Learning path planning method based on habit analysis

Abstract

The application provides a learning path planning method based on habit analysis, belongs to the field of learning path planning, solves the problem of complex learning path planning, and specifically comprises the following steps: S1, acquiring personalized information of a learner; acquiring learning goals and learning preset time length of the learner to form planning data; S2, judging the continuous learning time length of the learner according to the personalized information of the learner; combining the learning goals of the learner, preliminarily planning the learning path of the learner to obtain a first planning route; S3, analyzing the learning state of the learner according to the first planning route, arranging a review plan for the learner, and optimizing the first planning route according to the review plan; and S4, acquiring the learning preset time length of the learner, adjusting the optimized route, and completing route planning; the application plans the learning path of the learner, provides data reference for the learner, and improves the learning efficiency of the learner.

Description

TECHNICAL FIELD

[0001] The application discloses a learning path planning method based on habit analysis, and relates to the field of learning path planning. BACKGROUND

[0002] The existing learning path planning method has the following disadvantages:

[0003] Lack of personalization: The existing method usually plans the path based on fixed rules (such as "first learn the basic course, then learn the advanced course") or preset path templates, and fails to fully consider the individual differences (such as knowledge base, learning style, cognitive ability, etc.) of learners, resulting in low learning efficiency and poor learning experience of learners;

[0004] Lack of explainability and transparency: The existing learning path planning is mainly generated by a black box model such as deep learning, and it is difficult to explain the recommendation logic, so learners cannot understand why they are recommended a specific course or order; during the learning process, learners cannot intuitively perceive the learning progress and goal achievement;

[0005] Lack of dynamic adaptability: The path planning lacks a dynamic adjustment mechanism and cannot flexibly adjust the content difficulty or progress according to the real-time performance (such as the correct answer rate and attention dispersion degree) of learners, so when the ability of learners does not match the difficulty, time is wasted or frustration is caused. SUMMARY

[0006] In view of the deficiencies of the prior art, the application aims to provide a learning path planning method based on habit analysis, which aims to solve the problem of complex learning path planning.

[0007] In order to achieve the above-mentioned purpose, the application is realized by the following technical scheme: a learning path planning method based on habit analysis, the planning method comprising:

[0008] Step S1: acquiring the individual information of learners and planning data;

[0009] Step S2: judging the continuous learning time length of learners according to the individual information of learners, and preliminarily planning the learning path of learners according to the continuous learning time length and combining the planning data to obtain a first planning route;

[0010] Step S3: verifying the learning completion degree of learners according to the first planning route, analyzing the learning state of learners according to the learning completion degree of learners and combining the individual information of learners, arranging a review plan for learners, and optimizing the first planning route according to the review plan;

[0011] Step S4: dynamically adjusting the optimized route according to the planning data to complete the route planning.

[0012] Further, the specific steps of step S1 are as follows:

[0013] Step S11: Obtain real-time learning video data of the learner through a video capture device, and pre-process the video data to generate high-quality video data;

[0014] Step S12: Input the high-quality video data into a multi-modal large model, which performs end-to-end deep visual understanding on video frames to generate visual features containing information such as learner expressions, attention, and posture; the multi-modal large model further fuses the visual features with synchronously collected speech and other modal data to generate comprehensive feature representations;

[0015] Step S13: Based on the comprehensive feature representations, analyze the video sequence using a self-attention mechanism to generate quantized student state parameters, which constitute the personalized information of the learner.

[0016] Further, the specific steps of step S13 are as follows:

[0017] Obtain the correlation set tzj(i) and extract the number of correlation values t in the correlation set; the correlation values in the correlation set are denoted as tzz(1), tzz(2), …, tzz(t);

[0018] Step S1312: Obtain the correlation values tzz(1), tzz(2), …, tzz(t); construct a correlation function hs according to the correlation values, which is specifically as follows:

[0019] ;

[0020] Obtain the correlation values of the learner's facial image, obtaining xtz(1) to xtz(t); combine the correlation function hs to calculate the learner's correlation function xhs;

[0021] ;

[0022] Calculate the difference between the learner's correlation function xhs and the correlation function hs to obtain the difference in the x1 to xt direction;

[0023] ;

[0024] Calculate the absolute value of the difference in the x1 to xt direction to obtain the detection value; obtain the expression type corresponding to the smallest detection value, obtain the emotional intensity of the learner according to the self-attention mechanism, and combine the expression category and emotional intensity to obtain the expression information of the learner.

[0025] Further, the specific steps of the step S13 further include:

[0026] Obtaining the attention concentration area, connecting the starting point of the learner's line of sight and the center point of the attention concentration area as a standard line, and recording the length of the standard line as bzc; performing edge detection on the attention concentration area; testing the included angle between the line of sight and the standard line when the line of sight is at the edge of the attention concentration area to obtain an edge included angle by; obtaining the included angle interval [0, by] of attention from the edge included angle;

[0027] Obtaining the landing point of the learner's line of sight, connecting the landing point of the line of sight and the center point of the attention concentration area, recording the distance between the landing point of the line of sight and the center point of the attention concentration area as lyz; in combination with the length bzc of the standard line, the line of sight angle a of the learner is calculated;

[0028] When the line of sight angle a of the learner is in [0, by], it is judged that the line of sight is concentrated;

[0029] When the line of sight of the learner is concentrated, the blinking frequency is detected in combination to obtain the blinking frequency zpl(1) to zpl(z) of the learner in z daily states; the mean value of the blinking frequency zpl(1) to zpl(z) is calculated to obtain the mean frequency jpl; the real-time blinking frequency spl of the learner is obtained; if spl>jpl, it is judged that the learner's attention is not concentrated.

[0030] Further, the multi-modal large model adopts a Transformer-based architecture, divides video frames into image blocks and directly inputs them into an encoder, learns the internal correlation between image blocks through a self-attention mechanism, and comprehensively understands image content, so that it is not necessary to pre-extract facial key points or design handcrafted features.

[0031] Further, the student state parameters include but are not limited to: the expression category of the student; the concentration level; the emotional intensity; and the learning behavior.

[0032] Further, the specific steps of the step S2 are as follows:

[0033] Step S21: obtaining personalized information of the learner during learning, monitoring the expression category, emotional intensity and concentration level of the learner in real time during learning; according to the expression category, emotional intensity and concentration level of the learner during learning, the continuous learning duration of the learner is judged;

[0034] Step S22: obtaining the learning goal of the learner, decomposing the learning goal of the learner in combination with the continuous learning duration of the learner to obtain learning units; the correlation between the learning units is calculated to obtain a correlation value, and the learning path of the learner is preliminarily planned according to the correlation value.

[0035] Further, the specific steps of the step S22 are as follows:

[0036] Step S221: Obtain the learning goal of the learner, the continuous learning time of the learner; obtain the estimated learning time of the learning goal; divide the estimated learning time into learning units according to the continuous learning time of the learner;

[0037] Step S222: Obtain the number xgs of learning units; obtain the required time dsj(1), dsj(2), …, dsj(xgs) of each learning unit of the learner; combine the learning units two by two to obtain a combination unit; obtain the required time zh2 of the learner to learn the combination unit;

[0038] The correlation value glz2 between the learning units is calculated from the required time dsj(1) to dsj(xgs) of each learning unit of the learner and the required time zh of the learner to learn the combination unit;

[0039] ;

[0040] Wherein: glz2(u, v) represents the correlation value between the learning units u and v; zh2(u, v) represents the required time of the combination unit composed of the learning unit u and the learning unit v; dsj(u) represents the required time to learn the u-th learning unit, and dsj(v) represents the required time to learn the v-th learning unit;

[0041] Step S223: Add learning units on the basis of the combination unit; obtain the required time zh3 of the learner to learn the combination unit; calculate the correlation value according to the calculation formula of step S222 to obtain glz3; similarly, obtain zh4 to zhxgs; calculate the corresponding correlation value; accumulate the correlation values, and take the combination route corresponding to the maximum correlation value as the first planning route.

[0042] Further, the specific steps of the step S3 are as follows:

[0043] Step S31: The learner learns according to the first planning route; conduct in-class exercises on the learner during the learning process of the learner to obtain the learning behavior of the learner; obtain the answer information of the learner according to the learning behavior of the learner; calculate the learning completion degree of the learner according to the answer information;

[0044] Step S32: Analyze the learning state of the learner from the personalized information of the learner combined with the learning completion degree of the learner, calculate the review weight according to the learning state of the learner, arrange the review plan for the learner according to the review weight, and optimize the first route by combining the review plan with the first planning route.

[0045] Further, the specific steps of step S31 are as follows:

[0046] Step S311: Obtain the total score value zfz of the in-class exercise; obtain the score value kfz deducted by the learner due to errors in the in-class exercise; obtain the score value wfz deducted by the learner due to unanswered questions in the in-class exercise; calculate the error value cwz of the learner;

[0047] Step S312: Obtain the estimated time ygt of the in-class exercise; obtain the number of questions tsl; obtain the hand video of the learner; record the hand pause time tds of the learner when completing each question t ; obtain the completion time wct of the in-class exercise of the learner; calculate the stutter value kdz of the learner;

[0048] ;

[0049] Step S313: Calculate the learning completion degree of the learner according to the error value of the learner and the stutter value of the learner.

[0050] Further, the specific steps of step S32 are as follows:

[0051] Step S321: Obtain the personalized information of the learner, i.e., the time series data of the student state parameters generated by the multi-modal large model; from the time series data, extract the total time length qxt of the learner in a negative emotional state and the total time length zyt of the learner in an inattentive state during the learning process; obtain the total learning time xxt of the learner; take the ratio of the time of negative emotions of the learner during the learning process to the learning time of the learner as the first weight, and take the ratio of the time of inattention of the learner to the learning time of the learner as the second weight;

[0052] Step S322: Obtain the learning completion degree wcd of the learner; calculate by combining the first weight and the second weight, multiply the learning completion degree by the first weight and the second weight, to obtain the review weight fxq;

[0053] According to the review weight, arrange the review time of the learning unit, sort the review weight in ascending order, and review in order according to the sorting position of the review weight, to obtain a review plan; supplement the first planning route with the review plan, and optimize the first planning route to obtain an optimized route.

[0054] Further, the specific steps of step S4 are as follows:

[0055] Step S41: obtaining a current learning duration dqt, a learning preset duration yst, and a ranking position of a current learning unit in an optimized route; performing proportional calculation according to the ranking position and the number of learning units to obtain a learning progress; and comparing a ratio of the current learning duration and the learning preset duration with the learning progress;

[0056] Step S42: taking the ratio of the current learning duration and the learning preset duration as an expected value; if the expected value is greater than the learning progress, adjusting the optimized route:

[0057] performing difference calculation on the expected value and the learning progress to obtain a learning deviation; performing operation on the learning deviation and the learning preset duration to obtain a deviation duration; obtaining the number of remaining learning units in the optimized route; obtaining an adjustment duration according to the deviation duration and the number of remaining learning units; and performing time compression on the learning unit in the optimized route according to the adjustment duration, thereby completing route planning.

[0058] Compared with the prior art, the present application has the following beneficial effects:

[0059] Improving personalized selection of learners: learners autonomously select learning objectives and learning time, and improve learning time through attention judgment and expression analysis of the learners; the learning objectives are decomposed into learning units based on the improved learning time, and learning path planning is performed according to the learning units; the learning needs of learners are optimized through autonomous selection of learning objectives and learning time by the learners, unnecessary content is avoided from being repeatedly learned, and the learning efficiency of learners is improved through the learning unit mode;

[0060] Planning learning path based on time: based on learning units; analyzing learning duration required by different learning unit combinations through big data; obtaining time variation according to the combination mode of learning units; optimizing learning duration of learners; and helping learners to complete learning objectives within a certain time through the path planning mode;

[0061] Real-time adjustment of learning path: the present application adjusts the learning path in real time through the learning state of learners; the learning experience of learners is optimized, more review time is arranged for learning content with high difficulty, and the time of different learning units is reasonably arranged in combination with preset time. BRIEF DESCRIPTION OF DRAWINGS

[0062] Other features, objects and advantages of the present application will become more apparent from the following detailed description of non-limiting embodiments, made with reference to the accompanying drawings:

[0063] Figure 1 The figure is a schematic diagram of the method of the present application;

[0064] Figure 2A facial expression detection diagram of the present application;

[0065] Figure 3 A attention recognition diagram of the present application;

[0066] Figure 4 A path planning diagram of the present application. DETAILED DESCRIPTION

[0067] In order to make the above-mentioned purposes, features and advantages of the present application more obvious and easy to understand, the present application will be further described in detail below in combination with the drawings and specific embodiments.

[0068] Embodiment one

[0069] Please refer to Figure 1 The learning path planning method based on habit analysis includes:

[0070] Step S1: obtaining personalized information of the learner; obtaining learning goals and learning preset time length of the learner to form planning data;

[0071] Step S11: obtaining real-time learning video data of the learner through a video capture device, and pre-processing the video data to generate high-quality video data;

[0072] Step S12: inputting the high-quality video data into a multi-modal large model, the multi-modal large model performing end-to-end deep visual understanding on video frames to generate visual features containing information such as learner's facial expression, attention, and posture; the multi-modal large model further fuses the visual features and other modal data such as synchronous collected speech to generate comprehensive feature performance;

[0073] It should be noted that: the multi-modal large model adopts a Transformer-based architecture, divides video frames into image blocks and directly inputs them into an encoder, learns the internal correlation between image blocks through a self-attention mechanism, and comprehensively understands image content, so that it is not necessary to pre-extract facial key points or design handcrafted features.

[0074] Step S13: based on the comprehensive feature performance, analyzing video sequences using a self-attention mechanism to generate quantized student state parameters, the student state parameters constituting personalized information of the learner.

[0075] It should be noted that: the student state parameters include but are not limited to: student's facial expression categories such as confusion, concentration or fatigue; concentration level such as high, medium or low; emotional intensity; and learning behavior such as eye wandering, frequent blinking or change of sitting posture.

[0076] Please refer to Figure 2Step S131: Perform facial expression detection on the learner; obtain the facial expression recognition dataset, classify facial expression categories, such as happy, sad, surprised, and confused; learn the inherent relationship between image patches for the same type of expression through a self-attention mechanism to obtain the association value; statistically analyze the association value to obtain the association set tzj(i); construct the association function hs based on the association set;

[0077] It should be noted that: Facial expression classification refers to the detection and classification of human facial expressions using the ultra-lightweight face detection model RFB-320;

[0078] The system acquires real-time learning videos of learners; it also acquires facial images of learners, obtains the association values ​​of these images based on a self-attention mechanism, and uses the association function to detect learners' expressions, thus obtaining the detection results.

[0079] Step S1311: Obtain the association set tzj(i), extract the number of association values ​​t in the association set; denote the association values ​​in the association set as tzz(1), tzz(2), ..., tzz(t);

[0080] Step S1312: Obtain the correlation values ​​tzz(1), tzz(2), ..., tzz(t); construct the correlation function hs based on the correlation values. The specific correlation function is as follows:

[0081] ;

[0082] It should be noted that x1, x2, ..., xt are not calculated values; they represent correlation values ​​in different directions and are only used in the judgment process.

[0083] Step S1313: Obtain the association values ​​of the learner's facial images, denoted as xtz(1), xtz(2), ..., xtz(t); calculate the learner's association function xhs by combining it with the association function hs;

[0084] ;

[0085] The learner's correlation function xhs and correlation function hs are compared to obtain the difference in the x1, x2, ..., xt directions;

[0086] ;

[0087] For example, xtz(1)-tzz(1) is the difference in the x1 direction;

[0088] The absolute values of the differences in the x1, x2, …, xt directions are accumulated to obtain a detection value; the expression category corresponding to the smallest detection value is obtained, the emotional intensity of the learner is obtained according to the self-attention mechanism, and the expression information of the learner is obtained in combination with the expression category and the emotional intensity.

[0089] Step S132: Obtain the visual learning area of the learner, denoted as the attention concentration area; perform gaze detection on the learner; determine whether the attention of the learner is concentrated, and optimize the attention determination in combination with the blink frequency of the learner to obtain the attention information of the learner.

[0090] It should be noted that: Gaze Tracking / Eye Tracking is a technology that captures and analyzes the movement trajectory of the human eye to determine the gaze direction or focus area of the user; the gaze is detected according to the infrared light source + optical imaging: the eye is irradiated by a near-infrared light source (such as an LED), the position relationship between the Purkinje spot and the pupil center is captured by a camera, and the gaze point is calculated in combination with a geometric model (such as a 3D eye model).

[0091] Please refer to Figure 3 Step S1321: Obtain the attention concentration area, connect the gaze starting point of the learner and the center point of the attention concentration area as a standard line; the length of the standard line is denoted as bzc; perform edge detection on the attention concentration area; test the included angle between the gaze and the standard line when the gaze is at the edge of the attention concentration area to obtain the edge included angle by; obtain the included angle interval [0, by] of the attention from the edge included angle.

[0092] It should be noted that the attention concentration area is the display area of the learning content of the learner during the learning process.

[0093] Step S1322: Obtain the gaze landing point of the learner through the gaze detection device, connect the gaze landing point and the center point of the attention concentration area, and denote the distance between the gaze landing point and the center point of the attention concentration area as lyz; obtain the gaze angle a of the learner in combination with the length bzc of the standard line; the specific solving process is as follows:

[0094] ;

[0095] When the gaze angle a of the learner is in [0, by], it is determined that the gaze of the learner is concentrated.

[0096] Step S1323: When the learner's line of sight is concentrated, the blinking frequency of the learner in z daily states zpl(1), zpl(2),..., zpl(z) is obtained by combining the detection of the blinking frequency; the mean value of the blinking frequency zpl(1) to zpl(z) is obtained, and the mean frequency jpl is obtained; the real-time blinking frequency spl of the learner is obtained; if spl>jpl, it is judged that the learner's attention is not concentrated; the concentration of the learner is obtained according to the learner's attention;

[0097] Step S2: According to the personalized information of the learner, the continuous learning time of the learner is judged; according to the planning data, the learning goal of the learner is obtained, and the learning path of the learner is preliminarily planned according to the continuous learning time, and the first planning route is obtained.

[0098] Step S21: Obtain the personalized information of the learner when learning, and monitor the expression category, emotion intensity and concentration level of the learner in real time; according to the expression category, emotion intensity and concentration level of the learner when learning, the continuous learning time of the learner is judged;

[0099] Step S22: Obtain the learning goal of the learner, and decompose the learning goal of the learner according to the continuous learning time of the learner, and obtain the learning unit; the correlation between the learning units is obtained by big data, and the correlation value is obtained, and the learning path of the learner is preliminarily planned according to the correlation value;

[0100] Step S221: Obtain the learning goal of the learner and the continuous learning time of the learner; obtain the estimated learning time of the learning goal by big data; divide the estimated learning time into learning units according to the continuous learning time of the learner;

[0101] It should be noted that: the estimated learning time of the learning goal is obtained by big data, which means that the learning time of most people for the learning goal is obtained, such as the learning time of more than 100 learners for mathematical functions, the average of the results and the advantages and disadvantages, and the total learning time is added and summed, and the average learning time of each learner is obtained as the estimated learning time according to the number of learners.

[0102] Please refer to Figure 4 Step S222: Obtain the number of learning units xgs; obtain the required time dsj(1), dsj(2),..., dsj(xgs) of each learning unit of the learner by big data; combine the learning units two by two to obtain the combined unit; obtain the required time zh2 of the learner to learn the combined unit;

[0103] ;

[0104] It should be noted that: such as zh2(1,1), zh2(2,2) is only to express the data integrity, not involved in the specific calculation process; zh2(1,2) indicates learning the second learning unit after learning the first learning unit, the time required to learn two learning units;

[0105] The correlation value glz2 between the learning units is calculated by the time required by the learner for each learning unit dsj(1), dsj(2), …, dsj(xgs) and the time required by the learner for the combined unit zh2;

[0106] ;

[0107] It should be noted that: glz2(u,v) represents the correlation value between learning units u and v, zh2(u,v) represents the time required to learn the combined unit composed of learning unit u and learning unit v; dsj(u) refers to the time required to learn the u-th learning unit, and dsj(v) refers to the time required to learn the v-th learning unit; the correlation value is calculated according to the ratio of the time required to learn two learning units and the time required to learn two units respectively, the ratio reflects the reduced time of learning two learning units together, and the difference between the correlation value and the reduced time is taken to make the proportional relationship consistent;

[0108] Step S223: On the basis of the combined unit, the learning unit is added; the time required by the learner to learn the combined unit zh3 is obtained; taking zh2(1,2) as an example, zh3(1,2,3), zh3(1,2,4), …, zh3(1,2,xgs) are obtained; the correlation value is calculated according to the calculation formula of step S222, and glz3 is obtained; Similarly, zh4, zh5, …, zhxgs are obtained; the corresponding correlation values are calculated; the correlation values are accumulated, and the combined route corresponding to the maximum correlation value is taken as the first planning route;

[0109] Step S3: According to the first planning route, the learning completion degree of the learner is verified, and the individual information of the learner is obtained in real time, the learning state of the learner is analyzed by combining the learning completion degree of the learner, the review plan of the learner is arranged, and the first planning route is optimized according to the review plan;

[0110] Step S31: The learner learns according to the first planning route; the learning behavior of the learner is obtained by conducting in-class exercises on the learner during the learning process of the learner; the answer information of the learner is obtained according to the learning behavior of the learner; the learning completion degree of the learner is calculated according to the answer information;

[0111] Step S311: obtaining a total score value zfz of the instant exercise; obtaining a score value kfz of the instant exercise deducted for errors; obtaining a score value wfz of the instant exercise deducted for unanswered questions; calculating an error value cwz of the learner;

[0112] ;

[0113] It should be noted that: for the score value kfz deducted for errors, it indicates that the learner understands the question, but there are still problems; for the score value wfz deducted for unanswered questions, it indicates that the learner does not understand the question at all, and the two should be distinguished by different weights, and the present application classifies the score value deducted for errors by adding a weight of 1 / 2;

[0114] Step S312: obtaining an estimated time ygt of the instant exercise; obtaining a number of questions tsl; obtaining a hand video of the learner; recording the hand pause time tds of the learner when completing each question t ; obtaining the completion time wct of the instant exercise of the learner; calculating a stutter value kdz of the learner;

[0115] ;

[0116] It should be noted that: the hand pause time of the learner when completing each question refers to the stopping time of the learner in the question answering process from the time the learner starts answering the question to the end time period of the question. For example, for the first question, the learner starts answering at time 0 and ends answering at time 10; during the answering process from 0 to 10, the learner pauses at time 2-4 and time 6-8, so the hand pause time of the learner when completing each question is (4-2)+(8-6).

[0117] It should be noted that: the stutter value is preset to 1; the sign is extracted by the ratio of the difference between the completion time wct of the instant exercise of the learner and the estimated time ygt of the instant exercise to the absolute value, and when it is negative, it indicates that the learner learns well for the learning unit; the stutter value is reduced; if the completion time of the learner is long and the stutter time is long, the stutter value is enlarged.

[0118] Step S313: calculating the learning completion degree wcd of the learner according to the error value of the learner and the stutter value of the learner;

[0119] ;

[0120] Step S32: analyzing the learning state of the learner from the personalized information of the learner and the learning completion degree of the learner, calculating the review weight according to the learning state of the learner, arranging the review plan for the learner according to the review weight, and combining the review plan with the first planning route to optimize the first route.

[0121] Step S321: Obtain the personalized information of the learner, i.e., the time series data of the student state parameters generated by the multi-modal large model; from the time series data, extract the total time length qxt of the learner in a negative emotional state and the total time length zyt of the learner in an inattentive state during the learning process; obtain the total learning time xxt of the learner; take the ratio of the time of negative emotion of the learner in the learning process to the learning time of the learner as the first weight, and take the ratio of the time of inattention of the learner to the learning time of the learner as the second weight;

[0122] Step S322: Obtain the learning completion degree wcd of the learner; calculate in combination with the first weight and the second weight, multiply the learning completion degree by the first weight and the second weight to obtain the review weight fxq;

[0123] It should be noted that the learning completion degree, the first weight, the second weight and the review weight are in a proportional relationship, and when any of the learning completion degree, the first weight and the second weight increases, the review weight increases.

[0124] According to the review weight, arrange the time for reviewing the learning unit, sort the review weights in ascending order, and review in turn according to the sorting position of the review weights to obtain a review plan; supplement the first planning route with the review plan to optimize the first planning route to obtain an optimized route;

[0125] Step S4: Obtain the learning preset time length of the learner, and dynamically adjust the optimized route according to the learning preset time length to complete the route planning;

[0126] Step S41: Obtain the current learning time dqt, the learning preset time yst, and the sorting position of the current learning unit in the optimized route; calculate the learning progress according to the sorting position and the number of learning units; compare the ratio of the current learning time to the learning preset time with the learning progress;

[0127] Step S42: take the ratio of the current learning time to the learning preset time as the expected value; if the expected value is greater than the learning progress, adjust the optimized route:

[0128] Calculate the difference between the expected value and the learning progress to obtain a learning deviation; multiply the learning deviation by the learning preset time to obtain a deviation time, obtain the number of remaining learning units in the optimized route; divide the deviation time by the number of remaining learning units to obtain an adjustment time, and time-compress the learning unit in the optimized route according to the adjustment time to complete the route planning.

[0129] The above formulas are all dimensionless values calculated, the formulas are obtained by collecting a large amount of data to simulate a formula of the most recent real situation, preset parameters in the formula are set by a person skilled in the art according to actual conditions, such as weight coefficients and proportional coefficients, sizes of the weight coefficients and the proportional coefficients are set to obtain a specific value of quantization of each parameter, to facilitate subsequent comparison, as long as the sizes of the weight coefficients and the proportional coefficients do not affect the proportional relationship between the parameters and the quantized values.

[0130] Finally, it should be noted that the above-described embodiments are only specific implementations of the present application, which are used to illustrate the technical solutions of the present application, but not to limit the present application. The protection scope of the present application is not limited to this. Although the present application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that any person skilled in the art can modify or easily think of changes to the technical solutions recorded in the foregoing embodiments, or make equivalent replacements to some technical features, within the technical range disclosed by the present application. These modifications, changes or replacements do not make the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present application, and should be covered within the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the protection scope of the claims.

Claims

1. A learning path planning method based on habit analysis, characterized in that, Planning methods include: Step S1: Obtain learners' personalized information and planning data; The specific steps of step S1 are as follows: Step S11: Acquire real-time learning video data of learners through a video capture device, and preprocess the video data to generate high-quality video data; Step S12: Input the high-quality video data into the multimodal large model. The multimodal large model performs end-to-end deep visual understanding of the video frames and generates visual features including learner expressions, attention, and posture. The multimodal large model further integrates the visual features with synchronously acquired speech data to generate a comprehensive feature representation. Step S13: Based on the comprehensive feature performance, analyze the video sequence using a self-attention mechanism to generate quantified student state parameters, which constitute the learner's personalized information; Step S131: Perform facial expression detection on the learner; obtain the facial expression recognition dataset, classify facial expression categories, learn the inherent relationship between image patches of the same type of expression through a self-attention mechanism, and obtain the association value; perform statistics on the association value to obtain the association set tzj(i); construct the association function hs based on the association set; Acquire learners' real-time learning videos; acquire learners' facial images, obtain the association values ​​of learners' facial images based on the self-attention mechanism, and perform expression detection on learners using the association function to obtain the detection results; Step S1311: Obtain the association set tzj(i), extract the number of association values ​​t in the association set; denote the association values ​​in the association set as tzz(1), tzz(2), ..., tzz(t); Step S1312: Obtain the correlation values ​​tzz(1), tzz(2), ..., tzz(t); construct the correlation function hs based on the correlation values. The specific correlation function is as follows: ； It should be noted that x1, x2, ..., xt are not calculated values; they represent correlation values ​​in different directions. Step S1313: Obtain the association values ​​of the learner's facial images, denoted as xtz(1), xtz(2), ..., xtz(t); calculate the learner's association function xhs by combining it with the association function hs; ； The learner's correlation function xhs and correlation function hs are calculated by difference, and the sum of the absolute values ​​of the differences in the x1, x2, ..., xt directions is obtained. ； For example, xtz(1)-tzz(1) is the difference in the x1 direction; The absolute values ​​of the differences in the x1, x2, ..., xt directions are summed to obtain the detection value; the expression category corresponding to the smallest detection value is obtained; the emotional intensity of the learner is obtained based on the self-attention mechanism; and the expression information of the learner is obtained by combining the expression category and the emotional intensity. Step S132: Obtain the learner's visual learning area and record it as the attention concentration area; perform gaze detection on the learner; determine whether the learner's attention is focused, and optimize the attention judgment by combining the learner's blink frequency to obtain the learner's attention information; Step S2: Based on the learner's personalized information, determine the learner's continuous learning duration; based on the continuous learning duration and combined with the planning data, make a preliminary plan for the learner's learning path to obtain the first planned route; Step S3: Based on the first planned route, check the learner's learning completion rate. Based on the learner's learning completion rate and combined with the learner's personalized information, analyze the learner's learning status, arrange a review plan for the learner, and optimize the first planned route based on the review plan. Step S4: Dynamically adjust the optimized route based on the planning data to complete the route planning.

2. The learning path planning method based on habit analysis according to claim 1, characterized in that, The multimodal large model adopts a Transformer-based architecture, which segments video frames into image blocks and directly inputs them into the encoder. It learns the intrinsic relationships between image blocks through a self-attention mechanism to understand the image content holistically, thus eliminating the need to pre-extract facial key points or design handcrafted features.

3. The learning path planning method based on habit analysis according to claim 1, characterized in that, The student status parameters include, but are not limited to: the student's facial expression type; level of concentration; emotional intensity; and learning behavior.

4. The learning path planning method based on habit analysis according to claim 1, characterized in that, The specific steps of step S2 are as follows: Step S21: Obtain personalized information about the learner during learning, and monitor the learner's facial expressions, emotional intensity, and concentration level in real time; determine the learner's continuous learning duration based on the learner's facial expressions, emotional intensity, and concentration level during learning. Step S22: Obtain the learner's learning objectives, and decompose the learner's learning objectives into learning units based on the learner's continuous learning time; The correlation between learning units is calculated to obtain correlation values, and the learning path of learners is initially planned based on the correlation values.

5. The learning path planning method based on habit analysis according to claim 4, characterized in that, The specific steps of step S22 are as follows: Step S221: Obtain learner learning objectives and learner's continuous learning duration; obtain the estimated learning time for learning objectives; divide the estimated learning time into learning units according to the learner's continuous learning time; Step S222: Obtain the number of learning units xgs; obtain the time required for each learning unit dsj(1), dsj(2), ..., dsj(xgs); combine the learning units in pairs to obtain combined units; obtain the time required for the learner to learn the combined units zh2; The correlation value glz2 between learning units is calculated from the time required for each learning unit by the learner, dsj(1) to dsj(xgs), and the time required for the learner to learn a combination of units, zh2. ； Where: glz2(u,v) represents the association value between learning units u and v; zh2(u,v) represents the time required for the combined unit formed by learning units u and v; dsj(u) refers to the time required to learn the u-th learning unit, and dsj(v) refers to the time required to learn the v-th learning unit. Step S223: Based on the combined units, add learning units; obtain the time required for the learner to learn the combined units zh3; calculate the correlation value according to the calculation formula in step S222 to obtain glz3; similarly obtain zh4 to zhxgs; calculate the corresponding correlation value; accumulate the correlation values ​​and take the combined route corresponding to the largest correlation value as the first planning route.

6. The learning path planning method based on habit analysis according to claim 1, characterized in that, The specific steps of step S3 are as follows: Step S31: The learner learns according to the first planned route; during the learning process, the learner performs in-class exercises to obtain the learner's learning behavior; based on the learner's learning behavior, the learner's answer information is obtained; based on the answer information, the learner's learning completion rate is calculated. Step S32: Analyze the learner's learning status based on the learner's personalized information and learning completion rate, calculate the review weight based on the learner's learning status, and arrange a review plan for the learner based on the review weight; Combine the review plan with the first planned route and optimize the first route.

7. The learning path planning method based on habit analysis according to claim 6, characterized in that, The specific steps of step S31 are as follows: Step S311: Obtain the total score of the in-class exercises zfz; obtain the score deducted by the learner for errors in the in-class exercises kfz; obtain the score deducted by the learner for not answering questions in the in-class exercises wfz; calculate the learner's error value cwz; Step S312: Obtain the estimated time ygt for in-class exercises; obtain the number of questions tsl; acquire video recordings of the learners' hand movements; record the pause time tds of the learners' hands when completing each question. t ; Obtain the learner's completion time wct for in-class exercises; Calculate the learner's stuttering value kdz; ； Step S313: Calculate the learner's learning completion rate based on the learner's error value and stuttering value.

8. The learning path planning method based on habit analysis according to claim 6, characterized in that, The specific steps of step S32 are as follows: Step S321: Obtain the learner's personalized information, namely, the time series data of student state parameters generated by the multimodal large model; extract the total duration qxt of the learner's negative emotional state and the total duration zyt of the learner's inattention state during the learning process from the time series data; obtain the learner's total learning time xt; use the ratio of the learner's negative emotional time to the learner's learning time as the first weight, and the ratio of the learner's inattention time to the learner's learning time as the second weight; Step S322: Obtain the learner's learning completion rate wcd; calculate the review weight fxq by combining the first weight and the second weight, and multiply the learning completion rate with the first weight and the second weight. The study units are scheduled for review based on their review weights. The review weights are sorted in ascending order, and the review is carried out in order of their review weights to obtain a review plan. The review plan is used as a supplement to the first planned route to optimize it, resulting in an optimized route.

9. The learning path planning method based on habit analysis according to claim 1, characterized in that, The specific steps of step S4 are as follows: Step S41: Obtain the current learning time dqt, the preset learning time yst, and the ranking of the current learning unit in the optimization route; calculate the learning progress by proportionally calculating the ranking and the number of learning units; compare the ratio of the current learning time to the preset learning time with the learning progress. Step S42: Record the ratio of the current learning time to the preset learning time as the expected value; if the expected value is greater than the learning progress, adjust the optimized route; The learning deviation is calculated by comparing the expected value with the learning progress. The learning deviation is then calculated with the preset learning duration to obtain the deviation duration, and the number of remaining learning units in the optimized route is obtained. Based on the deviation duration and the number of remaining learning units, the adjustment duration is obtained, and the learning units in the optimized route are compressed according to the adjustment duration to complete the route planning.

Images