly can be improved in less time, while errors that improve slowly are relatively weak and require more time to be invested in order to make targeted changes.
[0040] As a further limitation of this technical solution, the specific steps of step one are as follows:
[0041] Step 11: Data preparation; data cleaning, correlation analysis and attribute subset selection, data transformation and reduction;
[0042] Steps 1 and 2: Data mining; using the Naive Bayes classification algorithm to mine the data;
[0043] Step 13: Model Evaluation Phase.
[0044] As a further limitation of this technical solution, when the data mining stage ends, the data patterns generated may not all be useful. Therefore, a measure of interest in a certain pattern is set, such as confidence or rule support, and the useful data patterns that users are truly interested in are discovered based on these criteria.
[0045] The evaluation criteria mainly include precision (P), recall (R), and F1 score. A indicates that the text belongs to a certain category and the prediction result also belongs to a certain category. D indicates that the text does not belong to a certain category but the prediction result is that it belongs to a certain category. C indicates that the text belongs to a certain category but the prediction result shows that it does not belong to a certain category.
[0046]
[0047]
[0048]
[0049] As a further limitation of this technical solution, the knowledge-related errors include knowledge point defects, knowledge system defects, and incomplete knowledge frameworks, while the non-knowledge-related errors include carelessness, incorrect operation, calculation errors, lack of focus when doing problems, and guessing answers.
[0050] As a further limitation of this technical solution, the error correction module categorizes the questions of this type in the learning stage, determines the difficulty based on the total error rate of the errors, and selects a mode of easy-to-difficult or evenly distributed questions for students to practice.
[0051] As a further limitation of this technical solution, the learning server provides the class error rate for the incorrect question and the overall error rate for all users of this system, determines whether there is a general gap in the knowledge point, and pushes it to the teacher's end for overall reinforcement learning at the class level.
[0052] A learning information server, characterized in that it includes:
[0053] Databases are used to store computer programs and big data analysis systems for student learning.
[0054] A processor is used to execute the computer program and the learning data analysis system to implement the learning data analysis system as described in any one of claims 1 to 9.
[0055] Compared with the prior art, the advantages and positive effects of the present invention are:
[0056] 1. Existing education, including intelligent learning, focuses on correct answers and analyzes whether students are qualified and their accuracy rate. This system specifically summarizes and categorizes the reasons for errors, and comprehensively evaluates students' shortcomings from multiple directions and algorithms.
[0057] 2. This invention implements an improved multinomial Naive Bayes algorithm based on the Hadoop framework. It summarizes and organizes a large amount of incorrect question data to form a huge incorrect question database. Data mining is then performed on the database to extract the knowledge and information that is useful and interesting to users, which is to obtain the classification of incorrect question data.
[0058] 3. This device. Detailed Implementation
[0059] The following is a detailed description of a specific embodiment of the present invention, but it should be understood that the scope of protection of the present invention is not limited to the specific embodiment.
[0060] This invention includes the following steps:
[0061] Step 1: Using the learning server, the Naive Bayes classification algorithm is used to classify and summarize the reasons for incorrect answers, including knowledge-based errors and non-knowledge-based errors;
[0062] The specific steps of step one are as follows:
[0063] Step 11: Data preparation; data cleaning, correlation analysis and attribute subset selection, data transformation and reduction.
[0064] Data cleaning. Data cleaning is used to eliminate or reduce data noise and handle missing values. For data noise, we can use smoothing techniques; for missing values, we can replace them with the most frequent value in the attribute or the most likely value derived from statistics.
[0065] Correlation analysis and attribute subset selection. In applications, some data may be redundant. We can use correlation analysis to determine whether two given attributes are statistically correlated, thus eliminating redundant data. Additionally, the data may contain attributes unrelated to the classification. For these unrelated attributes, we can use attribute subset selection to find a reduced subset of attributes, making the probability distribution of the data class results as close as possible to the distribution generated using all attributes.
[0066] Data transformation and reduction. Normalization can transform data, converting it into patterns suitable for data mining. It can also transform data by generalizing to higher-level concepts, compressing the original data and reducing the time and operations required for classification learning. Furthermore, methods ranging from wavelet transform and principal component analysis to histogram analysis and clustering can be used to reduce data, shrinking a large dataset into a smaller one while maintaining the integrity of the larger dataset.
[0067] Steps 1 and 2: Data mining; use the Naive Bayes classification algorithm to mine the data.
[0068] This is the basic and core stage of the entire process. In this stage, the transformed data from the previous stage is mined. First, appropriate data mining algorithms, such as classification, clustering, association rules, and regression, are used based on relevant information. Then, the data is mined to finally derive useful data patterns.
[0069] This algorithm selects a classification algorithm for data mining. Classification methods reflect how to identify characteristic knowledge of things belonging to the same class and characteristic knowledge of things that do not belong to the same class. We choose the Naive Bayes method from statistical methods to construct the classification model.
[0070] Naive Bayes classification is a type of Bayesian classification algorithm, based on probabilistic analysis. It's also known as simple Bayesian classification because, compared to other Bayesian methods, it assumes class conditional independence. This means that for a given class, all attributes are independent, and the decision attributes of that class are equally affected by conditional attributes. This assumption significantly reduces computational cost, hence the name "naive." Naive Bayes is built on a solid mathematical foundation, employing Bayes' theorem:
[0071]
[0072] Where: P(A|B) is called the posterior probability, or the posterior probability of A under condition B, which represents the probability of A occurring under condition B.
[0073] P(A) is called the prior probability, or the prior probability of A. It simply represents the probability of A occurring without any conditions.
[0074] The basic idea of the Naive Bayes classification algorithm is to assume that the attributes of different classes are independent and do not affect each other. Under this premise, given the data to be classified, the algorithm calculates the probability of different classes under the given conditions and assigns the data to the class with the highest probability. This can be basically expressed as:
[0075] Suppose there are n classes C1, C2, ..., Cn. Given a sample X to be classified, then the classification is successful if and only if:
[0076] P(C,|X)>P(C,|X)1≤j≤n,j≠i
[0077] At this point, sample X is considered to belong to Ci, meaning the Naive Bayes classification algorithm assigns the sample X to the class with the highest posterior probability.
[0078] The implementation process of the Naive Bayes classification algorithm consists of the following steps:
[0079] 1) For each of the n attributes A1, A2...An in the sample set, use an n-dimensional feature vector X = {x1, x2, ..., xn} to represent the data;
[0080] 3) Assume there are m classes C1, C2, ..., Cm, and a sample X of unknown class;
[0081] P(Ci|X) represents the probability that Ci will occur under the condition of X;
[0082] Under the condition of X, the hypothesis that maximizes the probability of Ci occurring, i.e., when P(Ci|X) is maximized, is considered that X can be assigned to the category Ci. This is called the maximum a posteriori hypothesis, where P(Ci|X) can be calculated using the formula mentioned earlier:
[0083]
[0084] For P(X|Ci), due to the assumption of class conditional independence, it is assumed that:
[0085]
[0086] 3) Based on the analysis in step 2), when classifying X, it is necessary to calculate P(X|Ci)P(Ci) for each class Ci. Then, among classes C1, C2, ... Cm, the class Ci that maximizes P(X|Ci)P(Ci) is the class to which the sample X is finally assigned.
[0087] Step 13: Model Evaluation Phase.
[0088] When the data mining phase ends, the resulting data patterns may not all be useful. Therefore, it is necessary to set a measure of interest in a pattern, such as confidence or rule support, and to discover the truly useful data patterns that users are interested in based on these criteria.
[0089] The evaluation criteria mainly include precision (P), recall (R), and F1 score. A indicates that the text belongs to a certain category and the prediction result also belongs to a certain category. D indicates that the text does not belong to a certain category but the prediction result is that it belongs to a certain category. C indicates that the text belongs to a certain category but the prediction result shows that it does not belong to a certain category.
[0090]
[0091]
[0092]
[0093] Step 2: Using the learning data server, we will analyze and summarize the weight of the same error in the overall error reasons, the weight of the error type in the score, and the weight of the error type across all students using factor analysis.
[0094] Calculating the proportion and weight of data is a common statistical analysis method. In actual statistical research, it is necessary to select the appropriate method based on the specific characteristics of the data. For example, if the volatility between data is a kind of information, then the CRITIC weighting method or information weighting method can be considered; or if it is expert scoring data, then the AHP hierarchical method or the maximal order graph method can be used.
[0095] Both factor analysis and principal component analysis (PCA) can calculate weights when determining proportions, and they utilize the exact same principle: information condensation. The difference lies in the fact that factor analysis incorporates a 'rotation' function, while PCA focuses more on condensing information. The 'rotation' function makes factors more interpretable; if interpretability is desired, factor analysis is generally preferred. This doesn't mean that PCA results are entirely uninterpretable; it simply means their interpretability is sometimes relatively poor. However, PCA is faster and therefore more widely used.
[0096] This invention employs factor analysis to calculate the weighted proportion of errors in different categories, the weighted proportion of errors in scores, and the weighted proportion of errors among all students. Eight different categories are categorized and statistically analyzed: knowledge point deficiencies, knowledge system deficiencies, incomplete knowledge frameworks, carelessness, operational errors, calculation errors, lack of focus during problem-solving, and guessing. These eight analytical items can be condensed into two aspects: principal component factors, namely, knowledge-based errors and non-knowledge-based errors.
[0097] ① The incorrect question bank was classified according to the Naive Bayes classification algorithm, and the total number of each category was obtained as m1, m2, ... m8. The proportion of each type of error among all errors was calculated:
[0098] P1 = m1 / (m1 + m2 + m3 + ... + m8)
[0099] Based on the above formula, the proportion of each error cause in the total number of errors can be calculated;
[0100] ② Based on the Naive Bayes classification algorithm, calculate the proportion of each incorrectly answered question's score in the overall score:
[0101] R-errors = Error score / Overall score
[0102] ③ Based on the Naive Bayes classification algorithm, the total number and score of all incorrect questions can be used to calculate the proportion of each question among all students:
[0103] R_student_errors = Number of student errors / (Number of student errors + Number of student correct answers).
[0104] Step 3: Students take multiple exams or practice sessions through the student terminal. The learning server calculates the percentage of students who improve their scores for different types of mistakes, and the improvement effect after targeted learning for different mistakes.
[0105] Data analysis of students' exam scores is conducted, focusing on statistical summarization and analysis of incorrect question types, and calculating the percentage of students who improved their scores for each type of incorrect question.
[0106] (1) There are 8 types of incorrect questions. For each exam, the scores Ga of all incorrect questions are calculated based on the first algorithm, and the scores Ga1, Ga2, ... Ga8 of each type of incorrect question are calculated.
[0107] (2) The statistical results of the next exam are the total score of all wrong questions Gb, and the score of each type of wrong question Gb1, Gb2, ... Gb8;
[0108] (3) Calculate the percentage increase in student scores for each type of incorrect question:
[0109] R = (Ga - Gb) / Ga
[0110] Based on the above formula, by substituting the scores corresponding to different errors in different exams, we can calculate the student's score improvement rates R1, R2, ... R8 for different errors, as well as the total error improvement rate R.
[0111] (4) By repeating steps (2) and (3), the improvement of students' scores in different or even multiple exams can be obtained, so that different improvement methods can be used for different improvement rates.
[0112] It also includes step four: The learning server calculates the speed at which students improve after learning in different directions, and then analyzes and judges the efficiency of different targeted learning methods in improving grades based on the achieved scores and other students' parallel analysis, and calculates the optimal time allocation for targeted learning.
[0113] In order to provide targeted practice for different types of errors and to assess the time and effort invested in learning, it is necessary to quantify the rate of improvement for each type of error:
[0114] (i) There are a total of 8 types of incorrect questions. In each exam, the scores Ga of all incorrect questions are calculated based on the first algorithm, and the scores Ga1, Ga2, ... Ga8 of each type of incorrect question.
[0115] (ii) Based on the actual situation of students, the tolerance level for each type of error, that is, the standard score for each type of error, is determined. The tolerance levels for different errors are denoted as t1, t2, ... t8, which represent the standard for allowing points to be lost for different errors. In other words, if the points lost for different errors are below the t value, it is considered normal and indicates that this part has met the standard.
[0116] (III) Calculate the improvement rate v for different errors. The first time, each type of error scores the highest because it did not undergo the targeted practice and improvement of this invention. The first time is denoted as Ga1, Ga2, ... Ga8, and the final passing score is denoted as t1, t2, ... t8. The number of attempts n1, n2, ... n8 are used to achieve the passing score.
[0117] v=(Ga-t) / n
[0118] Based on the above formula, the boost rates v1, v2, ... v8 for different errors can be calculated;
[0119] (iv) By comparing the improvement speed v of different errors and ranking them, we can determine the time, effort and resources required to improve different errors. Errors that improve quickly can be improved in less time, while errors that improve slowly are relatively weak and require more time to be invested in order to make targeted changes.
[0120] The knowledge-related errors include deficiencies in knowledge points, deficiencies in the knowledge system, and incomplete knowledge frameworks. The non-knowledge-related errors include carelessness, incorrect operation, calculation errors, lack of focus when doing problems, and guessing answers.
[0121] The "Incorrect Questions" module categorizes questions of this type for this learning stage, determines the difficulty based on the overall error rate of incorrect questions, and selects a pattern of easy-to-difficult or evenly distributed questions for students to practice.
[0122] The learning server provides the class error rate for the incorrect question and the overall error rate for all users of the system, determines whether there are any gaps in the knowledge point, and pushes the information to the teacher for overall reinforcement learning at the class level.
[0123] A learning information server includes:
[0124] Databases are used to store computer programs and big data analysis systems for student learning.
[0125] A processor is used to execute the computer program and the learning data analysis system to implement the learning data analysis system as described in any one of claims 1 to 9.
[0126] The learning server includes: exam analysis module, test paper analysis module, grade analysis module, ranking analysis module, knowledge point analysis module, score improvement strategy module, and printing module.
[0127] The exam analysis module provides an exam analysis page, which displays exam data for each school, grade, and class, as well as exam data and ranking information for each student.
[0128] The exam paper analysis module provides status information for exam papers in each subject, allows viewing total scores, multi-subject and individual subject score analysis, as well as score comparisons and trends in past exams; it also provides a subject analysis selection area for students or teachers to query their chosen subjects.
[0129] The interface for analyzing academic performance is also used to analyze the difficulty level of exam papers.
[0130] The performance analysis module provides a performance analysis page for students. The performance analysis page displays students' test scores, average scores for each test area and school, distribution of questions with different difficulty levels, and answer performance.
[0131] The ranking analysis module provides a ranking analysis page for students, which displays the total number of test takers, the highest score, the lowest score, the average score, the median, the mode, and the student's ranking.
[0132] The knowledge point analysis module provides students with a knowledge point analysis page, which includes an analysis of the knowledge points for each test paper, the score, and a prediction of the subject's difficulty level.
[0133] The score improvement strategy module provides a score improvement strategy page for students, analyzes weak knowledge points based on assessment results, and analyzes the direction for students to improve their scores in different knowledge points.
[0134] The printing module is used to provide teachers and students with the function of printing system data.
[0135] The teacher's client, student's client, and error correction module are all connected to the learning progress server. The teacher's client can access the student's client, error correction module, and learning progress server, and the student's client can access the error correction module and learning progress server.
[0136] The workflow of this invention is as follows:
[0137] After each exam, the exam analysis module provides status information for each subject's exam paper, allowing students to view total scores, multi-subject and individual subject score analysis, as well as score comparisons and trends across different exams. The exam analysis module collects student data and provides student exam data and ranking information.
[0138] The system performs data cleaning, correlation analysis, attribute subset selection, data transformation, and reduction on exam errors. The Naive Bayes classification algorithm is used to categorize and summarize the reasons for errors, and these errors are added to the error module, forming an error database for all students and an individual student's error database.
[0139] Through the learning data server, factor analysis is used to analyze and summarize the weighting of the same errors in the overall error causes, the weighting of error types in scores, and the weighting of error types across all students. The knowledge point analysis module provides students with a knowledge point analysis page, showing the knowledge points analyzed for each test paper, scores, and predicted scores for the subject at the same difficulty level. The score improvement strategy module provides students with a score improvement strategy page, analyzing weak knowledge points based on test results and outlining directions for improvement for different knowledge points.
[0140] Students access their own error database through the student client and take multiple tests or exercises. The learning server calculates the percentage of students who improve their scores for different types of errors, and the improvement effect after targeted learning for different errors.
[0141] The learning server calculates the speed at which students improve after learning in different directions, and then analyzes the efficiency of different targeted learning methods in improving grades based on the students' scores and in parallel with other students, and calculates the optimal time allocation for targeted learning.
[0142] Teachers can access the student interface, the error correction module, and the learning progress server through the teacher's terminal to view student exam data and rankings, as well as compare scores and trends across different exams. Based on the error rate of a particular question in the class and the overall error rate for all users of the system, it can be determined whether there are any gaps in knowledge on that topic, allowing for targeted learning and specific training to address knowledge-related errors, deficiencies in knowledge points, deficiencies in the knowledge system, and incomplete knowledge frameworks.
[0143] The above-disclosed embodiments are merely specific examples of the present invention. However, the present invention is not limited thereto, and any variations that can be conceived by those skilled in the art should fall within the protection scope of the present invention.
Claims
1. A method for error classification analysis and improvement based on big data, characterized in that, Includes the following steps: Step 1: Using the learning server, the Naive Bayes classification algorithm is used to classify and summarize the reasons for incorrect answers, including knowledge-based errors and non-knowledge-based errors; The implementation process of the Naive Bayes classification algorithm consists of the following steps: 1) For each of the n attributes A1, A2...An in the sample set, use an n-dimensional feature vector. The data can be represented by X = {x1, x2, ..., xn}; 2) Assume there are m classes C1, C2, ..., Cm, and a sample X of unknown class; P(Ci|X) represents the probability that Ci will occur under the condition of X; Given X, the hypothesis that maximizes the probability of Ci occurring, i.e., when P(Ci|X) is maximized, is considered to assign X to the category Ci. This is called the maximum a posteriori hypothesis, where P(Ci|X) is calculated using the formula mentioned earlier. For P(X|Ci), due to the assumption of class conditional independence, it is assumed that: 3) Based on the analysis in step 2), when classifying X, it is necessary to calculate P(X|Ci)P(Ci) for each class Ci. Then, among classes C1, C2, ... Cm, the class Ci that maximizes P(X|Ci)P(Ci) is the class to which the sample X is finally assigned. Step 2: Using the learning data server, we will analyze and summarize the weight of the same error in the overall error reasons, the weight of the error type in the score, and the weight of the error type across all students using factor analysis. ① The incorrect question bank was classified according to the Naive Bayes classification algorithm, and the total number of each category was obtained as m1, m2, ... m8. The proportion of each type of error among all errors was calculated: P1 = m1 / (m1 + m2 + m3 + ... + m8) Based on the above formula, calculate the proportion of each error cause in the total number of errors; ② Based on the Naive Bayes classification algorithm, calculate the proportion of each incorrectly answered question's score in the overall score: R-errors = Error score / Overall score ③ Based on the Naive Bayes classification algorithm, the total number and score of all incorrect questions are obtained, and their proportion among all students is calculated: R_student_errors = Number of student errors / (Number of student errors + Number of student correct answers); Step 3: Students take multiple exams or practice sessions through the student terminal. The learning server calculates the percentage of students who improve their scores for different types of mistakes, and the improvement effect after targeted learning for different mistakes. (1) There are 8 types of incorrect questions. For each exam, the scores Ga of all incorrect questions are calculated based on the first algorithm, and the scores Ga1, Ga2, ... Ga8 of each type of incorrect question are calculated. (2) The statistical results of the next exam are the total score of all wrong questions Gb, and the scores of wrong questions in each category Gb1, Gb2, ... Gb8; (3) Calculate the percentage increase in student scores for each type of incorrect question: R=(Ga-Gb) / Ga Based on the above formula, by substituting the scores corresponding to different errors in different exams, we can calculate the student's score improvement rates R1, R2, ... R8 for different errors, as well as the total error improvement rate R. (4) By repeating steps (2) and (3), the improvement of students' scores in different tests can be obtained, so that different improvement methods can be used for different improvement rates.
2. The method for classifying and improving incorrect answers based on big data according to claim 1, characterized in that: It also includes step four: The learning server calculates the speed at which students improve after learning in different directions, and then analyzes and judges the efficiency of different targeted learning methods in improving grades based on the achieved scores and other students' parallel analysis, and calculates the optimal time allocation for targeted learning.
3. The method for classifying and improving incorrect answers based on big data according to claim 2, characterized in that: In order to provide targeted practice for different types of errors and to assess the time and effort invested in learning, it is necessary to quantify the rate of improvement for each type of error: (a) There are a total of 8 types of incorrect questions. In each exam, the scores Ga of all incorrect questions are calculated based on the first algorithm, and the scores Ga1, Ga2, ... Ga8 of each type of incorrect question. (ii) Based on the actual situation of students, the tolerance level for each type of error, that is, the standard score for each type of error, is determined. The tolerance levels for different errors are denoted as t1, t2, ... t8, which represent the standard for allowing points to be lost for different errors. In other words, if the points lost for different errors are below the t value, it is considered normal and indicates that this part has met the standard. (III) Calculate the improvement rate v for different errors. The first time, each type of error scores the highest because there is no targeted practice or improvement. The first time is denoted as Ga1, Ga2, ... Ga8, and the final passing score is denoted as t1, t2, ... t8. The number of attempts n1, n2, ... n8 are used to achieve the passing score. v=(Ga-t) / n Based on the above formula, the boost rates v1, v2, ... v8 for different errors are calculated; (iv) By comparing the improvement speed v of different errors and ranking them, we can determine the time, effort and resources required to improve different errors. Errors that improve quickly require less time, while errors that improve slowly are relatively weak and require more time to be improved in order to make targeted changes.
4. The method for error classification analysis and improvement based on big data according to claim 1, characterized in that: The specific steps of step one are as follows: Step 11: Data preparation; data cleaning, correlation analysis and attribute subset selection, data transformation and reduction; Steps 1 and 2: Data mining; using the Naive Bayes classification algorithm to mine the data; Step 13: Model Evaluation Phase.
5. The method for error classification analysis and improvement based on big data according to claim 4, characterized in that: When the data mining phase ends, the resulting data patterns may not all be useful. Therefore, it is necessary to set a measure of interest in a certain pattern, set confidence levels or rule support levels, and use these criteria to discover the data patterns that users are truly interested in and that are useful. The evaluation criteria are precision (P), recall (R), and F1 score. A indicates that the text belongs to a certain category and the prediction result also belongs to a certain category; D indicates that the text does not belong to a certain category but the prediction result is that it belongs to a certain category; and C indicates that the text belongs to a certain category but the prediction result shows that it does not belong to a certain category. 。 6. The method for error classification analysis and improvement based on big data according to claim 1, characterized in that: The knowledge-related errors include deficiencies in knowledge points, deficiencies in the knowledge system, and incomplete knowledge frameworks. The non-knowledge-related errors include carelessness, incorrect operation, calculation errors, lack of focus when doing problems, and guessing answers.
7. The method for error classification analysis and improvement based on big data according to claim 1, characterized in that: The error correction module determines the difficulty level based on the total error rate of the questions and offers a choice between easy-to-difficult or evenly distributed modes for students to practice.
8. The method for error classification analysis and improvement based on big data according to claim 1, characterized in that: The learning server provides the error rate of each class that made a mistake and the overall error rate of all users of the system to determine whether there are any gaps in the knowledge points. The server then pushes the information to the teachers for overall reinforcement learning at the class level.
9. A learning information server, characterized in that, include: Databases are used to store computer programs and big data analysis systems for student learning. A processor is used to execute the computer program and the learning big data analysis system to implement the big data-based error classification analysis and improvement method as described in any one of claims 1 to 8.