Autonomous learning auxiliary system based on Chinese education

Abstract

The invention discloses an autonomous learning assistance system based on Chinese education, which relates to the technical field of learning assistance and comprises a master node analysis unit, a user tag generation unit, a slave node analysis unit, a recommendation output unit, a Chinese character recognition unit and a Chinese character detection unit. According to the invention, the recommendation output unit is utilized to push different numbers of video introduction data to each user according to the slave node recommendation coefficient corresponding to each master node, so that the screened video can be ensured to accord with the learning ability and acceptance degree of the current user to a great extent; meanwhile, the Chinese character recognition unit and the Chinese character detection unit can detect single Chinese character image data provided by the user, compare Chinese characters in the image with Chinese character templates in a database, and inspect the quality of the Chinese characters from multiple aspects, so that the system can reflect the Chinese character writing quality of the user more accurately, and the user experience is improved. And the Chinese learning efficiency of the user is effectively improved.

Description

Technical Field

[0001] The present invention relates to the technical field of learning assistance, and specifically to an autonomous learning assistance system based on Chinese education. Background Art

[0002] The autonomous learning assistance system based on Chinese education provides functions such as personalized learning recommendations, intelligent evaluation, and rich resources, which is conducive to different users learning Chinese more efficiently, thereby enhancing the ability of autonomous learning. The invention patent with the application number 202310527982.9 discloses a "language autonomous learning system based on big data voice interaction, which relates to the technical field of language learning; and the system includes a user learning information collection module, a user word mastery analysis module, a user oral language ability analysis module, a user foreign language level analysis module, a user listening article screening module, a user listening article playback module, and a cloud database; by screening the corresponding recommended listening articles for the target user according to the foreign language ability level of the target user, and then realizing the automatic playback of the listening articles, it solves the problem of insufficient intelligence in the playback of listening articles in the current technology, realizes the intelligent selection and automatic playback of listening articles on the learning machine, ensures the adaptability of the listening articles and the user's foreign language ability, thereby improving the effect of the user's foreign language listening practice, effectively ensuring the efficiency of the user's self-study, and also improving the convenience of the user during self-study".

[0003] The above-mentioned prior art solves problems such as the inability to enhance the user experience and the sense of interaction with the learning machine. However, when the system is running, since the learning degree value of the user is not analyzed, it is impossible to accurately reflect the current user's language learning ability, and the difficulty coefficients of different videos are not evaluated, resulting in the pushed content not meeting the acceptance level of the user. At the same time, the quality of writing cannot be reasonably and reliably detected. Summary of the Invention

[0004] The purpose of the present invention is to provide an autonomous learning assistance system based on Chinese education to solve the problems raised in the above background art.

[0005] To achieve the above purpose, the present invention provides the following technical solution: An autonomous learning assistance system based on Chinese education, including a recommendation output unit, a Chinese character recognition unit, and a Chinese character detection unit; A main node analysis unit, the main node analysis unit obtains all Chinese education video data and corresponding user data in the online course platform, constructs a plurality of main nodes and slave nodes according to the user data and video data, calculates after counting the video viewing quantity, viewing progress rate, test accuracy rate of the main nodes and the total number of slave nodes, obtains the learning degree value of the main nodes, sets the initial in-group center point, and adds each main node to different groups according to the learning degree value and video viewing quantity; The user label generation unit adds all the master nodes to different groups, and then calculates new in-group central points according to the learning degree values and video viewing quantities of all the master nodes in the groups. If all the new in-group central points are the same as the original in-group central points, it determines the number of master nodes in each group, sets the adjustment coefficient and three types of label information, and analyzes the label information of all the master nodes in the corresponding group according to the adjustment coefficient, the learning degree values of the in-group central points, and the video viewing quantities. The slave node analysis unit, after counting the total number of viewers of each slave node and the viewing duration, playback times, and test scores of each video among the master nodes, calculates the difficulty coefficient of the slave node by using the viewing duration, playback times, test scores of each video among the master nodes and the total number of viewers of the slave node, and uses the difficulty coefficient to count the known node viewing progress values and difficulty difference vectors corresponding to the master nodes, and then analyzes the viewing progress values and difficulty difference vectors to obtain the attention distribution matrix of the master nodes.

[0006] Preferably, the master node analysis unit includes a data collection module, a node construction module, a degree calculation module, and a grouping generation module. The data collection module obtains all the Chinese education video data and the corresponding user data in the online course platform. The video data includes video number, total video duration, total number of viewers, and total test score. The user data includes user number, video viewing quantity, viewing duration of different videos, playback times, and test scores. The node construction module constructs multiple master nodes according to the user data and constructs multiple slave nodes by using the video data , where is the total number of master nodes, is the total number of slave nodes. After counting the video viewing quantity of the th master node , the viewing duration of each video, the test score , and the corresponding total video duration and total test score , according to and it calculates the viewing progress rate of , where . By using in and in it analyzes the test accuracy rate of , where . is the main node number, and the degree calculation module sets three weight coefficients After that, through the number of video views and the viewing progress rate and the test accuracy rate as well as the total number of slave nodes calculate the learning degree value , where , after the grouping generation module repeats the operation until the learning degree value of each main node is determined, set three initial in-group center points, calculate the distance value between each main node and different in-group center points according to the learning degree value and the number of video views, and add each main node to the group where the center point with the smallest distance value is located.

[0007] Preferably, the user label generation unit includes a center point update module, a node number determination module, and a label addition module. After the center point update module adds all the main nodes to different groups, it calculates the current in-group average value according to the learning degree value and the number of video views of all the main nodes in the group, and determines the new in-group center point by using the in-group average value of each group. If the node number determination module finds that there is a new in-group center point that is inconsistent with the original in-group center point, it calculates the distance value between each main node and the current in-group center point, adds the node to the group where the center point with the smallest distance value is located, and then calculates the new in-group center point according to the main nodes in the group. If all the new in-group center points are consistent with the original in-group center points, it determines the number of main nodes in each group. The label addition module sets an adjustment coefficient and three types of label information, namely, positive user label, stable user label, and negative user label. According to the adjustment coefficient, the learning degree value of each group center point, and the number of video views, it calculates the actual value of the corresponding group, adds the label information of positive users to all the main nodes in the group with the largest actual value, adds the label information of negative users to all the main nodes in the group with the smallest actual value, and adds the label information of stable users to all the main nodes in the remaining groups.

[0008] Preferably, the slave node analysis unit includes a difficulty coefficient calculation module, a known matrix determination module, a participation vector calculation module, and a distribution determination module. After the difficulty coefficient calculation module counts the total number of viewers of each slave node and the viewing duration, playback times, and test scores of each video in the master node, it calculates the participation value, completion rate, playback rate, and scoring rate corresponding to each slave node by using the viewing duration, playback times, test scores of each video in the master node and the total number of viewers of the slave node, sets three weight coefficients, and calculates the difficulty coefficient of the slave node according to the weight coefficients and the completion rate, playback rate, and scoring rate corresponding to each slave node. The known matrix determination module constructs a corresponding attribute vector according to the participation value, completion rate, playback rate, scoring rate, and difficulty coefficient of each slave node, combines the attribute vectors of all slave nodes to obtain a global matrix, counts the video numbers with non-zero viewing duration in the master node, determines the corresponding slave nodes according to the video numbers, takes the slave nodes as the known nodes of the current master node, and combines the attribute vectors of the known nodes to obtain the known matrix of the master node. The participation vector calculation module sets a corresponding identity matrix according to the number of parameters in the slave node attribute vector, combines the known matrix of each master node with the identity matrix to obtain the transformation matrix of the master node, extracts the total video duration of the known nodes and the viewing durations of different videos of the master node, determines the viewing progress values of all the known nodes corresponding to the master node according to the ratio between the viewing duration and the total duration, analyzes the content revenue vector of the master node by using the viewing progress value and the known matrix, and calculates the content revenue vector of the master node and the transformation matrix to determine the attribute participation vector of the master node. The distribution determination module counts the difficulty coefficients of all the known nodes, calculates the distance values between the difficulty coefficients of the known nodes and the difficulty coefficients of all the slave nodes, constructs a difficulty difference vector according to the multiple distance values of the known nodes, and jointly analyzes all the difficulty difference vectors and the viewing progress values of the known nodes to obtain the attention distribution matrix of the current master node.

[0009] Preferably, the recommended output unit includes a selection coefficient calculation module, an association matrix calculation module, a recommended coefficient calculation module, and an introduction data query module. After the selection coefficient calculation module determines the attention distribution matrix, learning degree value, and the total number of slave nodes of the master node, it calculates according to the attention distribution matrix, learning degree value, and the total number of slave nodes to obtain the selection coefficient of the master node. The association matrix calculation module analyzes the association matrix of the master node through the global matrix corresponding to all slave nodes and the transformation matrix of the master node. The recommended coefficient calculation module counts the attribute participation vector, association matrix, selection coefficient of each master node, and the global matrix of the slave nodes, and then analyzes them using the recommended vector analysis algorithm to obtain the slave node recommended coefficient corresponding to each master node. After the introduction data query module sets the corresponding recommended number according to the tag information, it screens multiple slave nodes for each master node according to the recommended coefficient and the recommended number, determines the number after according to the video data corresponding to the slave node, and queries in the database using the video number to obtain the introduction data of the video, and transmits it to the user recommendation interface. The recommended vector analysis algorithm is specifically: Wherein, represents the global matrix corresponding to all slave nodes, represents the known matrix of the th master node , represents the transformation matrix of the th master node , represents the viewing progress vector of the known nodes of the th master node , represents the attribute participation vector of the th master node , represents the selection coefficient of the th master node , represents the attention distribution matrix of the th master node , represents the learning degree value of the th master node , represents the number of slave nodes, represents the th master node 's association matrix, represents the th master node 's recommended coefficient vector, represents the th master node The known matrix after transposition, represents the identity matrix, represents the main node number, represents the th main node corresponding slave node recommendation coefficient vector.

[0010] Preferably, the Chinese character recognition unit includes a field setting module, a pixel point marking module, an actual type determination module, and a similarity calculation module. After the field setting module obtains the single Chinese character image data provided by the user, it performs denoising and grayscale operations on the image data to obtain a grayscale image, sets the size of the field box. Arbitrarily select an unmarked pixel point as the center point of the field box. After counting the grayscale values and relative positions of all pixel points within the box, calculate the cumulative value of each pixel point within the box according to the grayscale values and relative positions of the pixel points. Calculate the field average value and standard deviation using the cumulative values of all pixel points in the current field box. After the pixel point marking module sets the range value, analyze the corresponding field threshold using the field average value, standard deviation, and range value, and determine the grayscale value of the center point of the field box. If the grayscale value is greater than the field threshold, set the grayscale value of the center point to the maximum brightness value; otherwise, set the grayscale value of the center point to zero, and mark the current center point. Repeat the operation until all pixel points are marked, thereby obtaining the processed image data. The actual type determination module obtains all the processed image data, transmits it to the CRNN model, ResNet model, and Transformer model for analysis. After obtaining three prediction results, take the Chinese character category corresponding to the maximum probability value in the prediction results as the actual type of the image data. The similarity calculation module queries in the database according to the actual type to obtain the handwritten Chinese character template corresponding to the Chinese character type. After extracting the attribute representations of the Chinese characters in the image data and the attribute representations of the three Chinese character templates, calculate the similarities between the attribute representations of the Chinese characters in the image data and the attribute representations of the three Chinese character templates respectively. Set the adjustment coefficient, and use the similarity analysis algorithm to combine the adjustment coefficient with the three similarities to obtain the actual similarity between the Chinese characters in the current image data and the template.

[0011] Preferably, the Chinese character detection unit includes a gradient magnitude determination module and an average value calculation module. After the gradient magnitude determination module performs a convolution operation on the image data using a Gaussian filter, it determines the gradient values of the image data in different directions, calculates the gradient direction and gradient magnitude of each pixel point according to the gradient values, determines the adjacent pixel points of each pixel point according to the gradient direction, and then judges the magnitude of the gradient magnitude between the current pixel point and the adjacent pixel points. If the gradient magnitude of the current pixel point is greater than that of the adjacent pixel points, the gradient magnitude of the current pixel point is retained; otherwise, the gradient magnitude is set to zero. After the average value calculation module statistically calculates the gradient magnitudes of all pixel points in the image data, it calculates the average value and standard deviation corresponding to the image using the gradient magnitudes, accumulates the average value and the standard deviation to obtain a first threshold value, and takes the difference between the average value and the standard deviation as the second threshold value. If the gradient magnitude of a pixel point is greater than the first threshold value, it is determined that the current pixel point is an edge point; if the gradient magnitude is less than or equal to the first threshold value and greater than the second threshold value, it is determined that the current pixel point is a candidate point; if the gradient magnitude is less than or equal to the second threshold value, it is determined that the current pixel point is not an edge point.

[0012] Preferably, the Chinese character detection unit further includes a candidate point analysis module, a dimension setting module, and a determination result generation module. After the candidate point analysis module sets a window, it uses the candidate point as the center point of the window and statistically calculates the number of edge points within the window. If the number is zero, it is determined that the current candidate point is not an edge point; otherwise, it is determined that the current candidate point is an edge point. Connect all the edge points to the nearest edge point to determine the contour features of the current image data. After the dimension setting module determines the length, width, and area of the Chinese character according to the contour features in the image data, it calculates the actual dimension of the Chinese character in the image data using the adjustment coefficient and the length, width, and area of the Chinese characters in the three templates. The determination result generation module statistically calculates the actual similarity and dimension of the Chinese characters in the image. If the actual similarity and dimension of the Chinese characters in the image are both greater than the threshold value, it is prompted that the current Chinese character is qualified; otherwise, it is prompted that the current Chinese character is unqualified.

[0013] Compared with the prior art, the beneficial effects of the present invention are: 1. The present invention constructs multiple master nodes and slave nodes by the master node analysis unit based on user data and video data, facilitating subsequent analysis of the relationship between the master nodes and the slave nodes. Moreover, the learning degree value of the master node can more accurately reflect the current user's Chinese learning ability. Meanwhile, after the user label generation unit divides each master node into different groups according to the learning degree value and the number of video views, and adds corresponding label information to the master nodes in these groups. This design, on the one hand, effectively reflects the user's enthusiasm for Chinese learning, and on the other hand, facilitates subsequent flexible adjustment of the number of video recommendations, enabling users with negative user labels to have a wider range of choices, thereby stimulating the interest in learning, while users with positive user labels and stable user labels can obtain more accurate and reasonable recommended content; 2. The present invention analyzes the difficulty difference vectors of different slave nodes by the slave node analysis unit based on the attribute vectors of the slave nodes and the video viewing duration of the master nodes, thereby showing the learning difficulties of different videos and the acceptance degree of users. The recommendation output unit pushes different numbers of video introduction data for each user according to the slave node recommendation coefficient corresponding to each master node, ensuring that the selected videos can largely conform to the current user's learning ability and acceptance degree. At the same time, the Chinese character recognition unit and the Chinese character detection unit can detect the single Chinese character image data provided by the user, compare the Chinese characters in the image with the Chinese character templates in the database, and examine the quality of the Chinese characters from multiple aspects to ensure that the system can more accurately reflect the user's Chinese character writing quality and effectively improve the user's Chinese learning efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] Figure 1 It is a schematic diagram of the overall system flow provided by an embodiment of the present invention; Figure 2 It is an internal module block diagram of the master node analysis unit provided by an embodiment of the present invention; Figure 3 It is an internal module block diagram of the user label generation unit provided by an embodiment of the present invention; Figure 4 It is an internal module block diagram of the recommendation output unit provided by an embodiment of the present invention; Figure 5 It is an internal module block diagram of the Chinese character recognition unit provided by an embodiment of the present invention; Figure 6 It is an internal module block diagram of the Chinese character detection unit provided by an embodiment of the present invention.

[0015] In the figure: 1. Main node analysis unit; 101. Data collection module; 102. Node construction module; 103. Degree calculation module; 104. Group generation module; 2. User label generation unit; 201. Central point update module; 202. Node number determination module; 203. Label addition module; 3. Slave node analysis unit; 301. Difficulty coefficient calculation module; 302. Known matrix determination module; 303. Participation vector calculation module; 304. Distribution determination module; 4. Recommendation output unit; 401. Selection coefficient calculation module; 402. Correlation matrix calculation module; 403. Recommendation coefficient calculation module; 404. Introduction data query module; 5. Chinese character recognition unit; 501. Field setting module; 502. Pixel point marking module; 503. Actual type determination module; 504. Similarity calculation module; 6. Chinese character detection unit; 601. Gradient amplitude determination module; 602. Mean value calculation module; 603. Candidate point analysis module; 604. Dimension setting module; 605. Judgment result generation module. Detailed implementation manner

[0016] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0017] Please refer to Figures 1 - 6 , the present invention provides a technical solution: an autonomous learning assistance system based on Chinese education, including a recommendation output unit 4, a Chinese character recognition unit 5, and a Chinese character detection unit 6; The main node analysis unit 1, the main node analysis unit 1 obtains all Chinese education video data and corresponding user data in the online course platform, constructs a plurality of main nodes and slave nodes according to the user data and video data, and after counting the video viewing quantity, viewing progress rate, test accuracy of the main nodes and the total number of slave nodes, calculates them to obtain the learning degree value of the main nodes, sets the initial in-group central point, and adds each main node to different groups according to the learning degree value and video viewing quantity; The user label generation unit 2, after adding all the main nodes to different groups, calculates a new in-group central point according to the learning degree value and video viewing quantity of all the main nodes in the group. If all the new in-group central points are the same as the original in-group central points, then determines the number of main nodes in each group, sets the adjustment coefficient and three types of label information, and analyzes the label information of all the main nodes in the corresponding group according to the adjustment coefficient, the learning degree value of each in-group central point, and the video viewing quantity; After the slave node analysis unit 3 counts the total number of viewers of each slave node and the viewing duration, playback times, and test scores of each video in the master node, it calculates the difficulty coefficient of the slave node using the viewing duration, playback times, test scores of each video in the master node and the total number of viewers of the slave node. After statistically obtaining the known node viewing progress value and difficulty difference vector corresponding to the master node using the difficulty coefficient, it analyzes the viewing progress value and difficulty difference vector to obtain the attention distribution matrix of the master node.

[0018] The master node analysis unit 1 includes a data collection module 101, a node construction module 102, a degree calculation module 103, and a grouping generation module 104. The data collection module 101 obtains all Chinese education video data and corresponding user data in the online course platform. The video data includes video number, total video duration, total number of viewers, and total test score. The user data includes user number, number of videos viewed, viewing duration of different videos, playback times, and test scores. The node construction module 102 constructs multiple master nodes according to the user data , and constructs multiple slave nodes using the video data , where is the total number of master nodes, is the total number of slave nodes. After counting the number of videos viewed of the master node , the viewing duration of each video, the test score , and the corresponding total video duration , the total test score , then according to and calculates the viewing progress rate , where , using in and in analyzes the test accuracy rate , where , is the master node number. The degree calculation module 103 sets three weight coefficients , and then through the number of videos viewed , the viewing progress rate , the test accuracy rate , and the total number of slave nodes calculates the learning degree value , where , after the grouping generation module 104 repeats the operation until the learning degree values of each master node are determined, three initial within-group center points are set, the distance values between each master node and different within-group center points are calculated according to the learning degree value and the number of video views, and each master node is added to the group where the center point with the smallest distance value is located; The user label generation unit 2 includes a center point update module 201, a node number determination module 202, and a label addition module 203. After the center point update module 201 adds all master nodes to different groups, the current within-group average value is calculated according to the learning degree values and the number of video views of all master nodes within the group, and the new within-group center points are determined using the within-group average values of each group. If there is a new within-group center point that is inconsistent with the original within-group center point, the node number determination module 202 calculates the distance value between each master node and the current within-group center point, adds the node to the group where the center point with the smallest distance value is located, and then calculates the new within-group center point according to the master nodes within the group. If all new within-group center points are consistent with the original within-group center points, the number of master nodes within each group is determined. The label addition module 203 sets the adjustment coefficient and three types of label information, where the three types of label information are positive user labels, stable user labels, and negative user labels. The actual value of the corresponding group is calculated according to the adjustment coefficient, the learning degree value of each within-group center point, and the number of video views. The label information of positive users is added to all master nodes within the group with the largest actual value, the label information of negative users is added to all master nodes within the group with the smallest actual value, and the label information of stable users is added to all master nodes within the remaining groups; The slave node analysis unit 3 includes a difficulty coefficient calculation module 301, a known matrix determination module 302, a participation vector calculation module 303, and a distribution determination module 304. After the difficulty coefficient calculation module 301 counts the total number of viewers of each slave node and the viewing duration, playback times, and test scores of each video in the master node, it calculates the participation value, completion rate, playback rate, and scoring rate corresponding to each slave node by using the viewing duration, playback times, test scores of each video in the master node and the total number of viewers of the slave node. Then, it sets three weight coefficients, and calculates the difficulty coefficient of the slave node according to the weight coefficients and the completion rate, playback rate, and scoring rate corresponding to each slave node. The known matrix determination module 302 constructs a corresponding attribute vector according to the participation value, completion rate, playback rate, scoring rate, and difficulty coefficient of each slave node, combines the attribute vectors of all slave nodes to obtain a global matrix, counts the video numbers with non-zero viewing duration in the master node, determines the corresponding slave nodes according to the video numbers, takes the slave nodes as the known nodes of the current master node, and combines the attribute vectors of the known nodes to obtain the known matrix of the master node. The participation vector calculation module 303 sets a corresponding identity matrix according to the number of parameters in the slave node attribute vector, combines the known matrix of each master node with the identity matrix to obtain the transformation matrix of the master node, extracts the total video duration of the known nodes and the viewing durations of different videos of the master node, determines the viewing progress values of all known nodes corresponding to the master node according to the ratio between the viewing duration and the total duration, and then analyzes the content revenue vector of the master node by using the viewing progress values and the known matrix. By calculating the content revenue vector of the master node and the transformation matrix, the attribute participation vector of the master node is determined. The distribution determination module 304 counts the difficulty coefficients of all known nodes, calculates the distance values between the difficulty coefficients of the known nodes and the difficulty coefficients of all slave nodes, constructs a difficulty difference vector according to the multiple distance values of the known nodes, and jointly analyzes all difficulty difference vectors and the viewing progress values of the known nodes to obtain the attention distribution matrix of the current master node; The recommended output unit 4 includes a selection coefficient calculation module 401, an association matrix calculation module 402, a recommended coefficient calculation module 403, and an introduction data query module 404. After the selection coefficient calculation module 401 determines the attention distribution matrix, learning degree value, and the total number of slave nodes of the master node, it calculates based on the attention distribution matrix, learning degree value, and the total number of slave nodes to obtain the selection coefficient of the master node. The association matrix calculation module 402 analyzes the association matrix of the master node through the global matrix corresponding to all slave nodes and the transformation matrix of the master node. After the recommended coefficient calculation module 403 counts the attribute participation vector, association matrix, selection coefficient of each master node, and the global matrix of the slave nodes, it analyzes them using the recommended vector analysis algorithm to obtain the slave node recommended coefficient corresponding to each master node. After the introduction data query module 404 sets the corresponding recommended number according to the label information, it screens out multiple slave nodes for each master node according to the recommended coefficient and the recommended number, determines the number after the video data corresponding to the slave node, and queries in the database using the video number to obtain the introduction data of the video, and transmits it to the user recommendation interface. The recommended vector analysis algorithm is specifically: Among them, represents the global matrix corresponding to all slave nodes, represents the known matrix of the th master node , represents the transformation matrix of the th master node , represents the viewing progress vector of the known nodes of the th master node , represents the attribute participation vector of the th master node , represents the selection coefficient of the th master node , represents the attention distribution matrix of the th master node , represents the learning degree value of the th master node , represents the number of slave nodes, represents the th master node 's association matrix, represents the th master node 's recommended coefficient vector, represents the th master node The known matrix after transposition, denotes the identity matrix, denotes the main node number, denotes the th main node corresponding slave node recommendation coefficient vector; The Chinese character recognition unit 5 includes a domain setting module 501, a pixel point marking module 502, an actual type determination module 503, and a similarity calculation module 504. After the domain setting module 501 obtains the single Chinese character image data provided by the user, it performs denoising and grayscale operations on the image data to obtain a grayscale image, sets the size to of the domain box. Arbitrarily select an unmarked pixel point as the center point of the domain box. After counting the grayscale values and relative positions of all pixel points within the box, calculate the cumulative value of each pixel point within the box according to the grayscale values and relative positions of the pixel points. Calculate the domain average value and standard deviation using the cumulative values of all pixel points in the current domain box. After the pixel point marking module 502 sets the range value, analyze the corresponding domain threshold using the domain average value, standard deviation, and range value, and determine the grayscale value of the center point of the domain box. If the grayscale value is greater than the domain threshold, set the grayscale value of the center point to the maximum brightness value; otherwise, set the grayscale value of the center point to zero and mark the current center point. Repeat the operation until all pixel points are marked, thus obtaining the processed image data. The actual type determination module 503 obtains all the processed image data, transmits it to the CRNN model, ResNet model, and Transformer model for analysis. After obtaining three prediction results, take the Chinese character category corresponding to the maximum probability value in the prediction results as the actual type of the image data. The similarity calculation module 504 queries in the database according to the actual type to obtain the handwritten Chinese character template corresponding to the Chinese character type. After extracting the attribute representations of the Chinese characters in the image data and the attribute representations of the three Chinese character templates, calculate the similarities between the attribute representations of the Chinese characters in the image data and the attribute representations of the three Chinese character templates respectively. Set the adjustment coefficient, and use the similarity analysis algorithm to combine the adjustment coefficient with the three similarities to obtain the actual similarity between the Chinese characters in the current image data and the template. The similarity analysis algorithm is specifically: where, denotes the actual similarity, denotes the adjustment coefficient corresponding to the CRNN model, denotes the adjustment coefficient corresponding to the ResNet model, denotes the adjustment coefficient corresponding to the Transformer model, denotes the attribute representation of the Chinese character in the image data corresponding to the CRNN model, It represents the Chinese character template attribute representation corresponding to the CRNN model. It represents the Chinese character attribute representation in the image data corresponding to the ResNet model. It represents the Chinese character template attribute representation corresponding to the ResNet model. It represents the Chinese character attribute representation in the image data corresponding to the Transformer model. It represents the Chinese character template attribute representation corresponding to the Transformer model. The Chinese character detection unit 6 includes a gradient magnitude determination module 601 and an average value calculation module 602. After the gradient magnitude determination module 601 performs a convolution operation on the image data using a Gaussian filter, it determines the gradient values of the image data in different directions, calculates the gradient direction and gradient magnitude of each pixel point according to the gradient values, determines the adjacent pixel points of each pixel point according to the gradient direction, and then judges the magnitude of the gradient magnitude between the current pixel point and the adjacent pixel points. If the gradient magnitude of the current pixel point is greater than that of the adjacent pixel points, the gradient magnitude of the current pixel point is retained; otherwise, the gradient magnitude is set to zero. After the average value calculation module 602 counts the gradient magnitudes of all pixel points in the image data, it calculates the average value and standard deviation corresponding to the image using the gradient magnitudes, accumulates the average value and the standard deviation to obtain a first threshold, and takes the difference between the average value and the standard deviation as the second threshold. If the gradient magnitude of a pixel point is greater than the first threshold, it is determined that the current pixel point is an edge point; if the gradient magnitude is less than or equal to the first threshold and greater than the second threshold, it is determined that the current pixel point is a candidate point; if the gradient magnitude is less than or equal to the second threshold, it is determined that the current pixel point is not an edge point. The Chinese character detection unit 6 further includes a candidate point analysis module 603, a dimension setting module 604, and a determination result generation module 605. After the candidate point analysis module 603 sets a window, it takes the candidate point as the center point of the window and counts the number of edge points within the window. If the number is zero, it is determined that the current candidate point is not an edge point; otherwise, it is determined that the current candidate point is an edge point. Connect all the edge points to the nearest edge point to determine the contour features of the current image data. After the dimension setting module 604 determines the length, width, and area of the Chinese character according to the contour features in the image data, it calculates the actual dimension of the Chinese character in the image data using the adjustment coefficient and the length, width, and area of the Chinese characters in the three templates. The determination result generation module 605 counts the actual similarity and dimension of the Chinese characters in the image. If both the actual similarity and dimension of the Chinese characters in the image are greater than the threshold, it is prompted that the current Chinese character is qualified; otherwise, it is prompted that the current Chinese character is unqualified.

[0019] Working principle: In the present invention, the data collection module 101 in the master node analysis unit 1 acquires video data and user data. The node construction module 102 determines the master node and slave nodes. The degree calculation module 103 analyzes the learning degree value of the master node. The grouping generation module 104 adds each master node to different groups. The center point update module 201 in the user label generation unit 2 determines a new in-group center point according to the in-group average value of each group. The node number determination module 202 analyzes the number of master nodes in each group. The label adding module 203 adds different label information to the master nodes in each group. The difficulty coefficient calculation module 301 in the slave node analysis unit 3 calculates the difficulty coefficient of the slave node according to the weight coefficient and the completion rate, playback rate, and scoring rate corresponding to each slave node. The known matrix determination module 302 obtains the known matrix of the master node. The participation vector calculation module 303 determines the attribute participation vector of the master node. The distribution determination module 304 calculates the attention distribution matrix of the master node. The selection coefficient calculation module 401 in the recommendation output unit 4 obtains the selection coefficient of the master node. The correlation matrix calculation module 402 analyzes the correlation matrix of the master node. The recommendation coefficient calculation module 403 determines the slave node recommendation coefficient corresponding to each master node. The introduction data query module 404 transmits the introduction data of the video to the user recommendation interface. The domain setting module 501 in the Chinese character recognition unit 5 calculates the domain average value and standard deviation. The pixel point marking module 502 obtains the processed image data. The actual type determination module 503 takes the Chinese character category corresponding to the maximum probability value in the prediction result as the actual type of the image data. The similarity calculation module 504 analyzes the actual similarity between the Chinese characters in the current image data and the template. The gradient amplitude determination module 601 in the Chinese character detection unit 6 calculates the gradient direction and gradient amplitude of each pixel point. The mean value calculation module 602 analyzes the average value and standard deviation corresponding to the image. The candidate point analysis module 603 determines the contour features of the current image data. The actual size of the Chinese characters in the image data is calculated according to the size setting module 604. The detection result of the Chinese characters in the image is output through the determination result generation module 605.

[0020] It should be noted that in this article, relational terms such as first and second are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the term "comprising", "including" or any other variation thereof is intended to cover non-exclusive inclusion, so that a process, method, article or device comprising a series of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article or device.

[0021] Although embodiments of the present invention have been shown and described, those of ordinary skill in the art will appreciate that various changes, modifications, substitutions and variations can be made to these embodiments without departing from the principles and spirit of the present invention. The scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. An autonomous learning assistance system based on Chinese education, comprising a recommendation output unit (4), a Chinese character recognition unit (5), and a Chinese character detection unit (6), characterized in that: A main node analysis unit (1), the main node analysis unit (1) obtains all Chinese education video data and corresponding user data in the online course platform, constructs a plurality of main nodes and slave nodes according to the user data and video data, and calculates the number of video views, viewing progress rate, test accuracy rate of the main nodes and the total number of slave nodes, and then calculates to obtain the learning degree value of the main nodes, sets the initial in-group center point, and adds each main node to different groups according to the learning degree value and the number of video views; A user label generation unit (2), after the user label generation unit (2) adds all the main nodes to different groups, calculates a new in-group center point according to the learning degree value and the number of video views of all the main nodes in the group. If all the new in-group center points are the same as the original in-group center points, then determines the number of main nodes in each group, sets an adjustment coefficient and three types of label information, and analyzes the label information of all the main nodes in the corresponding group according to the adjustment coefficient, the learning degree value of each in-group center point, and the number of video views; A slave node analysis unit (3), the slave node analysis unit (3) calculates the total number of viewers of each slave node and the viewing duration, playback times, and test scores of each video in the main nodes, and then calculates the difficulty coefficient of the slave nodes by using the viewing duration, playback times, test scores of each video in the main nodes and the total number of viewers of the slave nodes. After using the difficulty coefficient to count the known node viewing progress value and the difficulty difference vector corresponding to the main nodes, analyzes the viewing progress value and the difficulty difference vector to obtain the attention distribution matrix of the main nodes.

2. The autonomous learning assistance system based on Chinese education according to claim 1, characterized in that: The main node analysis unit (1) includes a data collection module (101), a node construction module (102), a degree calculation module (103), and a grouping generation module (104). The data collection module (101) obtains all Chinese education video data and corresponding user data in the online course platform. The video data includes video number, total video duration, total number of views, and total test score. The user data includes user number, number of video views, viewing duration of different videos, number of replays, and test scores. The node construction module (102) constructs multiple main nodes according to the user data and constructs multiple slave nodes by using the video data , where is the total number of main nodes, is the total number of slave nodes. After counting the number of video views of the th main node , the viewing duration of each video , the test score , and the corresponding total video duration , the total test score , according to in and in calculate the viewing progress rate of , where . By using in and in analyze the test accuracy rate of , where . is the main node number. After the degree calculation module (103) sets three weight coefficients , through the number of video views , the viewing progress rate , the test accuracy rate , and the total number of slave nodes calculate the learning degree value of , where . The grouping generation module (104) repeats the operation until the learning degree value of each main node is determined. Then, it sets three initial within-group center points, calculates the distance value between each main node and different within-group center points according to the learning degree value and the number of video views, and adds each main node to the group where the center point with the smallest distance value is located 3. The self-study assistance system based on Chinese education according to claim 1, characterized in that: The user label generation unit (2) includes a center point update module (201), a node number determination module (202), and a label addition module (203). After the center point update module (201) adds all the master nodes to different groups, it calculates the average value within the current group according to the learning degree values and video viewing quantities of all the master nodes within the group, and determines a new center point within the group by using the average value within each group. If the node number determination module (202) finds that there is a new center point within the group that is inconsistent with the original center point within the group, it calculates the distance value between each master node and the current center point within the group, adds the node to the group where the center point with the smallest distance value is located, and then calculates a new center point within the group according to the master nodes within the group. If all the new center points within the group are consistent with the original center point within the group, it determines the number of master nodes within each group. The label addition module (203) sets an adjustment coefficient and three types of label information, namely positive user label, stable user label, and negative user label. It calculates the actual value corresponding to each group according to the adjustment coefficient, the learning degree value of the center point within each group, and the video viewing quantity, adds the label information of positive users to all the master nodes within the group with the largest actual value, adds the label information of negative users to all the master nodes within the group with the smallest actual value, and adds the label information of stable users to all the master nodes within the remaining groups.

4. The autonomous learning assistance system based on Chinese education according to claim 1, characterized in that: The slave node analysis unit (3) includes a difficulty coefficient calculation module (301), a known matrix determination module (302), a participation vector calculation module (303), and a distribution determination module (304). After the difficulty coefficient calculation module (301) counts the total number of viewers of each slave node and the viewing duration, playback times, and test scores of each video in the master node, it calculates the participation value, completion rate, playback rate, and scoring rate corresponding to each slave node by using the viewing duration, playback times, test scores of each video in the master node and the total number of viewers of the slave node, sets three weight coefficients, and calculates the difficulty coefficient of the slave node according to the weight coefficients and the completion rate, playback rate, and scoring rate corresponding to each slave node. The known matrix determination module (302) constructs a corresponding attribute vector according to the participation value, completion rate, playback rate, scoring rate, and difficulty coefficient of each slave node, combines the attribute vectors of all slave nodes to obtain a global matrix, counts the video numbers with non-zero viewing duration in the master node, determines the corresponding slave nodes according to the video numbers, takes the slave nodes as the known nodes of the current master node, and combines the attribute vectors of the known nodes to obtain the known matrix of the master node. The participation vector calculation module (303) sets a corresponding identity matrix according to the number of parameters in the slave node attribute vector, combines the known matrix of each master node with the identity matrix to obtain the transformation matrix of the master node, extracts the total video duration of the known nodes and the viewing durations of different videos of the master node, determines the viewing progress values of all known nodes corresponding to the master node according to the ratio between the viewing duration and the total duration, and then analyzes the content revenue vector of the master node by using the viewing progress value and the known matrix, and calculates the content revenue vector of the master node and the transformation matrix to determine the attribute participation vector of the master node. The distribution determination module (304) counts the difficulty coefficients of all known nodes, calculates the distance values between the difficulty coefficients of the known nodes and the difficulty coefficients of all slave nodes, constructs a difficulty difference vector according to the multiple distance values of the known nodes, and jointly analyzes all difficulty difference vectors and the viewing progress values of the known nodes to obtain the attention distribution matrix of the current master node.

5. The self-learning assistance system based on Chinese education according to claim 1, wherein: The recommended output unit (4) includes a selection coefficient calculation module (401), a correlation matrix calculation module (402), a recommended coefficient calculation module (403), and an introduction data query module (404). After the selection coefficient calculation module (401) determines the attention distribution matrix, learning degree value, and the total number of slave nodes of the master node, it calculates according to the attention distribution matrix, learning degree value, and the total number of slave nodes to obtain the selection coefficient of the master node. The correlation matrix calculation module (402) analyzes the correlation matrix of the master node through the global matrix corresponding to all slave nodes and the transformation matrix of the master node. The recommended coefficient calculation module (403) statistically analyzes the attribute participation vector, correlation matrix, selection coefficient of each master node, and the global matrix of the slave nodes, and uses the recommended vector analysis algorithm to analyze them to obtain the slave node recommended coefficient corresponding to each master node. After the introduction data query module (404) sets the corresponding recommended number according to the tag information, it screens out multiple slave nodes for each master node according to the recommended coefficient and the recommended number. After determining the number according to the video data corresponding to the slave node, it queries in the database using the video number to obtain the introduction data of the video and transmits it to the user recommendation interface.

6. The self-learning assistance system based on Chinese education according to claim 1, wherein: The Chinese character recognition unit (5) includes a domain setting module (501), a pixel point marking module (502), an actual type determination module (503), and a similarity calculation module (504). After the domain setting module (501) obtains the single Chinese character image data provided by the user, it performs denoising and grayscale operations on the image data to obtain a grayscale image, and sets the size to of the domain box. Arbitrarily select an unmarked pixel point as the center point of the domain box. After statistically calculating the grayscale values and relative positions of all pixel points within the box, calculate the cumulative value of each pixel point within the box according to the grayscale values and relative positions of the pixel points. Use the cumulative values of all pixel points in the current domain box to calculate the domain average value and standard deviation. After the pixel point marking module (502) sets the range value, use the domain average value, standard deviation, and range value to analyze the corresponding domain threshold, and determine the grayscale value of the center point of the domain box. If the grayscale value is greater than the domain threshold, set the grayscale value of the center point to the maximum brightness value; otherwise, set the grayscale value of the center point to zero, and mark the current center point. Repeat the operation until all pixel points are marked, so as to obtain the processed image data. The actual type determination module (503) obtains all the processed image data, transmits it to the CRNN model, ResNet model, and Transformer model for analysis. After obtaining three prediction results, take the Chinese character category corresponding to the maximum probability value in the prediction results as the actual type of the image data. The similarity calculation module (504) queries in the database according to the actual type to obtain the handwritten Chinese character template corresponding to the Chinese character type. After extracting the attribute representations of the Chinese characters in the image data and the attribute representations of the three Chinese character templates, calculate the similarities between the attribute representations of the Chinese characters in the image data and the attribute representations of the three Chinese character templates respectively. Set the adjustment coefficient, and use the similarity analysis algorithm to combine the adjustment coefficient with the three similarities to obtain the actual similarity between the Chinese characters in the current image data and the template.

7. The autonomous learning assistance system based on Chinese education according to claim 1, characterized in that: The Chinese character detection unit (6) includes a gradient magnitude determination module (601) and a mean value calculation module (602). The gradient magnitude determination module (601) performs a convolution operation on the image data using a Gaussian filter, determines the gradient values of the image data in different directions, calculates the gradient direction and gradient magnitude of each pixel point according to the gradient values, determines the adjacent pixel points of each pixel point according to the gradient direction, and judges the magnitude of the gradient magnitude between the current pixel point and the adjacent pixel points. If the gradient magnitude of the current pixel point is greater than the gradient magnitude of the adjacent pixel points, the gradient magnitude of the current pixel point is retained; otherwise, the gradient magnitude is set to zero. The mean value calculation module (602) statistically analyzes the gradient magnitudes of all pixel points in the image data, calculates the average value and standard deviation corresponding to the image using the gradient magnitudes, accumulates the average value and the standard deviation to obtain a first threshold, and takes the difference between the average value and the standard deviation as a second threshold. If the gradient magnitude of a pixel point is greater than the first threshold, the current pixel point is determined to be an edge point; if the gradient magnitude is less than or equal to the first threshold and greater than the second threshold, the current pixel point is determined to be a candidate point; if the gradient magnitude is less than or equal to the second threshold, the current pixel point is determined not to be an edge point.

8. The autonomous learning assistance system based on Chinese education according to claim 7, characterized in that: The Chinese character detection unit (6) further includes a candidate point analysis module (603), a dimension setting module (604), and a determination result generation module (605). After setting a window by the candidate point analysis module (603), taking the candidate point as the center point of the window, counting the number of edge points within the window. If the number is zero, it is determined that the current candidate point is not an edge point; otherwise, it is determined that the current candidate point is an edge point. Connect all the edge points to the nearest edge point to determine the contour features of the current image data. After determining the length, width, and area of the Chinese character according to the contour features in the image data by the dimension setting module (604), calculate the actual dimension of the Chinese character in the image data using the adjustment coefficient and the length, width, and area of the Chinese characters in the three templates. The determination result generation module (605) counts the actual similarity and dimension of the Chinese characters in the image. If both the actual similarity and dimension of the Chinese characters in the image are greater than the threshold, it is prompted that the current Chinese character is qualified; otherwise, it is prompted that the current Chinese character is unqualified.

Images