Rapid memory result test method based on AI learning
Through the fast memory outcome test method based on AI learning, adaptive test questionnaire is generated and memory errors are analyzed, and subjectivity and adaptability problems of traditional memory detection are solved, and efficient and personalized memory evaluation and review plan generation are achieved.
Patent Information
- Application Number
- CN202510364808.6
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-03-26
- Publication Date
- 2025-07-04
AI Technical Summary
Traditional memory detection methods are subjective and difficult to adaptively adjust the test difficulty and review plan, and fail to effectively analyze the causes of memory errors.
Using the fast memory outcome test method based on AI learning, we automatically generate review strategies by generating preliminary test questionnaires, memory test questionnaires, generating memory network diagrams and analyzing error sets, and using large language models and word vector models to build memory detection models.
It realizes adaptive adjustment of the test difficulty according to the user's cognitive level, eliminates the subjectivity of manual assessment, automatically analyzes the causes of errors and generates a personalized review plan, which improves the efficiency of users' mastery of knowledge points.
Smart Images
Figure CN120259044A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of deep learning, and more specifically, it relates to a method for quickly testing memory results based on AI learning. Background Art
[0002] In traditional memory detection methods, manual evaluation, questionnaire evaluation and other detection methods are usually used to evaluate the memory level of learners. Although these methods can measure the mastery of learners in memory tests to a certain extent, there are still the following limitations: 1. Traditional manual evaluation is somewhat subjective, and there are differences in memory and expression methods among individuals, resulting in difficulty in providing an objective and quantitative memory level measurement standard; 2. Traditional questionnaire tests usually use a fixed test set, and there are differences in cognitive levels among individuals, resulting in difficulty in adaptively adjusting the test difficulty according to the user's cognitive level; 3. Traditional evaluation methods usually only focus on the correctness of memory and ignore the reasons for errors, such as conceptual understanding errors or forgetting memory errors, resulting in difficulty in automatically generating a review plan according to the memory situation.
[0003] With the development of artificial intelligence and deep learning, the present invention provides a method for quickly testing memory results based on AI learning to overcome the limitations of traditional memory detection and achieve more accurate, efficient and personalized memory evaluation and optimization. Summary of the Invention
[0004] The present invention provides a method for quickly testing memory results based on AI learning to solve the technical problems in the above background art.
[0005] The present invention provides a method for quickly testing memory results based on AI learning, including the following steps: Step S101, randomly select A questions from the basic question bank according to the user's age and educational level to generate a preliminary test questionnaire, and record the user's answering time and the number of correct answers; The fields of the data table corresponding to the basic question bank include: question type, question content, question answer, maximum applicable age, minimum applicable age, applicable educational level and difficulty score; Where A is a custom parameter; Step S102, calculate a preliminary test score based on the user's answering time and the number of correct answers, and automatically generate a memory question bank according to the preliminary test score; The value range of the preliminary test score is between 0 and 1; Step S103, randomly select B questions from the memory question bank to generate a memory test questionnaire, and record the user's memory situation; The memory situation includes memory errors and memory correctness; where B is a custom parameter; Step S104: Generate a memory network diagram based on the user's memory situation, and analyze the memory network diagram through a memory detection model to obtain a memory error set; The memory network diagram is represented by a triple in the form of (knowledge point, connection relationship, attribute). Among them, the knowledge point is represented by the combination of the question content and the question answer, and the attribute is represented by the combination of the word vector of the keyword label corresponding to the knowledge point and the user's memory situation of the knowledge point; Each element of the memory error set is represented by a binary tuple in the form of (knowledge point, error reason). Among them, the error reason includes conceptual understanding errors and forgetting memory errors; Step S105: Automatically generate a review strategy based on the memory error set.
[0006] Furthermore, construct a basic question bank through public education data sets, large language models, and manual input. The question types include: single-choice questions, fill-in-the-blank questions, and essay questions. The applicable education levels include: primary school, junior high school, senior high school, and university. The value ranges of the difficulty score and the preliminary test score are the same, and the difficulty score is obtained by inputting the question content and the question answer into the large language model.
[0007] Furthermore, the calculation formula for the preliminary test score Score is as follows: ; where and represent the number of correct answers and the total number of answers in the preliminary test respectively, and represent the answering duration and the maximum allowed answering duration of the preliminary test respectively, and represent the custom first weight coefficient and second weight coefficient respectively, and and The sum value of is 1.
[0008] Furthermore, select the questions that meet the screening conditions from the data table corresponding to the basic question bank according to the preliminary test score as the memory question bank. The screening conditions are that the difficulty score ≥ preliminary test score - C and the difficulty score ≤ preliminary test score + C, where C is a custom parameter.
[0009] Furthermore, generating a memory network diagram according to the user's memory situation includes the following steps: Step S201: Input the question content and the question answer into the large language model, and extract the keyword labels of the knowledge points through the large language model; The keyword label of each knowledge point does not exceed D, where D is a custom parameter; Step S202: Convert the keyword tags into word vector representations through the word vector model, binarize the user's memory of knowledge points, and combine the two as the attributes of the knowledge points. Step S203: Use the keyword tags of one knowledge point as the source tag set, and the keyword tags of another knowledge point as the target tag set. Calculate the correlation coefficient between the source tag set and the target tag set, and if the correlation coefficient is greater than or equal to the preset threshold, establish a connection relationship between the two knowledge points. The preset threshold is a custom parameter.
[0010] Furthermore, the calculation formula for the correlation coefficient Relation is as follows: ; ; ; where represents the first correlation coefficient, represents the second correlation coefficient, 1 ≤ i ≤ m, m represents the number of keyword tags in the source tag set, 1 ≤ j ≤ n, n represents the number of keyword tags in the target tag set, d represents the dimension number of the word vectors corresponding to the keyword tags, and respectively represent the k-th dimension value and the average dimension value of the word vector corresponding to the i-th keyword tag in the source tag set, and respectively represent the k-th dimension value and the average dimension value of the word vector corresponding to the j-th keyword tag in the target tag set, represents the target tag set, and respectively represent the custom third weight coefficient and fourth weight coefficient, and and The sum of the values is 1.
[0011] Furthermore, the memory detection model includes a hidden layer, a classifier, and an output layer; The hidden layer is used to update the word vectors of all keyword tags corresponding to the knowledge points in the memory network diagram to obtain updated vectors; Input the updated vectors corresponding to all the knowledge points with memory errors into the classifier, and the classification space of the classifier represents the reasons for the errors; The output layer outputs all the knowledge points with memory errors and the reasons for the errors.
[0012] Furthermore, the calculation formula of the hidden layer includes: ; ; ; where \(1\leq u\leq M\), and \(M\) represents the number of all knowledge points in the memory network graph, represents the set of knowledge points having a connection relationship with the \(u\)-th knowledge point, represents the number of knowledge points having a connection relationship with the \(u\)-th knowledge point, represents the update vector of the \(u\)-th knowledge point, and respectively represent the word vectors of the keyword tags corresponding to the \(u\)-th and \(v\)-th knowledge points, represents the retention probability of the \(u\)-th knowledge point, represents the transition probability from the \(v\)-th knowledge point to the \(u\)-th knowledge point, \(W\) represents the weight parameter, and \(b\) represents the bias parameter, represents the first perceptron, represents the second perceptron, COMBINE represents the splicing function, and Swish represents the Swish activation function.
[0013] Furthermore, a review strategy is automatically generated according to the memory error set, including: increasing the detection frequency for the knowledge points with forgetting memory errors; for the knowledge points with conceptual understanding errors, in addition to increasing the detection frequency, also increasing the detection frequency of the knowledge points having a connection relationship with them.
[0014] The beneficial effects of the present invention are as follows: The present invention can automatically generate a memory question bank according to the user's cognitive level, so as to achieve adaptive adjustment of the test difficulty, and the memory detection model provided by the present invention can automatically analyze the error causes of knowledge points, eliminate the subjective differences of manual evaluation, and automatically generate a review plan according to the error causes, thereby improving the efficiency of the user's mastery of knowledge points. BRIEF DESCRIPTION OF THE DRAWINGS
[0015] Figure 1 is a flowchart of the method for quickly verifying memory results based on AI learning of the present invention; Figure 2 is a flowchart of generating a memory network graph according to the user's memory situation of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS
[0016] Now, the subject matter described herein will be discussed with reference to example embodiments. It should be understood that discussing these embodiments is only to enable those skilled in the art to better understand and thus implement the subject matter described herein, and the functions and arrangements of the elements discussed can be changed without departing from the protection scope of the content of this specification. Each example can omit, substitute, or add various processes or components as needed. Additionally, the features described for some examples can also be combined in other examples.
[0017] It should be noted that, unless otherwise defined, the technical terms or scientific terms used in one or more embodiments of the present invention should have the ordinary meaning understood by those of ordinary skill in the art to which the present invention pertains. The "first", "second" and similar terms used in one or more embodiments of the present invention do not denote any order, quantity or importance, but are only used to distinguish different components. Words such as "including" or "comprising" mean that the elements or objects appearing before this word cover the elements or objects listed after this word and their equivalents, without excluding other elements or objects. Words such as "connected" or "coupled" are not limited to physical or mechanical connections, but may include electrical connections, whether direct or indirect. "Upper", "lower", "left", "right", etc. are only used to represent relative positional relationships, and when the absolute position of the object being described changes, the relative positional relationship may also change accordingly.
[0018] As Figures 1 to 2 shown, a method for testing the results of rapid memory based on AI learning includes the following steps: Step S101, randomly select A questions from the basic question bank according to the user's age and educational level to generate a preliminary test questionnaire, and record the user's answering time and the number of correct answers; The fields of the data table corresponding to the basic question bank include: question type, question content, question answer, maximum applicable age, minimum applicable age, applicable educational level, and difficulty score; Where A is a custom parameter, preferably, A is set to 20; Step S102, calculate a preliminary test score based on the user's answering time and the number of correct answers, and automatically generate a memory question bank according to the preliminary test score; The value range of the preliminary test score is between 0 and 1; Step S103, randomly select B questions from the memory question bank to generate a memory test questionnaire, and record the user's memory situation; The memory situation includes memory errors and correct memories; Where B is a custom parameter, preferably, B is set to 30; Step S104, generate a memory network diagram according to the user's memory situation, and analyze the memory network diagram through a memory detection model to obtain a memory error set; The memory network diagram is represented by a triple in the form of (knowledge point, connection relationship, attribute), where the knowledge point is represented by the combination of the question content and the question answer, and the attribute is represented by the combination of the word vector of the keyword label corresponding to the knowledge point and the user's memory situation of the knowledge point; Each element of the memory error set is represented by a binary tuple in the form of (knowledge point, error reason), where the error reason includes conceptual understanding errors and forgetting memory errors; Step S105, automatically generate a review strategy according to the memory error set.
[0019] In an embodiment of the present invention, a basic question bank is constructed by means of a publicly available education dataset, a large language model, and manual input. The question types include: single-choice questions, fill-in-the-blank questions, and essay questions. The applicable education levels include: primary school, junior high school, senior high school, and university. The value ranges of the difficulty score and the preliminary test score are the same, and the difficulty score is obtained by inputting the question content and the question answer into the large language model, or can also be obtained by manual annotation.
[0020] It should be noted that the publicly available education dataset can be Math23K (a mathematics question bank dataset released by Tencent AI Lab), Exam-Question-Bank-Dataset-zh (a general exam question bank dataset, including various question types such as multiple-choice questions, fill-in-the-blank questions, and short-answer questions), etc. The large language model can be ChatGPT, DeepSeek, etc. The data table corresponding to the basic question bank can be stored in a MySQL database. Then, the query statement for randomly extracting A questions from the basic question bank to generate a preliminary test questionnaire according to the user's age and education level is as follows: "SELECT test_type, test_content, test_answer FROM question_data WHERE age_min >= user age AND age_max <= user age AND education_level = 'user education level' ORDER BY RAND() LIMIT A;", where question_data represents the table name of the data table corresponding to the basic question bank, test_type represents the question type, test_content represents the question content, test_answer represents the question answer, age_min represents the minimum applicable age, age_max represents the maximum applicable age, and education_level represents the applicable education level.
[0021] In an embodiment of the present invention, the calculation formula for the preliminary test score Score is as follows: ; where and respectively represent the number of correct answers and the total number of answers in the preliminary test, and respectively represent the answering duration and the maximum allowed answering duration of the preliminary test, and respectively represent the custom first weight coefficient and second weight coefficient, and and The sum value of is set to 0.7, and
[0022]
[0023]
[0024]
[0025] Figure 2 For example, if the number of correct answers is 18, the total number of questions is 20, the answering time is 15 minutes, the maximum allowed answering time is 30 minutes, and the first weight coefficient and the second weight coefficient are 0.7 and 0.3 respectively, then the preliminary test score Score = 0.7×0.9 + 0.3×0.5 = 0.78. Step S201, input the question content and the question answer into the large language model, and extract the keyword tags of the knowledge points through the large language model; The keyword tags of each knowledge point cannot exceed D, where D is a custom parameter, and preferably, D is set to 5; Step S202, convert the keyword tags into word vector representations through the word vector model, and binarize the user's memory situation of the knowledge points. The two are combined as the attributes of the knowledge points; The word vector model can be Word2Vec, FastText, etc., which will not be elaborated here; Step S203: Use the keyword tags of one knowledge point as the source tag set, and the keyword tags of another knowledge point as the target tag set. Calculate the correlation coefficient between the source tag set and the target tag set, and if the correlation coefficient is greater than or equal to the preset threshold, a connection relationship is constructed between these two knowledge points. The preset threshold is a custom parameter. Preferably, the preset threshold is set to 0.7.
[0026] In an embodiment of the present invention, the calculation formula of the correlation coefficient Relation is as follows: ; ; ; where represents the first correlation coefficient, represents the second correlation coefficient, 1 ≤ i ≤ m, m represents the number of keyword tags in the source tag set, 1 ≤ j ≤ n, n represents the number of keyword tags in the target tag set, d represents the number of dimensions of the word vector corresponding to the keyword tag, and respectively represent the k-th dimension value and the dimension average value of the word vector corresponding to the i-th keyword tag in the source tag set, and respectively represent the k-th dimension value and the dimension average value of the word vector corresponding to the j-th keyword tag in the target tag set, represents the target tag set, and respectively represent the custom third weight coefficient and the fourth weight coefficient, and and the sum value of is 1. Preferably, is set to 0.6, is set to 0.4.
[0027] It should be noted that before calculating the first correlation coefficient and the second correlation coefficient of the word vector corresponding to the keyword tag, it is necessary to ensure that the number of dimensions of the word vector corresponding to each keyword tag is the same. If not, zero values are filled at the end of the shorter word vector until the number of dimensions is the same as that of the longer word vector.
[0028] In an embodiment of the present invention, the memory detection model includes a hidden layer, a classifier, and an output layer; The hidden layer is used to update the word vectors of all keyword tags corresponding to the knowledge points in the memory network diagram to obtain updated vectors; Input the updated vectors corresponding to all the knowledge points with memory errors into the classifier, and the classification space of the classifier represents the reasons for the errors; The output layer outputs all the knowledge points with memory errors and the reasons for the errors.
[0029] In one embodiment of the present invention, the calculation formula of the hidden layer includes: ; ; ; where 1 ≤ u ≤ M, M represents the number of all knowledge points in the memory network diagram, represents the set of knowledge points having a connection relationship with the u-th knowledge point, represents the number of knowledge points having a connection relationship with the u-th knowledge point, represents the update vector of the u-th knowledge point, and respectively represent the word vectors of the keyword tags corresponding to the u-th and v-th knowledge points, represents the retention probability of the u-th knowledge point, represents the transition probability from the v-th knowledge point to the u-th knowledge point, W represents the weight parameter, b represents the bias parameter, represents the first perceptron, represents the second perceptron, COMBINE represents the splicing function, and Swish represents the Swish activation function.
[0030] It should be noted that the weight parameters and bias parameters in the memory detection model are all learnable hyperparameters, the activation function corresponding to the classifier is the softmax activation function. Through the self-retention mechanism, it can ensure that its own information will not be overly diluted when aggregating neighboring knowledge points. Through the state transition mechanism, it can calculate the state transition probability between knowledge points, and can clarify the dependence relationship and information transmission relationship between adjacent knowledge points, which helps to better understand the reasons for the errors of knowledge points, thereby improving the detection accuracy of the memory detection model.
[0031] In one embodiment of the present invention, a review strategy is automatically generated according to the memory error set, including: increasing the detection frequency for the knowledge points with forgetting memory errors; for the knowledge points with conceptual understanding errors, in addition to increasing the detection frequency, also increasing the detection frequency of the knowledge points having a connection relationship with them.
[0032] The above has described the embodiments of this embodiment, but this embodiment is not limited to the above specific implementation manners. The above specific implementation manners are only illustrative and not restrictive. Under the inspiration of this embodiment, those of ordinary skill in the art can also make many forms, all of which fall within the protection scope of this embodiment.
Claims
1. A method for quickly testing the results of AI-based learning, characterized in that It includes the following steps: Step S101: Randomly select A questions from the basic question bank according to the user's age and education level to generate a preliminary test questionnaire, and record the user's answering time and the number of correct answers; The fields of the data table corresponding to the basic question bank include: question type, question content, question answer, maximum applicable age, minimum applicable age, applicable education level, and difficulty score; where A is a custom parameter; Step S102: Calculate the preliminary test score based on the user's answering time and the number of correct answers, and automatically generate a memory question bank according to the preliminary test score; The value range of the preliminary test score is between 0 and 1; Step S103: Randomly select B questions from the memory question bank to generate a memory test questionnaire, and record the user's memory situation; The memory situation includes memory errors and correct memories; where B is a custom parameter; Step S104: Generate a memory network diagram according to the user's memory situation, and analyze the memory network diagram through a memory detection model to obtain a memory error set; The memory network diagram is represented by a triple in the form of (knowledge point, connection relationship, attribute), where the knowledge point is represented by the combination of the question content and the question answer, and the attribute is represented by the combination of the word vector of the keyword label corresponding to the knowledge point and the user's memory situation of the knowledge point; Each element of the memory error set is represented by a binary tuple in the form of (knowledge point, error reason), where the error reason includes conceptual understanding errors and forgetting memory errors; Step S105: Automatically generate a review strategy according to the memory error set.
2. The method for quickly verifying the results of memorization based on AI learning according to claim 1, wherein The basic question bank is constructed through public education data sets, large language models, and manual input. The question types include: single-choice questions, fill-in-the-blank questions, and essay questions. The applicable education levels include: primary school, junior high school, senior high school, and university. The value range of the difficulty score is the same as that of the preliminary test score, and the difficulty score is obtained by inputting the question content and the question answer into the large language model.
3. The method for quickly verifying the results of memorization based on AI learning according to claim 1, characterized in that, The calculation formula for the preliminary test score Score is as follows: ； where and represent the number of correct answers and the total number of questions in the preliminary test respectively, and represent the answering duration and the maximum allowed answering duration of the preliminary test respectively, and represent the first weight coefficient and the second weight coefficient defined by oneself respectively, and and the sum value of them is 1.
4. The method for quickly verifying the results of memorization based on AI learning according to claim 1, wherein, Select the questions that meet the screening conditions from the data table corresponding to the basic question bank as the memory question bank according to the preliminary test score. The screening conditions are that the difficulty score ≥ preliminary test score - C and the difficulty score ≤ preliminary test score + C, where C is a custom parameter.
5. The method for quickly verifying the results of memorization based on AI learning according to claim 1, characterized in that, Generating a memory network diagram according to the user's memory situation includes the following steps: Step S201: Input the question content and the question answer into the large language model, and extract the keyword labels of the knowledge points through the large language model; The keyword labels of each knowledge point do not exceed D, where D is a custom parameter; Step S202: Convert the keyword labels into word vector representations through a word vector model, and binarize the user's memory situation of the knowledge points. The two are combined as the attribute of the knowledge point; Step S203: Use the keyword labels of one knowledge point as the source label set, and the keyword labels of another knowledge point as the target label set. Calculate the correlation coefficient between the source label set and the target label set, and if the correlation coefficient is greater than or equal to the preset threshold, a connection relationship is established between these two knowledge points; where the preset threshold is a custom parameter.
6. The method for quickly verifying the results of memorization based on AI learning according to claim 5, wherein The calculation formula for the correlation coefficient Relation is as follows: ； ； ； Among them represents the first association coefficient represents the second association coefficient, where 1 ≤ i ≤ m, m represents the number of keyword tags in the source tag set, 1 ≤ j ≤ n, n represents the number of keyword tags in the target tag set, and d represents the dimensionality of the word vectors corresponding to the keyword tags and respectively represent the k-th dimensional value and the dimensional average value of the word vector corresponding to the i-th keyword tag in the source tag set and respectively represent the k-th dimensional value and the dimensional average value of the word vector corresponding to the j-th keyword tag in the target tag set represents the target tag set and respectively represent the user-defined third weight coefficient and fourth weight coefficient, and and The sum value of is 1 7. The method for quickly verifying the results of memorization based on AI learning according to claim 1, wherein The memory detection model includes a hidden layer, a classifier, and an output layer; The hidden layer is used to update the word vectors of the keyword tags corresponding to all knowledge points in the memory network diagram to obtain updated vectors; The updated vectors corresponding to all the knowledge points with memory errors are input into the classifier, and the classification space of the classifier represents the reasons for the errors; The output layer outputs all the knowledge points with memory errors and the reasons for the errors.
8. The method for quickly verifying the results of memorization based on AI learning according to claim 7, wherein The calculation formula of the hidden layer includes: ； ； ； where \(1\leq u\leq M\), and \(M\) represents the number of all knowledge points in the memory network graph. represents the set of knowledge points that have a connection relationship with the \(u\)-th knowledge point. represents the number of knowledge points that have a connection relationship with the \(u\)-th knowledge point. represents the update vector of the \(u\)-th knowledge point. and represent the word vectors of the keyword tags corresponding to the \(u\)-th and \(v\)-th knowledge points respectively. represents the retention probability of the \(u\)-th knowledge point. represents the transition probability from the \(v\)-th knowledge point to the \(u\)-th knowledge point, \(W\) represents the weight parameter, and \(b\) represents the bias parameter. represents the first perceptron. represents the second perceptron, COMBINE represents the concatenation function, and Swish represents the Swish activation function.
9. The method for testing the results of rapid memorization based on AI learning according to claim 1, characterized in that, Automatically generate a review strategy according to the memory error set, including: increasing the detection frequency for the knowledge points with forgetting memory errors; for the knowledge points with conceptual understanding errors, in addition to increasing the detection frequency, also increase the detection frequency of the knowledge points that have a connection relationship with them.