A resource allocation method and system based on deep learning

Abstract

The present application provides a resource allocation method and system based on deep learning, which relates to the technical field of resource allocation, wherein the method includes steps such as database construction, pre-class testing, resource allocation, concentration evaluation, maintenance and adjustment of resource allocation; the system includes a database construction module, a pre-class testing module, a resource allocation module, a concentration evaluation module, a maintenance module and a resource allocation adjustment module. The present application can match learning resources according to students' learning effects and improve learning efficiency.

Description

Technical Field

[0001] The present application relates to the technical field of resource allocation, and in particular, to a resource allocation method and system based on deep learning. Background Art

[0002] At present, personalized matching of learning resources means that the system accurately matches learning resources according to students' individual situations and needs. Learning resources are composed of a large number of digital resources such as teaching materials, tutoring materials, handouts, books and periodicals, audio and video lectures, exercise books, and test questions. The learning resources in the system are matched according to students' grade or hobbies, and the learning resources are pushed to the user terminal.

[0003] In the prior art, a Chinese patent with publication number CN110704737A discloses a method for matching online teaching resources, which includes: identifying and determining candidate training objects and candidate teaching objects based on a user's viewing history of multimedia content in an electronic device; pushing online teaching matching requests to the candidate training objects and candidate teaching objects, and determining formal training objects and formal teaching objects based on feedback results; and performing online teaching matching based on the attributes and states of the formal training objects and formal teaching objects.

[0004] However, different students have different learning abilities and different degrees of mastery of knowledge points. When the difficulty of the knowledge points in the learning resources is higher than the students' learning ability, the students will not be able to fully master the knowledge points, and their attention will be diverted. At this time, if they continue to study online according to the original learning plan, the students' learning effect and efficiency will be reduced. Summary of the invention

[0005] In order to match learning resources according to students' learning outcomes and improve learning efficiency, this application provides a resource allocation method and system based on deep learning.

[0006] In the first aspect, the present application provides a resource allocation method based on deep learning, which adopts the following technical solution:

[0007] A resource allocation method based on deep learning comprises the following steps:

[0008] Building a database: Building a knowledge base, which stores learning resources and question-and-answer questions;

[0009] Pre-class test: Conduct pre-class tests to determine students’ mastery of knowledge points;

[0010] Resource allocation: Allocate learning resources based on students’ mastery of knowledge points and generate a learning plan for this section;

[0011] Concentration assessment: including obtaining key learning points, setting questions and judging concentration;

[0012] Get the key points of learning: Based on the learning plan in this section, get the key points of learning;

[0013] Set questions: pop up questions t seconds before the key learning point, and get the answer results;

[0014] Concentration judgment: judge whether the answer result meets expectations. If so, execute the maintenance step; if not, execute the step of adjusting resource allocation;

[0015] Maintain: Continue to implement the original learning plan;

[0016] Adjust resource allocation: adjust the difficulty of the knowledge points and learning resources in the original learning plan, generate a new learning plan, and perform concentration evaluation steps.

[0017] By adopting the above technical solution, this application tests students before class, which can quickly and accurately evaluate students' mastery of the learned knowledge points (especially the knowledge points involved in this course), help identify students' knowledge blind spots and learning needs, and provide important basis for subsequent learning resource allocation. According to the results of the pre-class test, students are intelligently allocated the learning resources that best suit their learning needs, and a personalized learning plan is generated. This on-demand resource allocation method can effectively improve learning efficiency, reduce unnecessary learning burdens, and promote the pertinence and effectiveness of learning. Clarifying the key learning points based on the learning plan helps students achieve clear goals in the learning process and improves learning efficiency. Popping up questions and answers before learning key points can not only test students' understanding of the key points, but also prompt students that the following knowledge points are key knowledge points by popping up questions and answers, reducing the workload of teachers, that is, teachers no longer need to record the location of key knowledge points, and prompt students to concentrate before explaining key knowledge points, so that teachers can focus on teaching. By judging whether the answer results are in line with expectations, students' concentration and learning effects can be evaluated in real time, which helps teachers to obtain students' status in time and take corresponding measures. When students' answer results are not in line with expectations, that is, when students' attention is diverted due to poor grasp of knowledge points, the difficulty of knowledge points in the learning plan can be adjusted in time, a new learning plan can be generated, and concentration can be re-evaluated.

[0018] Optionally, after performing the pre-class test step and before performing the resource allocation step, the method further includes:

[0019] Setting difficulty: including question classification and setting weights;

[0020] Question classification: According to the difficulty of the questions, the questions are divided into three categories: difficult questions, medium questions and easy questions;

[0021] Set weights: Set the weights for difficult questions, medium questions, and easy questions separately.

[0022] By adopting the above technical solution, by refining the classification of question-and-answer questions and setting corresponding weights, we can more accurately understand the students' knowledge mastery at different difficulty levels. The classification of questions not only helps to evaluate the students' overall learning level, but also reveals the students' specific performance at different difficulty levels, and provides direction for the subsequent adjustment of resource allocation. This application sets different weights for different types of question-and-answer questions, which can reflect the importance of these question-and-answer questions in teaching evaluation. Through the classification and weight setting of question-and-answer questions, it is possible to more accurately judge which knowledge points students have weak links in, so as to allocate learning resources of corresponding difficulty and type in a targeted manner, and improve the pertinence and effectiveness of learning. Reasonable question classification and weight setting can make students feel the challenge and sense of accomplishment of learning, stimulate their interest and motivation in learning, promote the formation of a positive learning attitude, and enhance learning motivation.

[0023] Optionally, after executing the step of setting the question and before executing the step of judging the concentration, the method further includes:

[0024] Get the answer time: Calculate the answer time from the time the question pops up to the time the answer result is submitted ;

[0025] Accuracy calculation: Calculate the accuracy based on the answer results , the calculation model is as follows:

[0026] ;

[0027] Among them, A is the number of quiz questions answered correctly; B is the total number of quiz questions that pop up;

[0028] Concentration calculation: based on the time of answering questions and accuracy , calculate the concentration level C, the calculation model is as follows:

[0029] ;

[0030] Among them, T is the average answering time of the current question.

[0031] By adopting the above technical solution, by accurately calculating the time required for students from seeing the question to submitting the answer, that is, the answering time, it can directly reflect the actual time students spend on each question, which helps to more accurately judge students' thinking speed and coping ability. The accuracy rate directly reflects students' understanding and application ability of the knowledge they have learned, and is an important indicator for evaluating learning effects. The calculation of the concentration level C takes into account the answering time and accuracy rate, which can more comprehensively reflect the students' concentration and involvement in the answering process. When the student's answering time is close to or shorter than the average time and the accuracy rate is high, the value of the concentration level C will be relatively high, indicating that the student has maintained a high level of concentration in the answering process.

[0032] Optionally, in the step of judging concentration, the answer result meets the expected standard if the concentration level C is greater than a preset concentration threshold.

[0033] By adopting the above technical solution and setting the expected standard for the answer results, the student's concentration level can be quantified into a comparable value, which makes it easier for teachers or systems to quickly judge the student's concentration state.

[0034] Optionally, after executing the step of calculating the accuracy and before executing the step of calculating the concentration, the method further includes:

[0035] Obtaining standard answering time: extracting the answering time of students for the same type of questions from the knowledge base, and recording the question and answering time as the first data;

[0036] Second modeling: Building LSTM model based on deep learning algorithm;

[0037] Model training: Use the first data to train the LSTM model;

[0038] Prediction: Use the current question as the input of the trained LSTM model, output the predicted answering time, record it as the second data, and use the second data as the new average answering time for the current question.

[0039] By adopting the above technical solution, the historical answering time of the same type of question and answer questions is extracted from the knowledge base to provide basic data for subsequent model training. The LSTM model is good at processing long-term dependencies in sequence data and can more accurately predict the student's answering time. Use the first data to train the LSTM model so that it can learn the potential relationship between the question and answer questions and the answering time. Through training, the LSTM model can continuously optimize its internal parameters to improve the accuracy and reliability of the prediction. The current question and answer questions are used as the input of the trained LSTM model, and the predicted answering time is output. It is used as the new average answering time for concentration calculation. The predicted answering time can be dynamically adjusted according to the actual performance of the students, making the concentration calculation more personalized and accurate. Using the answering time predicted by the LSTM model as the average answering time can more accurately reflect the student's answering speed in the current state.

[0040] Optionally, in the accuracy calculation step, the accuracy It can also be calculated by the following calculation model:

[0041] ;

[0042] Among them, B is the total number of pop-up question and answer questions; is the weight of the i-th type of question-answering question; is the number of correctly answered questions in the i-th category.

[0043] By adopting the above technical solution, by assigning weights to different types of quiz questions, it is possible to reflect the importance or difficulty of certain quiz questions or knowledge points in teaching. In this way, even if students score high on some simple questions, if they perform poorly on important or difficult questions, their accuracy will be affected accordingly.

[0044] Optionally, after executing the step of determining the concentration and before executing the step of adjusting resource allocation, the method further includes:

[0045] Data collection: Collect students' learning behaviors after this course, including review time, test results, preview time and personal preferences;

[0046] The third modeling: Based on the reinforcement learning algorithm, establish a learning program evaluation model;

[0047] Select a learning plan: The learning plan evaluation model predicts students' future learning outcomes based on their learning behaviors and selects the optimal learning path.

[0048] By adopting the above technical solution, this application collects students' learning behaviors after the course ends, providing data support for predicting students' learning outcomes and selecting the optimal learning path. The collected learning behaviors cover multiple aspects such as students' learning time, test results, and personal preferences, which helps to more comprehensively evaluate students' learning status. The reinforcement learning algorithm can feedback learning outcomes based on students' learning behaviors, and continuously adjust and optimize learning plans to adapt to students' learning progress and changes in needs. The learning plan evaluation model can predict students' future learning outcomes and provide a scientific basis for selecting the optimal learning path. By selecting the optimal learning path, students' personalized learning needs can be met, and learning efficiency and learning effects can be improved. The formulation of personalized learning paths helps to optimize the allocation of learning resources and reduce resource waste and repeated investment.

[0049] Optionally, in the step of adjusting resource allocation, the difficulty of the knowledge points and the learning resources are adjusted based on the optimal learning path.

[0050] By adopting the above technical solutions, the difficulty of knowledge points can be dynamically adjusted according to the optimal learning path of students, so that students can encounter content suitable for their level at each learning stage. This adaptive adjustment can reduce the frustration or boredom caused by content that is too difficult or too easy, thereby improving concentration and learning efficiency. By gradually increasing the difficulty of knowledge points, students can steadily advance on the basis of existing knowledge and build a solid knowledge system. This step-by-step learning method helps to consolidate memory, deepen understanding, and promote long-term retention of knowledge. When students face learning content that is somewhat challenging, they tend to be more engaged and focused. After successfully overcoming these challenges, students will gain a strong sense of accomplishment and self-confidence, which will further stimulate their motivation and enthusiasm for learning. By adjusting the difficulty of knowledge points, teaching resources can be delivered more accurately. For students with faster learning progress, higher-level learning materials and challenges can be provided; for students with slower learning progress, more support and guidance can be provided. This precise delivery method helps to optimize the utilization efficiency of resources and reduce resource waste and repeated investment.

[0051] Optionally, after executing the step of allocating resources and before executing the step of acquiring learning key points, the method further includes:

[0052] First modeling: Building a BERT model based on deep learning algorithms;

[0053] Locating key points: Identify the key learning points in this section’s learning plan through the BERT model.

[0054] By adopting the above technical solutions, compared with traditional text analysis methods, the BERT model has higher accuracy in identifying keywords, key sentences, etc., which helps to locate the key positions of learning more accurately. The establishment and application of the BERT model makes the positioning of learning key points more accurate and efficient, thereby pushing question and answer questions more accurately, which helps to improve students' learning efficiency and teachers' teaching quality.

[0055] In the second aspect, the present application provides a resource allocation system based on deep learning, which adopts the following technical solutions:

[0056] A resource allocation system based on deep learning, comprising:

[0057] Database building module, used to build knowledge base;

[0058] Pre-class test module, used to conduct pre-class tests to determine students' mastery of knowledge points;

[0059] The resource allocation module is connected to the output end of the pre-class test module and is used to allocate learning resources and generate a learning plan for this lesson based on the students' mastery of knowledge points;

[0060] A concentration evaluation module, comprising a learning key point acquisition unit, a question setting unit and a concentration judgment unit, wherein the input end of the learning key point acquisition unit is communicatively connected to the output end of the resource allocation module, the output end of the learning key point acquisition unit is communicatively connected to the input end of the question setting unit, and the output end of the question setting unit is communicatively connected to the input end of the concentration judgment unit;

[0061] A unit for obtaining key learning points is used to obtain key learning points based on the learning plan of this section;

[0062] Set a question unit to pop up a question and answer question t seconds before the key learning point and obtain the answer result;

[0063] Concentration judgment unit, used to judge whether the answer results meet expectations;

[0064] A maintenance module, the input end of the maintenance module is communicatively connected to the output end of the concentration judgment unit, and is used to maintain the execution of the original learning plan;

[0065] The resource allocation adjustment module has an input end that is communicatively connected to an output end of the concentration judgment unit, and is used to adjust the difficulty of knowledge points and learning resources in the original learning plan to generate a new learning plan.

[0066] By adopting the above technical solutions, this application realizes comprehensive monitoring and personalized adjustment of students' learning process through close collaboration and real-time feedback mechanism of various modules. From establishing knowledge base to pre-class testing, resource allocation, concentration evaluation to adjustment of learning plan, each step is aimed at providing students with the most suitable learning resources and learning path. This highly personalized and flexible learning method helps to stimulate students' interest and motivation in learning and improve learning efficiency and effectiveness.

[0067] In summary, the present application includes at least one of the following beneficial technical effects:

[0068] 1. This application tests students before class, which can quickly and accurately evaluate students' mastery of the knowledge points they have learned (especially the knowledge points that will be involved in this course), help identify students' knowledge blind spots and learning needs, and provide an important basis for the subsequent allocation of learning resources. According to the results of the pre-class test, students are intelligently allocated the learning resources that best suit their learning needs, and a personalized learning plan is generated. This on-demand resource allocation method can effectively improve learning efficiency, reduce unnecessary learning burdens, and promote the pertinence and effectiveness of learning. Clarifying learning focus based on learning plans helps students achieve clear goals in the learning process and improves learning efficiency. Popping up questions and answers before learning key points can not only test students' understanding of key points, but also remind students that the following knowledge points are key knowledge points by popping up questions and answers, reducing the workload of teachers, that is, teachers no longer need to record the location of key knowledge points, and remind students to concentrate before explaining key knowledge points, so that teachers can focus on teaching. By judging whether the answer results are in line with expectations, students' concentration and learning effects can be evaluated in real time, which helps teachers to obtain students' status in time and take corresponding measures. When students' answer results are not in line with expectations, that is, when students' attention is diverted due to poor grasp of knowledge points, the difficulty of knowledge points in the learning plan can be adjusted in time, a new learning plan can be generated, and concentration can be re-evaluated.

[0069] 2. By accurately calculating the time it takes for students to see the question and submit their answers, that is, the answering time, it can directly reflect the actual time students spend on each question, which helps to more accurately judge students' thinking speed and response ability. The accuracy rate directly reflects students' understanding and application of the knowledge they have learned, and is an important indicator for evaluating learning effects. The calculation of the concentration level C takes into account the answering time and accuracy rate, which can more comprehensively reflect the students' concentration and involvement in the answering process. When the student's answering time is close to or shorter than the average time and the accuracy rate is high, the value of the concentration level C will be relatively high, indicating that the student has maintained a high level of concentration during the answering process. BRIEF DESCRIPTION OF THE DRAWINGS

[0070] Figure 1 is a flow chart of Example 1 of the present application;

[0071] Figure 2 This is a flowchart from S1 database building to S6 resource allocation adjustment in Example 2 of the present application;

[0072] Figure 3 This is a flowchart from S421 obtaining the answering time to S423 concentration calculation in Example 2 of the present application;

[0073] Figure 4 This is a flowchart from S4221 obtaining the standard answering time to S4224 prediction in Example 2 of the present application;

[0074] Figure 5 This is a flowchart from S71 data collection to S73 learning plan selection in Example 2 of the present application. DETAILED DESCRIPTION

[0075] The following combination Figures 1 to 5 This application is described in further detail.

[0076] Embodiment 1: This embodiment discloses a resource allocation method based on deep learning, referring to Figure 1 The method includes: S1 database building, S2 pre-class test, S3 resource allocation, S4 concentration evaluation, S5 maintenance and S6 resource allocation adjustment. In this embodiment, a knowledge base is first established, and the knowledge base stores learning resources and question-answering questions. Then, a pre-class test is conducted on the students, and then learning resources are allocated based on the test results to generate a learning plan. Then, the students' concentration is evaluated, and then it is determined whether the original learning plan needs to be adjusted based on the evaluation results. The specific steps of this embodiment are as follows:

[0077] S1: Build a knowledge base. The knowledge base stores learning resources and questions and answers. The learning resources include textbook essences, video tutorials, interactive demonstrations, case analyses, and question banks, etc., aiming to provide students with rich learning materials. The questions and answers cover various knowledge points, aiming to test students' understanding and application capabilities through questions and answers.

[0078] S2 is a pre-class test, which is conducted before class. The test questions cover the previous knowledge points involved in the knowledge points learned in this class, thereby measuring the students' mastery of basic knowledge.

[0079] S3 resource allocation dynamically adjusts the order and difficulty of learning resource push based on students' mastery of knowledge points and their strengths and weaknesses, and tailors a learning plan for this course for each student.

[0080] S4 concentration assessment, including S41 obtaining key learning points, S42 setting questions and S43 concentration judgment.

[0081] S41 Obtaining key learning points: Based on the specific content of this section’s learning plan, identify and obtain key learning points. These key learning points are often the core concepts, difficult points, or easy-to-make mistakes of the course, and are crucial to students’ learning outcomes.

[0082] S42 sets questions, and pops up a question t seconds before the key learning point is explained. t is a preset time interval. It is intended to test students before explaining the key learning point. First, it tests the mastery of the previously learned knowledge points, and second, it reminds students that the key learning point will be explained next and they need to concentrate. After the students finish answering, the answer results are automatically obtained.

[0083] S43 concentration judgment, intelligent analysis of the collected answer results, to determine whether the student's answer situation meets the expected standards. If so, it means that the answer accuracy rate is high, indicating that the student can accept the current learning plan, and the concentration and comprehension are good, there is no need to modify the learning plan, execute S5 to maintain; if not, it means that the answer accuracy rate is low, the student is not very adapted to the current learning plan, there is a situation of distraction, and it is necessary to adjust the difficulty of the knowledge points and learning resources in the learning plan, execute S6 to adjust resource allocation.

[0084] S5: Maintain and continue to implement the original learning plan.

[0085] S6 adjusts resource allocation, adjusts the difficulty of knowledge points and learning resources in the original learning plan, generates a new learning plan, and performs S4 concentration evaluation. The adjustment of the difficulty of knowledge points in the original learning plan includes replacing knowledge points with different difficulty levels (replacing with relatively simple knowledge points, breaking down and explaining current knowledge points, etc.), changing the order of knowledge points in the original learning plan, etc.

[0086] This embodiment constructs a knowledge base with rich learning resources and question-and-answer questions. Before class, the students' mastery of knowledge points is accurately assessed through tests, and then learning resources are intelligently allocated according to individual differences of students to customize personalized learning plans. During the learning process, the students' concentration is monitored in real time, and the learning difficulty and resources are dynamically adjusted by setting questions and answering questions before explaining the key points of learning and evaluating the results of the answers. If the student performs well, the current learning plan is maintained; if improvement is needed, the resource allocation is flexibly adjusted so that each student can complete the learning tasks efficiently and attentively at the pace that suits them best, and achieve continuous optimization of learning effects.

[0087] Example 2: Figure 2 As shown, the difference from Embodiment 1 is that after executing S2 pre-class test and before executing S3 resource allocation, it also includes:

[0088] S21 sets the difficulty, including S211 question classification and S212 setting weights.

[0089] S211 question classification, according to the difficulty of the question, the question and answer questions are divided into three categories: difficult questions, medium questions and easy questions.

[0090] The classification criteria for difficult questions, medium questions and easy questions are as follows:

[0091] If the number of knowledge points involved in this lesson is greater than three, the question is considered a difficult question; if the number of knowledge points involved in this lesson is not less than two and not greater than three, the question is considered a medium question; if the number of knowledge points involved in this lesson is not greater than one, the question is considered an easy question, for example:

[0092] The learning plan for this section includes: like terms, polynomials, polynomial addition and subtraction, and polynomial multiplication and division. If the question is "Determine whether 2ab is a polynomial", this question only involves one knowledge point in this lesson: the definition of polynomial, then this question is a simple question. If the question is "What polynomial is the result of 2ab+3a+4ba", this question involves three knowledge points in this lesson: like terms, polynomials, and polynomial addition and subtraction, then this question is a medium question. If the question is "What is the result of -2a(b+c)+b(a+c)", this question involves more than three knowledge points in this lesson: like terms, polynomials, polynomial addition and subtraction, and polynomial multiplication and division, then this question is a difficult question.

[0093] By judging the difficulty of the three types of questions by the number of knowledge points involved in the question, and automatically pushing the question, the teacher's workload can be reduced, and the entire teaching process is more intelligent. From the above content, we can see that the more knowledge points the question involves in the lesson, the closer it is to the end of the course, so the order in which the three types of questions appear in a course is: simple questions, medium questions and difficult questions.

[0094] S212 sets weights, setting weights for difficult questions, medium questions and simple questions respectively. Difficult questions are given higher weights due to their key position in the knowledge system and their role in training students' higher-order thinking ability. Medium questions are second, and simple questions have the smallest weight. The sum of the weights of all questions and answers that appear in this class is equal to one.

[0095] By setting weights reasonably, the fairness and efficiency of learning resource allocation can be improved, focusing on the consolidation of basic knowledge while encouraging students to challenge themselves and continuously improve.

[0096] This embodiment divides the difficulty of the questions and answers into different levels, which not only helps to evaluate students' mastery of knowledge points, but also reveals students' specific performance at different difficulty levels, providing direction for subsequent adjustment of resource allocation.

[0097] like Figure 3 As shown, in other embodiments, after executing S42 to set the question and before executing S43 to determine the concentration, the method further includes:

[0098] S421 obtains the answering time, which is calculated from the time the question pops up to the time the answer result is submitted. , the duration of answering the question It is equal to the time when the answer result is submitted minus the time when the question pops up.

[0099] This embodiment also sets a time threshold for answering questions. When the answering time exceeds the threshold, the pop-up box of the question will be automatically withdrawn. This is equal to the time threshold for answering questions. In case students have poor concentration, such as falling asleep or not sitting in their seats, the accuracy of students' answers can be counted in time.

[0100] S422 accuracy calculation, calculate the accuracy based on the answer results , the calculation model is as follows:

[0101] ;

[0102] Among them, A is the number of quiz questions answered correctly; B is the total number of quiz questions that pop up.

[0103] In other embodiments, the accuracy It can also be calculated by the following calculation model:

[0104] ;

[0105] Among them, B is the total number of pop-up question and answer questions; is the weight of the i-th type of question-answering question; is the number of correctly answered questions in the i-th category.

[0106] S423 concentration calculation, based on the time of answering questions and accuracy , calculate the concentration level C, the calculation model is as follows:

[0107] ;

[0108] Among them, T is the average answering time of the current question and answer. The average answering time T is obtained by statistical analysis of the number of answers to the current question and answer stored in the knowledge base and the time of each answer. For example, for question 1, there are 3 answering records in the knowledge base, and the answering times are 2 minutes, 3 minutes, and 4 minutes respectively. Then, through the calculation formula (2+3+4) / 3, the average answering time T of the current question and answer is equal to 3 minutes.

[0109] In S43 concentration judgment, the answer result meets the expected standard: the concentration level C is greater than the preset concentration threshold.

[0110] The concentration calculation in this embodiment not only focuses on the students' knowledge mastery (accuracy), but also takes into account their learning attitude and efficiency (answering time), providing strong support for the comprehensive evaluation of students' learning status. Based on the concentration calculation results, it is possible to more accurately identify whether the current learning plan is suitable for students, so as to take targeted measures to help them improve their concentration and learning efficiency.

[0111] like Figure 4 As shown, in other embodiments, after executing S422 accuracy calculation and before executing S423 concentration calculation, it also includes:

[0112] S4221 obtains the standard answering time, extracts the answering time of students for the same type of questions (difficult questions / medium questions / easy questions) from historical data in the knowledge base, and records the question and answering time as the first data.

[0113] Necessary preprocessing is performed on the first data, such as text cleaning, word segmentation, conversion into numerical input, etc., and the processed first data is used as new first data.

[0114] S4222 The second modeling is to build an LSTM model based on a deep learning algorithm. Design the architecture of the LSTM model, including the input layer, hidden layer (including multiple layers of LSTM units), and output layer.

[0115] S4223 model training, using S4221 to obtain the new first data in the standard answering time to train the LSTM model. The LSTM model parameters are adjusted through the back propagation algorithm until the performance of the LSTM model on the validation set reaches the preset stopping standard (such as the loss function no longer significantly decreases).

[0116] S4224 predicts that the current question and answer topic will be used as the input of the trained LSTM model, and the predicted answering time will be output, which will be recorded as the second data. The second data reflects the time that students may need to answer similar questions based on historical data. Therefore, the second data is used as the new average answering time for the current question and answer topic.

[0117] This embodiment collects historical data, trains an LSTM model with the historical data, and uses the trained LSTM model to predict the student's answering time for the current question. The answering time obtained by this solution is more suitable for the student. The answering time predicted by the LSTM model is used in the concentration calculation process, and the predicted concentration is more accurate.

[0118] like Figure 5 As shown, in other embodiments, after executing S43 concentration determination and before executing S6 adjusting resource allocation, the method further includes:

[0119] S71 data collection, collecting students' learning behaviors after this course, the learning behaviors include:

[0120] Review time, which records the total time students spend reviewing course content, completing exercises, or participating in review activities, helps to understand students' mastery of course content and their self-discipline in learning.

[0121] Test results: collect the scores and feedback of various tests (such as quizzes and unit tests) that students participate in after the course. The test results directly reflect the students' understanding and mastery of knowledge points.

[0122] Preview time tracks the length of time students actively preview new content before the next class begins. Preview behavior can predict students' learning attitudes and autonomous learning abilities.

[0123] Personal preferences: Through questionnaires, learning behavior analysis, etc., identify and record students' learning style preferences (such as visual, auditory, hands-on), learning time period preferences, and preferences for specific teaching methods or learning resources.

[0124] S72 is the third modeling, which is based on the rich learning behavior data collected in S71 and uses the reinforcement learning algorithm to establish a learning program evaluation model. The learning program evaluation model can adopt a DQN (Deep Q-Network) model or a Q-learning model.

[0125] The setting of the learning scheme evaluation model is as follows:

[0126] State definition: Students’ learning behavior data (review time, test results, preview time, personal preferences, etc.) are used as the state input of the learning program evaluation model to reflect the students’ current learning status and progress.

[0127] For example:

[0128] Revision time: The average number of hours students spent revising for a specific subject in the most recent week.

[0129] Test results: The score of the most recent unit test, and the performance trend (improvement or regression compared to the last test).

[0130] Preview time: the length of time students spend previewing new content in advance, especially the preview of difficult chapters.

[0131] Personal preferences: Students’ preferences for different types of learning resources (such as videos, graphics, and animations) and their acceptance of specific teaching styles based on historical learning behavior analysis.

[0132] Action space: A series of possible learning plans or intervention measures are defined as the output actions of the learning plan evaluation model, such as recommending exercises of a specific difficulty, adjusting the presentation of learning resources (videos, graphics, and text), setting a personalized learning schedule, etc.

[0133] For example:

[0134] Recommended exercises: Based on students’ current learning progress and test scores, we intelligently recommend exercises of appropriate difficulty to strengthen weak links.

[0135] Adjust learning resources: If students prefer video learning but have not been effective in using graphic resources recently, the system recommends switching to video resources.

[0136] Personalized study schedule: Customize daily / weekly study plans based on students’ study habits and progress, including reasonable arrangements for review, preview and rest time.

[0137] Emotional motivation: When it is detected that students lack motivation to learn, positive motivation is provided through system messages to improve their learning enthusiasm.

[0138] Reward mechanism: A reward mechanism is designed to evaluate the effectiveness of different learning plans. Rewards can be set based on students' long-term learning outcomes (such as improved final exam scores), short-term learning progress (such as improved test scores), or improved learning behaviors (such as increased pre-study time).

[0139] For example:

[0140] Long-term learning achievement rewards: Based on the improvement of students’ final exam scores, additional learning resource rewards or commendations will be given for every certain percentage increase (such as 5%).

[0141] Short-term learning progress rewards: After each unit test, students will be given learning points or small rewards based on their performance improvement, such as unlocking more difficult learning content. Learning behavior improvement rewards: Students are encouraged to develop good learning habits, such as learning medals or the right to use special learning tools if they preview for more than a specified period of time for a week.

[0142] S73 selects a learning plan. The learning plan evaluation model predicts students’ future learning outcomes under different learning plans based on their learning behaviors, and selects the learning path that can bring the best learning outcomes (optimal learning path) as the recommended plan.

[0143] In S6, adjust resource allocation, based on the optimal learning path, adjust the difficulty of knowledge points and the allocation strategy of learning resources, for example:

[0144] Adjustment of the difficulty of knowledge points: In response to the difficulties and bottlenecks that students may encounter in their current learning path, we will add auxiliary materials (such as more detailed explanations, case analysis) or reduce the difficulty requirements of some content in a timely manner to ensure that students can smoothly advance their learning process.

[0145] Learning resource allocation: Based on students' learning preferences and efficiency, the system dynamically adjusts the recommended proportions and presentation order of different types of learning resources (such as video tutorials, online discussions, and exercise question banks) to maximize students' learning effects and learning experience.

[0146] Feedback loop: Through continuous learning behavior monitoring and effectiveness evaluation, the system continuously optimizes the selection of learning plans and resource allocation strategies, forming a closed-loop personalized learning optimization system.

[0147] This embodiment can provide strong support for personalized teaching and optimization of learning paths through steps such as data collection, third-party modeling, and selection of learning plans. These steps not only help to more comprehensively evaluate students' learning status and needs, but also dynamically adjust learning plans and resource allocation based on students' learning behavior feedback and learning outcomes, thereby achieving more accurate and efficient personalized teaching.

[0148] In other embodiments, after executing S3 resource allocation and before executing S41 acquiring learning key points, the method further includes:

[0149] S81 First modeling, based on the deep learning algorithm, build a BERT model. In this step, a large amount of text data related to the current learning topic will be collected and processed, including but not limited to textbook content, reference books, academic papers, teaching video scripts, etc. These data will be used as training sets, and through a complex training process, the BERT model will learn the inherent structure and laws of subject knowledge. In addition, this step will also fine-tune the BERT model according to specific teaching needs and goals, so that it can more accurately capture the key points in the learning plan.

[0150] S82 locates the key points and identifies the key learning points in this section's learning plan through the BERT model. This step will input the text content of the learning plan into the BERT model, which will automatically analyze the semantic information, contextual relationships, and dependencies between words in the text to identify the key information in the text. Based on this information, the key learning points in the learning plan can be accurately located, and marks can be added at the key learning points, which will be used as an important reference for subsequent teaching sessions.

[0151] This embodiment can not only realize the personalized allocation of learning resources, but also more accurately identify the key points in the learning plan, providing a more scientific and accurate basis for subsequent concentration evaluation and learning plan adjustment, which will help to further improve students' learning effects and learning experience and achieve more intelligent and personalized teaching goals.

[0152] Embodiment 3: This embodiment discloses a resource allocation system based on deep learning, the system comprising:

[0153] The database building module is used to build a knowledge base.

[0154] The pre-class test module, through carefully selected questions, aims to quickly evaluate students' mastery of previous knowledge points related to the knowledge points to be learned through short tests.

[0155] The resource allocation module is connected to the output of the pre-class test module and is used to intelligently allocate learning resources suitable for students based on their knowledge mastery. The learning resources include video tutorials, graphic analysis, online exercises, etc., which are designed to help students consolidate their learned knowledge and guide them to explore new knowledge in depth. The resource allocation module will also generate a personalized learning plan for this section based on individual differences and learning progress of students, ensuring that each student can obtain the most suitable learning content and learning path for themselves.

[0156] The concentration evaluation module includes a learning key point acquisition unit, a question setting unit and a concentration judgment unit. The input end of the learning key point acquisition unit is communicatively connected to the output end of the resource allocation module, the output end of the learning key point acquisition unit is communicatively connected to the input end of the question setting unit, and the output end of the question setting unit is communicatively connected to the input end of the concentration judgment unit.

[0157] The unit for obtaining learning key points is used to obtain learning key points based on the learning plan in this section.

[0158] Set up question units to automatically pop up questions and answers t seconds before the key learning points appear, so as to test students' attention and understanding, and automatically obtain the answer results.

[0159] The concentration judgment unit is used to judge whether the answer results meet expectations.

[0160] A maintenance module, wherein the input end of the maintenance module is communicatively connected to the output end of the concentration judgment unit, and is used to maintain the execution of the original learning plan.

[0161] The resource allocation adjustment module has an input end that is communicatively connected to an output end of the concentration judgment unit, and is used to adjust the difficulty of knowledge points and learning resources in the original learning plan to generate a new learning plan.

[0162] The various modules of this embodiment work together to determine the students' mastery of knowledge points based on pre-class tests, then allocate student resources based on the knowledge points' mastery, generate a learning plan for this section, and determine whether the current learning plan and learning resources need to be adjusted by judging whether the students' concentration meets expectations.

[0163] The above are all preferred embodiments of the present application, and the protection scope of the present application is not limited thereto. Therefore, any equivalent changes made according to the structure, shape, and principle of the present application should be included in the protection scope of the present application.

Claims

1. A resource allocation method based on deep learning, characterized in that: include: Building a database: Building a knowledge base, which stores learning resources and question-and-answer questions; Pre-class test: Conduct pre-class tests to determine students’ mastery of knowledge points; Setting difficulty: including question classification and setting weights; Question classification: According to the difficulty of the quiz questions, the quiz questions are divided into three categories: difficult questions, medium questions and simple questions; if the number of knowledge points involved in this lesson is greater than three, the quiz question is judged to be a difficult question; if the number of knowledge points involved in this lesson is not less than two and not more than three, the quiz question is judged to be a medium question; if the number of knowledge points involved in this lesson is not more than one, the quiz question is judged to be a simple question; Set weights: Set the weights of difficult questions, medium questions and easy questions respectively; Resource allocation: Allocate learning resources based on students’ mastery of knowledge points and generate a learning plan for this section; Concentration assessment: including obtaining key learning points, setting questions and judging concentration; Get the key points of learning: Based on the learning plan in this section, get the key points of learning; Set questions: pop up questions t seconds before the key learning point, and get the answer results; Get the answer time: Calculate the answer time from the time the question pops up to the time the answer result is submitted ; Accuracy calculation: Calculate the accuracy based on the answer results , the calculation model is as follows: ； Among them, A is the number of quiz questions answered correctly; B is the total number of quiz questions that pop up; Obtaining standard answering time: extracting the answering time of students for the same type of questions from the knowledge base, and recording the question and answering time as the first data; Second modeling: Building LSTM model based on deep learning algorithm; Model training: Use the first data to train the LSTM model; Prediction: Use the current question as the input of the trained LSTM model, output the predicted answering time, record it as the second data, and use the second data as the new average answering time for the current question; Concentration calculation: based on the time of answering questions and accuracy , calculate the concentration level C, the calculation model is as follows: ； Among them, T is the average answering time of the current question; Concentration judgment: judge whether the answer result meets expectations. If so, execute the maintenance step; if not, execute the resource allocation adjustment step. The standard for the answer result to meet expectations is: the concentration level C is greater than the preset concentration threshold; Maintain: Continue to implement the original learning plan; Adjust resource allocation: adjust the difficulty of the knowledge points and learning resources in the original learning plan, generate a new learning plan, and perform concentration evaluation steps.

2. The resource allocation method based on deep learning according to claim 1, characterized in that: In the accuracy calculation step, the accuracy It can also be calculated by the following calculation model: ； Among them, B is the total number of pop-up question and answer questions; is the weight of the i-th type of question-answering question; is the number of correctly answered questions in the i-th category.

3. The resource allocation method based on deep learning according to claim 1, characterized in that: After executing the step of determining the concentration and before executing the step of adjusting resource allocation, the method further includes: Data collection: Collect students' learning behaviors after this course, including review time, test results, preview time and personal preferences; The third modeling: Based on the reinforcement learning algorithm, establish a learning program evaluation model; Select a learning plan: The learning plan evaluation model predicts students' future learning outcomes based on their learning behavior and selects the optimal learning path.

4. The resource allocation method based on deep learning according to claim 3, characterized in that: In the step of adjusting resource allocation, the difficulty of knowledge points and learning resources are adjusted based on the optimal learning path.

5. The resource allocation method based on deep learning according to claim 1, characterized in that: After executing the step of allocating resources and before executing the step of acquiring learning key points, the method further includes: First modeling: Building a BERT model based on deep learning algorithms; Locating key points: Identify the key learning points in this section’s learning plan through the BERT model.

6. A resource allocation system based on deep learning, characterized in that: The system is used to implement the method according to any one of claims 1 to 3, and the system further includes: Database building module, used to build knowledge base; Pre-class test module, used to conduct pre-class tests to determine students' mastery of knowledge points; The resource allocation module is connected to the output end of the pre-class test module and is used to allocate learning resources and generate a learning plan for this lesson based on the students' mastery of knowledge points; A concentration evaluation module, comprising a learning key point acquisition unit, a question setting unit and a concentration judgment unit, wherein the input end of the learning key point acquisition unit is communicatively connected to the output end of the resource allocation module, the output end of the learning key point acquisition unit is communicatively connected to the input end of the question setting unit, and the output end of the question setting unit is communicatively connected to the input end of the concentration judgment unit; A unit for obtaining key learning points is used to obtain key learning points based on the learning plan of this section; Set a question unit to pop up a question and answer question t seconds before the key learning point and obtain the answer result; Concentration judgment unit, used to judge whether the answer results meet expectations; A maintenance module, the input end of the maintenance module is communicatively connected to the output end of the concentration judgment unit, and is used to maintain the execution of the original learning plan; The resource allocation adjustment module has an input end that is communicatively connected to an output end of the concentration judgment unit, and is used to adjust the difficulty of knowledge points and learning resources in the original learning plan to generate a new learning plan.

Images