A personalized ability-based question recommendation method and system
By constructing correct and incorrect question banks, obtaining question vector representations of students' historical answer sequences, extracting knowledge points and question difficulty features, and using an ability status analysis model to conduct personalized ability analysis, the problem of inaccurate ability analysis and lack of individual adaptability in existing technologies is solved, and accurate recommendations for personalized teaching are realized.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- SOUTHWEST UNIV
- Filing Date
- 2022-07-15
- Publication Date
- 2026-06-12
AI Technical Summary
Existing methods for analyzing student abilities lack consideration for the rationality of ability evolution, resulting in inaccurate analysis results and a lack of individual adaptability.
By constructing correct and incorrect question banks, we obtain the question vector representation of students' historical answer sequences, extract the difficulty of knowledge points, question difficulty, and learning ability characteristics, use the ability status analysis model to conduct personalized ability analysis, and recommend matching questions based on the analysis results.
It enables personalized ability analysis based on students' historical answer sequences, improves individual adaptability, and can more accurately reflect the differences in learning abilities among different students, thus meeting the needs of personalized teaching.
Smart Images
Figure CN115329190B_ABST
Abstract
Description
Technical Field
[0001] This application relates to the field of intelligent analysis applications, and in particular to a method and system for recommending questions based on personalized abilities. Background Technology
[0002] In online learning systems, students train their skills by answering questions on specific knowledge points provided by the system. Existing knowledge tracing methods can capture students' skill levels based on their historical answer records and predict their mastery of given questions. Current learning analysis methods have evolved from traditional personalized Bayesian knowledge tracing models to deep knowledge tracing methods using improved Bayesian neural networks, and even to complex neural network algorithms for analyzing student abilities. However, these previous student ability analysis methods lack strong interpretability and fail to consider the rationality of ability evolution. Previous knowledge tracing methods assumed that students would only improve their abilities if they answered questions correctly, which is inaccurate from an educational theory perspective. Even if students answer questions incorrectly, they can still gain some skill improvement from the questions. Furthermore, previous models did not adequately characterize student abilities; answering a single question correctly or incorrectly could cause significant fluctuations in a student's ability. Summary of the Invention
[0003] In view of the problems existing in the prior art, this application proposes a question recommendation method and system based on personalized ability, which mainly solves the problem that the existing student ability analysis methods lack consideration of the rationality of ability evolution, resulting in inaccurate analysis results and a lack of individual adaptability in the analysis.
[0004] To achieve the above and other objectives, the technical solution adopted in this application is as follows.
[0005] This application provides a question recommendation method based on personalized ability, including:
[0006] Construct a correct question bank and an incorrect question bank based on a preset question bank, and obtain the question vector representation corresponding to each question in the correct question bank and the incorrect question bank respectively;
[0007] Obtain the historical answer sequence of the same student, obtain the question vector representation of each question in the historical answer sequence from the corresponding question bank, obtain the question vector set corresponding to the historical answer sequence, perform feature extraction on the question vector set, and obtain the knowledge point difficulty feature, question difficulty feature and learning ability feature of the corresponding student;
[0008] The difficulty characteristics of the knowledge points, the difficulty characteristics of the questions, and the learning ability characteristics are input into a preset ability status analysis model to obtain the ability status of students at each time point.
[0009] Based on the student's ability status at each time point, a matching question is retrieved from the question bank and pushed to the corresponding student's end.
[0010] In one embodiment of this application, a correct question bank and an incorrect question bank are constructed based on a preset question bank, and a question vector representation corresponding to each question in the correct question bank and the incorrect question bank is obtained respectively, including:
[0011] Obtain the question stems and corresponding correct answers for all questions in the question bank;
[0012] Based on the question stem and the corresponding correct answer, mark each question in the question bank as a correct question and an incorrect question;
[0013] The correct questions are stored in a preset correct question bank as positive samples of the corresponding questions, and the incorrect questions are stored in the incorrect question bank as negative samples of the corresponding questions;
[0014] The positive sample is input into a pre-trained contrastive learning model to obtain the question vector representation corresponding to the positive sample;
[0015] The negative sample is input into the contrastive learning model to obtain the question vector representation corresponding to the negative sample.
[0016] In one embodiment of this application, obtaining the historical answer sequence of the same student includes:
[0017] Obtain student identification;
[0018] The current learning progress of the corresponding student is determined based on the identity identifier;
[0019] Determine the relevant knowledge points based on the current learning progress;
[0020] Based on the pre-defined correspondence between knowledge points and questions, determine the set of questions corresponding to the associated knowledge points;
[0021] The question set is filtered based on the identity identifier to determine all the question-answering records of the corresponding student in the question set, and the historical answer sequence is generated according to the time sequence of the question-answering records.
[0022] In one embodiment of this application, the question vector representation corresponding to each question in the historical answer sequence is obtained from the corresponding question bank to obtain the question vector set corresponding to the historical answer sequence, including:
[0023] If a student answers a question correctly in the historical question-answering sequence, the question vector representation of the correctly answered question is obtained from the correct question bank and entered into the question vector set;
[0024] If a student answers a question incorrectly in the historical question-answering sequence, the question vector representation of the incorrectly answered question is obtained from the error question bank and entered into the question vector set.
[0025] In one embodiment of this application, feature extraction is performed on the set of question vectors to obtain the knowledge point difficulty features, question difficulty features, and learning ability features of the corresponding student, including:
[0026] Based on the preset initial difficulty value of each question and the correspondence between the questions and knowledge points, determine the initial difficulty value of the corresponding knowledge point for each question;
[0027] Based on the question vector representation corresponding to each time node in the question vector set, predict the probability that the student will answer the next question correctly at the next time node. Update the question difficulty of the corresponding question based on the probability and the initial value of the question difficulty to obtain the question difficulty feature.
[0028] Based on the question vector representation of all questions corresponding to each knowledge point in the question vector set, determine the knowledge point difficulty characteristics of the corresponding knowledge point;
[0029] Based on the question vector representation of students for the same type of questions, the learning ability characteristics of the corresponding students are determined. The knowledge point difficulty characteristic is used to represent the degree of students' mastery of the same knowledge point, the question difficulty characteristic is used to represent the degree of students' mastery of the same question, and the learning ability characteristic is used to represent the students' mastery of questions corresponding to the same knowledge point.
[0030] In one embodiment of this application, the difficulty characteristics of the knowledge points, the difficulty characteristics of the questions, and the learning ability characteristics are input into a preset ability status analysis model to obtain the student's ability status at each time point, including:
[0031] The difficulty characteristics of the knowledge points, the difficulty characteristics of the questions, and the learning ability characteristics are input into a preset ability status analysis model to obtain a set of ability statuses for the corresponding students. The set of ability statuses includes ability statuses at different time points, and the ability status analysis model includes a project response theory model.
[0032] In one embodiment of this application, after inputting the knowledge point difficulty features, question difficulty features, and learning ability features into a preset ability status analysis model to obtain the student's ability status at each time point, the method further includes:
[0033] By establishing a first constraint relationship between the capability states corresponding to adjacent time nodes through comparative learning, the capability state changes at adjacent time nodes are kept within a preset range.
[0034] By establishing a second constraint relationship through comparative learning, different students at the same point in time can be made to have different ability states at the same point in time.
[0035] In one embodiment of this application, retrieving matching questions from the question bank based on the student's ability status at each time point and pushing them to the corresponding student's end includes:
[0036] The learning ability status is compared with a preset ability status threshold, and the learning ability status below the ability status threshold is selected as the weak ability status.
[0037] Based on the stated weakness in ability, questions of matching difficulty are extracted from a pre-set question bank for question recommendation.
[0038] This application also provides a question recommendation system based on personalized ability, including:
[0039] The vector representation module is used to construct a correct question bank and an incorrect question bank based on a preset question bank, and to obtain the question vector representation corresponding to each question in the correct question bank and the incorrect question bank, respectively.
[0040] The student feature extraction module is used to obtain the historical answer sequence of the same student, obtain the question vector representation of each question in the historical answer sequence from the corresponding question bank, obtain the question vector set corresponding to the historical answer sequence, and perform feature extraction on the question vector set to obtain the knowledge point difficulty feature, question difficulty feature and learning ability feature of the corresponding student.
[0041] The ability status analysis module is used to input the knowledge point difficulty characteristics, question difficulty characteristics and learning ability characteristics into a preset ability status analysis model to obtain the student's ability status at each time point.
[0042] The question recommendation module is used to retrieve matching questions from the question bank based on the student's ability status at each time point and push them to the corresponding student's end.
[0043] As described above, the question recommendation method and system based on personalized ability proposed in this application have the following beneficial effects.
[0044] This application can perform personalized ability analysis based on students' historical answer sequences, and recommend questions based on the differences in learning abilities among different students, thereby improving individual adaptability. In addition, this application comprehensively analyzes students' ability status through three dimensions: the difficulty characteristics of knowledge points, the difficulty characteristics of questions, and the learning ability of the target audience. This can effectively reflect the personalized differences in the difficulty of different questions or knowledge points relative to different students, realize the matching recommendation of questions based on individual ability differences, and better meet the needs of personalized teaching. Attached Figure Description
[0045] Figure 1 This is a flowchart illustrating a personalized question recommendation method according to one embodiment of this application.
[0046] Figure 2 This is a block diagram of a question recommendation system based on personalized capabilities in one embodiment of this application.
[0047] Figure 3 This is a schematic diagram of the data on the evolution of a student's abilities during practice, as shown in one embodiment of this application.
[0048] Figure 4 This is a schematic diagram of the overall architecture of a question recommendation method based on personalized ability in one embodiment of this application. Detailed Implementation
[0049] The following specific examples illustrate the implementation of this application. Those skilled in the art can easily understand other advantages and effects of this application from the content disclosed in this specification. This application can also be implemented or applied through other different specific embodiments, and various details in this specification can also be modified or changed based on different viewpoints and applications without departing from the spirit of this application. It should be noted that, unless otherwise specified, the following embodiments and features in the embodiments can be combined with each other.
[0050] It should be noted that the illustrations provided in the following embodiments are only schematic representations of the basic concept of this application. Therefore, the drawings only show the components related to this application and are not drawn according to the actual number, shape and size of the components in the actual implementation. In the actual implementation, the form, quantity and proportion of each component can be arbitrarily changed, and the layout of the components may also be more complex.
[0051] Please see Figure 1 This application provides a method for recommending questions based on personalized learning abilities, which includes the following steps.
[0052] Step S01: Construct a correct question bank and an incorrect question bank based on a preset question bank, and obtain the question vector representation corresponding to each question in the correct question bank and the incorrect question bank respectively.
[0053] In one embodiment, this application primarily targets online learning systems. Based on existing question banks and knowledge point databases, and combined with students' own learning progress, it captures students' abilities during online learning exercises and quantitatively analyzes the evolution of these abilities throughout the entire knowledge point learning process. Once the student's ability status is understood, targeted reinforcement can be provided for weaknesses in specific knowledge points, enabling personalized teaching. The question recommendation method in this application primarily relies on knowledge tracking, i.e., capturing students' ability status based on their historical question-answering information. Existing question banks are divided based on knowledge points, with each knowledge point corresponding to numerous questions. The set of all exercises in the online learning system can be represented as E={e1, e2, e3, …, en}. The set of all concepts in the system can be represented as C={c1, c2, c3, …, c m The knowledge points contained in all questions within the system can be represented as r(e1) = {c1, c2}, r(e2) = {c2, c3}, r(e3) = {c4}. This allows us to obtain the correspondence between each question and the knowledge points in the question bank.
[0054] In one embodiment, a correct question bank and an incorrect question bank are constructed based on a preset question bank, and a question vector representation corresponding to each question in the correct question bank and the incorrect question bank is obtained respectively, including:
[0055] Obtain the question stems and corresponding correct answers for all questions in the question bank;
[0056] Based on the question stem and the corresponding correct answer, mark each question in the question bank as a correct question and an incorrect question;
[0057] The correct questions are stored in a preset correct question bank as positive samples of the corresponding questions, and the incorrect questions are stored in the incorrect question bank as negative samples of the corresponding questions;
[0058] The positive sample is input into a pre-trained contrastive learning model to obtain the question vector representation corresponding to the positive sample;
[0059] The negative sample is input into the contrastive learning model to obtain the question vector representation corresponding to the negative sample.
[0060] In one embodiment, taking multiple-choice questions as an example, if each question in the question bank corresponds to four options, with only one option being the correct answer and the other three being incorrect answers, after obtaining the question stem and corresponding correct options for each question in the question bank, the corresponding questions can be labeled based on the correct and incorrect options. If the correct option is selected, the corresponding question is labeled as a correct question; if the incorrect option is selected, the corresponding question is labeled as an incorrect question. This divides the original question bank into two question banks: a correct question bank and an incorrect question bank. Taking the questions in the original question bank as the original samples, the questions in the correct question bank are positive samples of the original samples, and the questions in the incorrect question bank are negative samples of the original samples. Contrastive learning can be used to narrow the semantic distance between the original samples and positive samples, and widen the semantic distance between the original samples and negative samples, thereby obtaining the question vector representation of each question in the correct question bank. This question vector contains the semantic information of the positive samples, which can be measured by the similarity between the original samples and positive samples. The specific semantic representation is not limited here. Simultaneously, through contrastive learning, a question vector representation can be obtained for each question in the error question bank. This question vector identifier contains semantic information of the negative sample, and this semantic information can be measured by the similarity between the original sample and the negative sample. Through the contrastive learning model, the corresponding question vector representation contains higher-order features such as the correctness of the question, ensuring that even incorrect answers can positively impact subsequent ability analysis, making the student's ability analysis more rational. Specifically, the loss function of the contrastive learning model can be constructed with the goal of maximizing the similarity between the positive sample and the original sample; the specific form of the function is not restricted here.
[0061] Step S02: Obtain the historical answer sequence of the same student, obtain the question vector representation of each question in the historical answer sequence from the corresponding question bank, obtain the question vector set corresponding to the historical answer sequence, and perform feature extraction on the question vector set to obtain the knowledge point difficulty feature, question difficulty feature and learning ability feature of the corresponding student.
[0062] In one embodiment, obtaining the historical answer sequence of the same student includes:
[0063] Obtain student identification;
[0064] The current learning progress of the corresponding student is determined based on the identity identifier;
[0065] Determine the relevant knowledge points based on the current learning progress;
[0066] Based on the pre-defined correspondence between knowledge points and questions, determine the set of questions corresponding to the associated knowledge points;
[0067] The question set is filtered based on the identity identifier to determine all the question-answering records of the corresponding student in the question set, and the historical answer sequence is generated according to the time sequence of the question-answering records.
[0068] In one embodiment, each time a student logs into the online learning system, the system records the student's identity information, such as account ID, name, and nickname. After completing each exercise, the system can associate the identity information with the completed questions. The system can retrieve all historical exercise records of the target student based on their identity information, or it can retrieve exercise records based on the target student's current learning progress. The system retrieves the question stem and correct answer for each question. It also retrieves all historical exercise records of the current student from the online learning system, including whether the answers were correct or not.
[0069] In one embodiment, when retrieving practice records based on the target object's current learning progress, the learning progress corresponds to the learning status of the current knowledge point. The system can pre-set different sets of knowledge points for different learning progresses, establishing a correlation between learning progress and knowledge points. For example, knowledge points can be divided by chapter. Based on the current learning progress and matching knowledge points, the set of questions for the current learning progress can be further determined according to the correspondence between knowledge points and questions. This yields a set of practice records used for personalized ability analysis. After obtaining the student's practice records, a corresponding historical answer sequence can be generated based on the student's practice sequence. This historical answer sequence corresponds to the student's order of answering questions. Since the questions pushed to each student by the online learning system differ, each student's historical answer sequence has personalized differences.
[0070] In one embodiment, the question vector representation corresponding to each question in the historical answer sequence is obtained from the corresponding question bank to obtain the question vector set corresponding to the historical answer sequence, including:
[0071] If a student answers a question correctly in the historical question-answering sequence, the question vector representation of the correctly answered question is obtained from the correct question bank and entered into the question vector set;
[0072] If a student answers a question incorrectly in the historical question-answering sequence, the question vector representation of the incorrectly answered question is obtained from the error question bank and entered into the question vector set.
[0073] In one embodiment, since the student's historical answer sequence records the student's correct or incorrect answers to different questions, a time-series set of question vectors can be generated based on the answer sequence. If the student answers a question correctly at a certain historical time point, the question vector representation corresponding to the question answered at that time point is selected from the correct question bank and recorded in the sequence corresponding to that time point. Similarly, if the student answers a question incorrectly at a certain historical time point, the question vector representation corresponding to the question answered at that time point is selected from the incorrect question bank and recorded in the sequence position of the time point corresponding to the incorrect question. In this way, a set of question vectors consisting of the question vector representations corresponding to all questions in the student's entire historical answer sequence is obtained.
[0074] In one embodiment, since students may answer the same question multiple times at different points in time during the problem-solving process, the same target object will have different problem-solving records for the same question at different points in time. Each question can be represented by a vector of length m. Furthermore, since each question may contain multiple knowledge points, these knowledge points can be encoded based on the time points.
[0075] In one embodiment, feature extraction is performed on the set of question vectors to obtain the knowledge point difficulty features, question difficulty features, and learning ability features of the corresponding student, including:
[0076] Based on the preset initial difficulty value of each question and the correspondence between the questions and knowledge points, determine the initial difficulty value of the corresponding knowledge point for each question;
[0077] Based on the question vector representation corresponding to each time node in the question vector set, predict the probability that the student will answer the next question correctly at the next time node. Update the question difficulty of the corresponding question based on the probability and the initial value of the question difficulty to obtain the question difficulty feature.
[0078] Based on the question vector representation of all questions corresponding to each knowledge point in the question vector set, determine the knowledge point difficulty characteristics of the corresponding knowledge point;
[0079] Based on the question vector representation of students for the same type of questions, the learning ability characteristics of the corresponding students are determined. The knowledge point difficulty characteristic is used to represent the degree of students' mastery of the same knowledge point, the question difficulty characteristic is used to represent the degree of students' mastery of the same question, and the learning ability characteristic is used to represent the students' mastery of questions corresponding to the same knowledge point.
[0080] In one embodiment, three feature extractors can be set up to extract knowledge point difficulty features, question difficulty features, and learning ability features from the question vector set, respectively. Each feature extractor can employ a transformer model, using the transformer model's encoder and decoder to convert the question vector representation into a corresponding feature vector. The question vector representation at each time point is input into the transformer model to dynamically update the corresponding features, where the question vector representation contains the correct and incorrect semantic information of the corresponding question. The specific architecture and implementation process of the transformer are well known to those skilled in the art and will not be described in detail here. Since each knowledge point can be contained in multiple questions, the difficulty information of that knowledge point can be determined based on students' feedback on multiple questions containing the same knowledge point, and then the difficulty information of each knowledge point for different students can be updated. An initial value for the difficulty of the knowledge point can be preset, and the knowledge point difficulty information of each knowledge point relative to the corresponding student can be dynamically updated based on the question vector representation corresponding to the historical answer sequence. Since there are individual differences in the historical answer sequence, the knowledge point difficulty information also has individual differences for different students. Similarly, each question can have an initial difficulty level. Based on students' continuous feedback, a transformer model predicts the probability of a student answering the next question correctly at the next time point, dynamically updating the question difficulty information. The model also obtains students' ability status from historical answer records and estimates their accuracy on the next question. Regardless of the prediction's accuracy, the parameters of the entire model are updated. Simultaneously, the difficulty characteristics of the questions themselves and the knowledge points are dynamically updated. Furthermore, if a student consistently answers the same type of question (i.e., questions corresponding to the same knowledge point) incorrectly over a period of time, it indicates that the student has not mastered the corresponding knowledge point, and the student's learning ability for that type of question can be adjusted according to a preset ratio.
[0081] In one embodiment, assume the current question e t The corresponding knowledge point is C. t If a student answers the current question correctly, then set r. t The value is 1 for the first option and 0 for the rest; if the answer is incorrect, then r is set to 1. t+1 The value is 1 for the first m and 0 for the rest. The knowledge tracking algorithm outputs a vector of length m as a representation of the student's mastery of m knowledge points.
[0082] Step S03: Input the knowledge point difficulty characteristics, question difficulty characteristics, and learning ability characteristics into the preset ability status analysis model to obtain the student's ability status at the current time node.
[0083] In one embodiment, the difficulty features of the knowledge points, the difficulty features of the questions, and the learning ability features are input into a preset ability status analysis model to obtain the student's ability status at the current time point, including:
[0084] The difficulty characteristics of the knowledge points, the difficulty characteristics of the questions, and the learning ability characteristics are input into a preset ability status analysis model to obtain a set of ability statuses for the corresponding students. The set of ability statuses includes ability statuses at different time points, and the ability status analysis model includes a project response theory model.
[0085] In one embodiment, the difficulty characteristics of the knowledge points, the difficulty characteristics of the questions, and the learning ability characteristics of the target student are fed into an IRT (Item Response Theory) model to obtain the student's ability status. IRT is an improvement on traditional educational measurement theory. It models the student's ability status by comprehensively considering the student's ability and the difficulty of the questions they participated in, making the results more accurate.
[0086] Step S04, based on the student's ability status at each time point, retrieves matching questions from the question bank and pushes them to the corresponding student's end, including:
[0087] The learning ability status is compared with a preset ability status threshold, and the learning ability status below the ability status threshold is selected as the weak ability status.
[0088] Based on the stated weakness in ability, questions of matching difficulty are extracted from a pre-set question bank for question recommendation.
[0089] In one embodiment, a threshold g can be provided to determine whether a student's mastery of the knowledge points meets the requirements. The theoretical range of this threshold is (0, 1], but it typically ranges from [0.7 to 0.9]. The student's ability level across all knowledge points can then be obtained, denoted as h. t The subscripts represent time. The set of ability states of a student within the system is h = {h1, h2, h3, …, h…}. t (The subscript represents time). Based on threshold judgment, questions of corresponding difficulty are recommended to students, achieving personalized recommendations.
[0090] In one embodiment, after inputting the knowledge point difficulty features, question difficulty features, and learning ability features into a preset ability status analysis model to obtain the student's ability status at each time point, the method further includes:
[0091] By establishing a first constraint relationship between the capability states corresponding to adjacent time nodes through comparative learning, the capability state changes at adjacent time nodes are kept within a preset range.
[0092] By establishing a second constraint relationship through comparative learning, different students at the same point in time can be made to have different ability states at the same point in time.
[0093] In one embodiment, please refer to Figure 4 The student ability status obtained through the ability status analysis model is temporally sequential. Since students only encounter one question at each time point, a comparative learning model can be pre-trained. Using the learning ability status of the same student at a certain time point t as the original sample, the learning ability status at the next time point t+1 as the positive sample, and the learning ability status of different students at time point t as the negative sample, the comparative learning model is trained. The model training process is conventional and will not be elaborated here. After completing the comparative learning model training, the learning ability status of a student at two adjacent time points is input into the comparative learning model to bring the learning ability status of adjacent time points closer together, preventing abrupt changes in the learning ability status between adjacent time points. When conducting comparative analysis of the learning abilities of different students, the learning ability status of different students at the same time point is input into the comparative learning model to distance the learning ability status of different students, allowing for individual differences in the learning ability status of different students.
[0094] In one embodiment, the set of student ability states h = {h1, h2, h3, …, h} is obtained. t To make the entire evolution process more reasonable, the method of this application establishes the following constraints: 1) The ability status changes smoothly between adjacent time steps (i.e., the student's ability level at two consecutive time points obtained by this method will not fluctuate significantly); 2) The knowledge and ability status of different students has individual differences. That is, h t and h t+1 The mastery of corresponding knowledge points between two adjacent time steps should not fluctuate too much. Even if different students are given the same set of questions, their ability state ht at time step t should be individualized.
[0095] Please see Figure 3 In one embodiment, 20 questions from a student were selected and recorded. These 20 questions belonged to 5 knowledge points. Each question was represented by its corresponding knowledge point. For example, 37 represents multiplication, and if the student answered incorrectly, it would be recorded as (37, 0). Figure 3 Each value in the table represents the student's level of mastery of that knowledge point, ranging from 0 to 1, with lighter colors indicating a higher level of mastery.
[0096] It can be observed that initially, the score remained around 0.5. With continued practice, the mastery of the corresponding knowledge points would either increase or decrease, but the entire process was gradual without sudden changes. For example, for knowledge point 70, as students consistently answered questions related to knowledge point 70 correctly, their ability improved. However, over time, this ability would slowly decline, but would still remain at a higher level than initially perceived.
[0097] Even when given the same questions, different students have different ability levels due to variations in their answer records. Even with the same knowledge point, the difficulty varies from person to person. The method proposed in this application effectively captures this personalized student ability status, as well as information on the difficulty of the questions and knowledge points. It can provide students with more accurate learning path planning (recommending suitable questions based on a comprehensive consideration of question difficulty, knowledge point difficulty, and student learning ability), thereby enhancing their weaker abilities.
[0098] In one embodiment, this application also provides a question recommendation system based on personalized capabilities, used to execute the question recommendation method based on personalized capabilities described in the foregoing method embodiments. Since the technical principles of the system embodiment are similar to those of the foregoing method embodiments, the same technical details will not be repeated.
[0099] Please see Figure 2 In one embodiment, the question recommendation system based on personalized ability includes:
[0100] The vector representation module 10 is used to construct a correct question bank and an incorrect question bank based on a preset question bank, and to obtain the question vector representation corresponding to each question in the correct question bank and the incorrect question bank, respectively.
[0101] The student feature extraction module 11 is used to obtain the historical answer sequence of the same student, obtain the question vector representation of each question in the historical answer sequence from the corresponding question bank, obtain the question vector set corresponding to the historical answer sequence, and perform feature extraction on the question vector set to obtain the knowledge point difficulty feature, question difficulty feature and learning ability feature of the corresponding student.
[0102] The ability status analysis module 12 is used to input the knowledge point difficulty characteristics, question difficulty characteristics and learning ability characteristics into a preset ability status analysis model to obtain the student's ability status at the current time point;
[0103] The question recommendation module 13 is used to retrieve matching questions from the question bank based on the student's ability status at the current time point and push them to the corresponding student's end.
[0104] The above embodiments are merely illustrative of the principles and effects of this application and are not intended to limit this application. Any person skilled in the art can modify or alter the above embodiments without departing from the spirit and scope of this application. Therefore, all equivalent modifications or alterations made by those skilled in the art without departing from the spirit and technical concept disclosed in this application should still be covered by the claims of this application.
Claims
1. A question recommendation method based on personalized ability, characterized in that, include: Construct a correct question bank and an incorrect question bank based on a preset question bank, and obtain the question vector representation corresponding to each question in the correct question bank and the incorrect question bank respectively; Obtain the historical answer sequence of the same student, obtain the question vector representation of each question in the historical answer sequence from the corresponding question bank, obtain the question vector set corresponding to the historical answer sequence, perform feature extraction on the question vector set, and obtain the knowledge point difficulty feature, question difficulty feature and learning ability feature of the corresponding student; The difficulty characteristics of the knowledge points, the difficulty characteristics of the questions, and the learning ability characteristics are input into a preset ability status analysis model to obtain the ability status of students at each time point. Based on the student's ability status at each time point, a matching question is retrieved from the question bank and pushed to the corresponding student's end; Feature extraction is performed on the set of question vectors to obtain the knowledge point difficulty features, question difficulty features, and learning ability features of the corresponding students, including: Based on the preset initial difficulty value of each question and the correspondence between the questions and knowledge points, determine the initial difficulty value of the corresponding knowledge point for each question; Based on the question vector representation corresponding to each time node in the question vector set, predict the probability that the student will answer the next question correctly at the next time node. Update the question difficulty of the corresponding question based on the probability and the initial value of the question difficulty to obtain the question difficulty feature. Based on the question vector representation of all questions corresponding to each knowledge point in the question vector set, determine the knowledge point difficulty characteristics of the corresponding knowledge point; Based on the question vector representation of students for the same type of questions, the learning ability characteristics of the corresponding students are determined. The knowledge point difficulty characteristic is used to represent the degree of students' mastery of the same knowledge point, the question difficulty characteristic is used to represent the degree of students' mastery of the same question, and the learning ability characteristic is used to represent the students' ability to master questions corresponding to the same knowledge point. The difficulty characteristics of the knowledge points, the difficulty characteristics of the questions, and the learning ability characteristics are input into a preset ability status analysis model to obtain the student's ability status at each time point, including: The difficulty characteristics of the knowledge points, the difficulty characteristics of the questions, and the learning ability characteristics are input into a preset ability status analysis model to obtain a set of ability statuses for the corresponding students. The set of ability statuses includes ability statuses at different time points, and the ability status analysis model includes a project response theory model. After inputting the knowledge point difficulty characteristics, question difficulty characteristics, and learning ability characteristics into a preset ability status analysis model to obtain the student's ability status at each time point, the model also includes: By establishing a first constraint relationship between the capability states corresponding to adjacent time nodes through comparative learning, the capability state changes at adjacent time nodes are kept within a preset range. By establishing a second constraint relationship through comparative learning, different students at the same point in time can be made to have different ability states at the same point in time.
2. The question recommendation method based on personalized ability according to claim 1, characterized in that, Based on a preset question bank, construct a correct question bank and an incorrect question bank, and obtain the question vector representation corresponding to each question in the correct question bank and the incorrect question bank, respectively, including: Obtain the question stems and corresponding correct answers for all questions in the question bank; Based on the question stem and the corresponding correct answer, mark each question in the question bank as a correct question and an incorrect question; The correct questions are stored in a preset correct question bank as positive samples of the corresponding questions, and the incorrect questions are stored in the incorrect question bank as negative samples of the corresponding questions; The positive sample is input into a pre-trained contrastive learning model to obtain the question vector representation corresponding to the positive sample; The negative sample is input into the contrastive learning model to obtain the question vector representation corresponding to the negative sample.
3. The question recommendation method based on personalized ability according to claim 1, characterized in that, Retrieve the historical answer sequence of the same student, including: Obtain student identification; The current learning progress of the corresponding student is determined based on the identity identifier; Determine the relevant knowledge points based on the current learning progress; Based on the pre-defined correspondence between knowledge points and questions, determine the set of questions corresponding to the associated knowledge points; The question set is filtered based on the identity identifier to determine all the question-answering records of the corresponding student in the question set, and the historical answer sequence is generated according to the time sequence of the question-answering records.
4. The question recommendation method based on personalized ability according to claim 1, characterized in that, The question vector representation corresponding to each question in the historical answer sequence is obtained from the corresponding question bank, resulting in a set of question vectors corresponding to the historical answer sequence, including: If a student answers a question correctly in the historical question-answering sequence, the question vector representation of the correctly answered question is obtained from the correct question bank and entered into the question vector set; If a student answers a question incorrectly in the historical question-answering sequence, the question vector representation of the incorrectly answered question is obtained from the error question bank and entered into the question vector set.
5. The question recommendation method based on personalized ability according to claim 1, characterized in that, Based on the student's ability status at each time point, matching questions are retrieved from the question bank and pushed to the corresponding student's end, including: The learning ability status is compared with a preset ability status threshold, and the learning ability status below the ability status threshold is selected as the weak ability status. Based on the stated weakness in ability, questions of matching difficulty are extracted from a pre-set question bank for question recommendation.
6. A system for performing the question recommendation method based on personalized ability as described in any one of claims 1-5, characterized in that, include: The vector representation module is used to construct a correct question bank and an incorrect question bank based on a preset question bank, and to obtain the question vector representation corresponding to each question in the correct question bank and the incorrect question bank, respectively. The student feature extraction module is used to obtain the historical answer sequence of the same student, obtain the question vector representation of each question in the historical answer sequence from the corresponding question bank, obtain the question vector set corresponding to the historical answer sequence, and perform feature extraction on the question vector set to obtain the knowledge point difficulty feature, question difficulty feature and learning ability feature of the corresponding student. The ability status analysis module is used to input the knowledge point difficulty characteristics, question difficulty characteristics and learning ability characteristics into a preset ability status analysis model to obtain the student's ability status at each time point. The question recommendation module is used to retrieve matching questions from the question bank based on the student's ability status at each time point and push them to the corresponding student's end.