A smart education system

Abstract

An embodiment of the present invention provides a smart education system, belonging to the technical field of smart education. The system includes: an anomaly analysis module analyzes the first behavior data to obtain a first abnormal behavior and the abnormal position of the first abnormal behavior; a data acquisition module obtains adjacent behavior data from the first behavior data according to the abnormal position; a behavior prediction module reconstructs the behavior based on the adjacent behavior data to obtain a target predicted behavior; an anomaly screening module screens out a second abnormal behavior according to the target predicted behavior and the first abnormal behavior; an anomaly classification module classifies the second abnormal behavior to obtain an anomaly type; an anomaly determination module determines a target abnormal behavior according to the anomaly type, and obtains an associated teaching text of the target abnormal behavior from the initial teaching text; a preference recognition module analyzes the second behavior data to obtain a target behavior preference; a data adjustment module adjusts the associated teaching text according to the target abnormal behavior and the target behavior preference to obtain a target teaching text.

Description

Technical Field

[0001] The present invention relates to the field of intelligent education technology, and in particular, to an intelligent education system. Background Art

[0002] In the process of education, teachers usually prepare detailed teaching plans before teaching to guide and standardize teaching activities. The formulation of teaching plans is an important part of teaching preparation, which helps to ensure the clarity of teaching objectives, the systematicness of content, and the rationality of teaching methods. However, the existing teaching plan design may be too rigid and difficult to adjust teaching strategies in a timely manner according to the performance and feedback of students, thus making it difficult to effectively respond to various changes in the classroom. That is to say, many teaching plans do not fully consider the individual differences of students, such as learning ability, interests and hobbies, cognitive level, and cultural background. This one-size-fits-all teaching design is difficult to meet the needs of different students, resulting in some students may not be able to keep up with the teaching progress or lose their interest in learning, thereby further reducing the teaching quality of teachers. Summary of the Invention

[0003] The main purpose of the embodiments of the present invention is to provide an intelligent education system, aiming to solve the problem in the related technology that the teaching plans of teachers are difficult to meet the needs of different students, resulting in some students being unable to keep up with the teaching progress and thus reducing the teaching quality of teachers.

[0004] In a first aspect, the embodiments of the present invention provide an intelligent education system, including:

[0005] A data acquisition module, configured to obtain first behavior data corresponding to a target student under an initial teaching text and obtain second behavior data corresponding to the target student watching a teaching video;

[0006] An anomaly analysis module, configured to perform anomaly behavior analysis on the first behavior data to obtain a first anomaly behavior corresponding to the target student and a first anomaly position corresponding to the first anomaly behavior;

[0007] A data obtaining module, configured to obtain adjacent behavior data corresponding to the target student from the first behavior data according to the first anomaly position;

[0008] A behavior prediction module, configured to perform behavior reconstruction according to the adjacent behavior data to obtain a target prediction behavior corresponding to the target student under the first anomaly behavior;

[0009] An anomaly screening module, configured to screen out a second anomaly behavior corresponding to the target student according to the target prediction behavior and the first anomaly behavior;

[0010] An anomaly classification module, configured to perform anomaly classification according to the second anomaly behavior to obtain an anomaly type corresponding to the second anomaly behavior;

[0011] Anomaly determination module, configured to determine a target abnormal behavior corresponding to the target student according to the anomaly type, and obtain associated teaching text corresponding to the target abnormal behavior from the initial teaching text;

[0012] Preference recognition module, configured to perform behavior analysis on the second behavior data to obtain a target behavior preference corresponding to the target student;

[0013] Data adjustment module, configured to adjust the associated teaching text according to the target abnormal behavior and the target behavior preference to obtain target teaching text corresponding to the target student.

[0014] An embodiment of the present invention provides an intelligent education system, which includes: a data acquisition module, configured to obtain first behavior data corresponding to a target student under an initial teaching text and obtain second behavior data corresponding to the target student's viewing of a teaching video. By collecting the behavior data of the target student when viewing the initial teaching text and the teaching video, the learning habits and interaction patterns of the student can be comprehensively understood, thereby providing support for subsequent provision of personalized teaching texts. An anomaly analysis module, configured to perform anomaly behavior analysis on the first behavior data to obtain a first anomaly behavior corresponding to the target student and a first anomaly position corresponding to the first anomaly behavior. By analyzing the first anomaly behavior in the first behavior data, possible learning obstacles or comprehension difficulties of the target student can be identified, providing support for subsequent adjustment of the teaching text. A data acquisition module, configured to obtain adjacent behavior data corresponding to the target student from the first behavior data according to the first anomaly position; a behavior prediction module, configured to perform behavior reconstruction based on the adjacent behavior data to obtain a target prediction behavior corresponding to the target student under the first anomaly behavior; an anomaly screening module, configured to screen out a second anomaly behavior corresponding to the target student according to the target prediction behavior and the first anomaly behavior; an anomaly classification module, configured to perform anomaly classification on the second anomaly behavior to obtain an anomaly type corresponding to the second anomaly behavior; an anomaly determination module, configured to determine a target anomaly behavior corresponding to the target student according to the anomaly type and obtain an associated teaching text corresponding to the target anomaly behavior from the initial teaching text; a preference recognition module, configured to perform behavior analysis on the second behavior data to obtain a target behavior preference corresponding to the target student, thereby timely discovering the learning method preference of the target student and providing a direction for subsequent adjustment of the teaching text; a data adjustment module, configured to adjust the associated teaching text according to the target anomaly behavior and the target behavior preference to obtain a target teaching text corresponding to the target student. By accurately identifying the target anomaly behavior and the target behavior preference, the pertinence of the teaching text can be significantly improved, thereby effectively supporting the subsequent learning of the target student and helping them master knowledge more efficiently. This can not only improve the teaching quality of teachers, but also solve the problem in related technologies that the teaching plans of teachers are difficult to meet the needs of different students, resulting in some students being unable to keep up with the teaching progress and thus reducing the teaching quality of teachers. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the accompanying drawings required for the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings in the following description are some embodiments of the present application. For those of ordinary skill in the art, other accompanying drawings can be obtained based on these drawings without creative efforts.

[0016] Figure 1 It is a schematic diagram of the module structure of an intelligent education system provided by an embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0017] The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are some, but not all, embodiments of the present invention. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0018] The flowcharts shown in the accompanying drawings are only illustrative examples, and do not necessarily include all contents and operations / steps, nor do they necessarily need to be executed in the described order. For example, some operations / steps can also be decomposed, combined, or partially merged, so the actual execution order may be changed according to the actual situation.

[0019] It should be understood that the terms used in the specification of the present invention are only for the purpose of describing specific embodiments and are not intended to limit the present invention. As used in the specification of the present invention and the appended claims, unless the context clearly indicates otherwise, the singular forms "a", "an", and "the" are intended to include the plural forms.

[0020] The embodiments of the present invention provide an intelligent education system. Among them, the intelligent education system can be applied to a terminal device, and the terminal device can be an electronic device such as a tablet computer, a notebook computer, a desktop computer, a personal digital assistant, and a wearable device. The terminal device can be a server or a server cluster.

[0021] The following will describe in detail some embodiments of the present invention with reference to the accompanying drawings. Without conflict, the following embodiments and the features in the embodiments can be combined with each other.

[0022] Please refer to Figure 1 , Figure 1Schematic diagram of the module structure of an intelligent education system provided by an embodiment of the present invention. An intelligent education system 200 provided by an embodiment of the present application, the intelligent education system 200 includes a data acquisition module 201, an anomaly analysis module 202, a data obtaining module 203, a behavior prediction module 204, an anomaly screening module 205, an anomaly classification module 206, an anomaly determination module 207, a preference recognition module 208, and a data adjustment module 209. Among them, the data acquisition module 201 is used to obtain the first behavior data corresponding to the target student under the initial teaching text and the second behavior data corresponding to the target student watching the teaching video; the anomaly analysis module 202 is used to perform anomaly behavior analysis on the first behavior data to obtain the first anomaly behavior corresponding to the target student and the first anomaly position corresponding to the first anomaly behavior; the data obtaining module 203 is used to obtain the adjacent behavior data corresponding to the target student from the first behavior data according to the first anomaly position; the behavior prediction module 204 is used to perform behavior reconstruction according to the adjacent behavior data to obtain the target prediction behavior corresponding to the target student under the first anomaly behavior; the anomaly screening module 205 is used to screen out the second anomaly behavior corresponding to the target student according to the target prediction behavior and the first anomaly behavior; the anomaly classification module 206 is used to perform anomaly classification according to the second anomaly behavior to obtain the anomaly type corresponding to the second anomaly behavior; the anomaly determination module 207 is used to determine the target anomaly behavior corresponding to the target student according to the anomaly type, and obtain the associated teaching text corresponding to the target anomaly behavior from the initial teaching text; the preference recognition module 208 is used to perform behavior analysis on the second behavior data to obtain the target behavior preference corresponding to the target student; the data adjustment module 209 is used to adjust the associated teaching text according to the target anomaly behavior and the target behavior preference to obtain the target teaching text corresponding to the target student.

[0023] Exemplarily, the data acquisition module 201 is used to obtain the first behavior data corresponding to the target student under the initial teaching text and the second behavior data corresponding to the target student watching the teaching video. Among them, the initial teaching text can be a teaching plan prepared by the teacher according to the textbook content, or a lecture summary prepared by the teacher according to the knowledge points to be explained.

[0024] For example, the first behavioral data of the target student is tracked through a video sensor when following the teacher to learn the initial teaching text. The first behavioral data is the video data corresponding to the target student listening to the teacher's explanation of the initial teaching text. In addition, the functions of the video platform can be used to track the second behavioral data of the target student watching other teaching videos, such as viewing type, viewing duration, number of replays, pause points, viewing speed, etc. In addition, throughout the process, it is important to ensure the privacy and security of the data, comply with relevant data protection regulations, and obtain the consent of the target student to collect and use their first and second behavioral data.

[0025] Exemplarily, the anomaly analysis module 202 determines normal behavioral postures based on historical experience, and then uses statistical methods or machine learning algorithms to identify behaviors in the first behavioral data that are significantly different from the normal behavioral posture benchmark. These first abnormal behaviors may include lying on the table, lowering the head for a long time, etc. Then, the corresponding frames of the first abnormal behavior in the first behavioral data are obtained, and the time information corresponding to the corresponding frames is obtained. Then, the time information is determined as the first abnormal position corresponding to the first abnormal behavior.

[0026] Exemplarily, the data acquisition module 203 intercepts the normal behavioral data corresponding to a preset range before the first abnormal position from the first behavioral data according to the first abnormal position, and determines it as the adjacent behavioral data corresponding to the target student under the first abnormal behavior.

[0027] Exemplarily, the behavior prediction module 204 selects a suitable behavior prediction model. Common models include decision trees, random forests, support vector machines, neural networks, etc. Then, the selected model is trained using the preprocessed historical adjacent behavioral data. In this process, the model will learn the patterns and relationships in the data, so as to be able to predict the target behavior based on the input adjacent behavioral data, and thus obtain the target prediction behavior corresponding to the first abnormal position of the target student under the first abnormal behavior.

[0028] Exemplarily, the anomaly screening module 205 calculates the similarity between the target prediction behavior and the first abnormal behavior to obtain the target similarity value. Then, when the target similarity value is less than the preset value, the first abnormal behavior corresponding to the target similarity value is determined as the second abnormal behavior corresponding to the target student; when the target similarity value is greater than or equal to the preset value, it means that the first abnormal behavior corresponding to the target similarity value is a misjudgment, and the first abnormal behavior can be directly deleted.

[0029] Exemplarily, the anomaly classification module 206 uses machine learning algorithms such as K-nearest neighbor, decision tree, support vector machine, etc. to train the model with historical abnormal data and the known abnormal types corresponding to the historical abnormal data, so as to obtain the target classification model. Then, the second abnormal behavior is classified according to the target classification model, and the abnormal type corresponding to the second abnormal behavior is obtained.

[0030] Exemplarily, the anomaly determination module 207 summarizes the anomaly types corresponding to all target students to obtain the target abnormal behavior corresponding to the anomaly type with a proportion exceeding the preset value among the anomaly types corresponding to all target students. Then, based on the target abnormal behavior, the second abnormal position corresponding to this target type is obtained from the first abnormal positions, and based on the second abnormal position, the content currently explained by the teacher is obtained. Then, the associated teaching text corresponding to the target abnormal behavior is obtained from the initial teaching text.

[0031] Exemplarily, the preference recognition module 208 defines behavior preference indicators according to the second behavior data. The behavior preference indicators are used to help quantify and evaluate behavior preferences, such as: participation, viewing duration, viewing type, etc. Then, the average value, median, standard deviation, etc. of each behavior indicator are calculated to understand the data distribution, and thus the changing trend of the second behavior data over time is analyzed to discover behavior patterns. Then, the correlation between different behavior indicators is calculated to identify the key factors affecting behavior preferences. Thus, the second behavior data of the students is clustered to discover different behavior patterns and preference types. Then, according to the analysis results, the target behavior preferences of the target students are identified.

[0032] For example, if the target student likes to learn the content related to geometry in mathematics, it can be found that the target behavior preference of the target student may be the content related to space and imagination.

[0033] Exemplarily, the data adjustment module 209 is used to adjust the associated teaching text according to the target abnormal behavior and the target behavior preference to obtain the target teaching text corresponding to the target student.

[0034] For example, if the target abnormal behavior is frequent distraction in class and the target behavior preference is liking interaction and participation, specific adjustment strategies are determined according to the target abnormal behavior and the target behavior preference. These strategies may include: increasing interactivity: increasing classroom interaction and discussion sessions; introducing interestingness: adding interesting cases, stories or practical activities; integrating students' interests: introducing relevant content or activities according to students' interests. For example, adding group discussion and classroom interaction sessions to the teaching text, and adding science-related practical activities such as science experiments and science competitions to the teaching text.

[0035] Exemplarily, according to the determined adjustment strategies, the associated teaching text is specifically adjusted to obtain the target teaching text. For example, changing the originally boring associated teaching text to be explained through experiments and actual operations to obtain the target teaching text, and thus obtaining the target teaching text that better meets the needs and preferences of the students. This process helps to improve the learning interest and participation of the target students, improve abnormal behaviors, and enhance learning effects.

[0036] Exemplarily, when the target teaching text is obtained, and there are students with the same situation as the target student, knowledge can be imparted according to the target teaching text, thereby improving the teaching quality of teachers and the learning effect of students in learning knowledge.

[0037] In some embodiments, the anomaly analysis module includes: a target recognition module for performing target recognition on the first behavior data to obtain target pose information corresponding to the target student in each frame; a pose clustering module for performing pose clustering on the target pose information to obtain a target clustering result corresponding to the target student; a key pose determination module for determining key pose information corresponding to the first behavior data according to the target clustering result; an anomaly scoring module for performing anomaly behavior scoring according to the key pose information to obtain a target anomaly score corresponding to the key pose information; an anomaly behavior determination module for determining the first anomaly behavior corresponding to the target student and the first anomaly position corresponding to the first anomaly behavior according to the target anomaly score.

[0038] Exemplarily, the anomaly analysis module includes a target recognition module, a pose clustering module, a key pose determination module, an anomaly scoring module, and an anomaly behavior determination module.

[0039] Exemplarily, the target recognition module performs target recognition on the first behavior data to obtain target pose information corresponding to the target student in each frame. For example, the first behavior data includes a video or an image sequence, and each frame of the image includes an image of the target student. Then, a target detection algorithm (such as YOLO, SSD, etc.) is used to detect the target student in each frame of the image, and a pose estimation algorithm (such as OpenPose, HRNet, etc.) is used to extract the joint position information of the target student, generating the target pose information for each frame.

[0040] Exemplarily, the pose clustering module performs pose clustering on the target pose information to obtain a target clustering result corresponding to the target student. For example, features are extracted from the target pose information of each frame, such as joint positions, angles, distances, etc. Then, a clustering algorithm (such as K-means, DBSCAN, etc.) is used to cluster the extracted features, grouping similar poses into one category, thereby generating the target clustering result of the target student, and each clustering represents a type of pose.

[0041] Exemplarily, the key pose determination module determines the key pose information corresponding to the first behavior data according to the target clustering result, and then analyzes the target clustering result to identify the clustering clusters with higher occurrence frequencies or significant features, thereby selecting the clustering with a higher occurrence frequency or significant features as the key pose information.

[0042] Exemplarily, the abnormal score module scores abnormal behaviors based on the key pose information to obtain the target abnormal score corresponding to the key pose information. For example, the abnormal behavior is defined as an overly long pose duration or an abnormal joint angle, and then a scoring criterion is formulated. For example, 1 point is given if the joint angle exceeds a certain threshold, and 2 points are given if the pose duration exceeds a certain threshold. Thus, according to the scoring criterion, the key pose information is scored to generate an abnormal score.

[0043] Exemplarily, the abnormal behavior determination module determines the first abnormal behavior corresponding to the target student and the first abnormal position corresponding to the first abnormal behavior according to the target abnormal score. For example, a threshold for the abnormal behavior is set, and a pose exceeding this threshold is considered an abnormal behavior. Thus, the target abnormal score is compared with the threshold to determine which poses are the first abnormal behaviors. Furthermore, the corresponding frames of the first abnormal behavior in the first behavior data are obtained, and then the time information corresponding to the corresponding frames is obtained. Furthermore, this time information is determined as the first abnormal position corresponding to the first abnormal behavior.

[0044] In some embodiments, the pose clustering module includes: a joint recognition module, configured to recognize the target joints in the target pose information by using the joint recognition layer of the feature recognition model to obtain the joint positions corresponding to the target joints of the target student in the target pose information; an information calculation module, configured to calculate the relative position information corresponding to the target joints in the target pose information according to the joint positions by using the information calculation layer of the feature recognition model; an emotion recognition module, configured to determine the target emotion information corresponding to the target student according to the relative position information and the corresponding facial information in the target pose information by using the emotion recognition layer of the feature recognition model; and an information clustering module, configured to perform pose clustering according to the relative position information and the target emotion information by using the information clustering layer of the feature recognition model to obtain the target clustering result corresponding to the target student.

[0045] Exemplarily, the pose clustering module includes a joint recognition module, an information calculation module, an emotion recognition module, and an information clustering module.

[0046] Exemplarily, the joint recognition module uses the joint recognition layer of the feature recognition model to recognize the target joints in the target pose information, and obtains the joint positions corresponding to the target joints of the target student in the target pose information. For example, the joint recognition layer is OpenPose, and thus the key joints of the target such as the shoulders, elbows, knees, etc. can be recognized. Furthermore, the position information of the target joints is extracted from the output of the joint recognition layer to generate the coordinate data of the target joints in the target pose information, that is, the joint positions corresponding to the target joints are obtained.

[0047] Exemplarily, the information calculation module uses the information calculation layer of the feature recognition model to calculate the relative position information corresponding to the target joint in the target pose information based on the joint positions. For example, it calculates the relative positions between the target joints according to the joint position information. For instance, it calculates the position of the elbow joint relative to the shoulder joint, the position of the knee joint relative to the hip joint, etc., so as to obtain the relative position information corresponding to the target joint.

[0048] Exemplarily, the emotion recognition module uses the emotion recognition layer of the feature recognition model to determine the target emotion information corresponding to the target student based on the relative position information and the corresponding facial information in the target pose information. For example, it extracts the facial information of the target student from the target pose information, such as facial expressions, eye states, etc., and then uses the emotion recognition layer (such as an emotion recognition model based on deep learning) to analyze the facial information and the relative position information to identify the target emotion information of the target student, such as happy, sad, angry, etc.

[0049] Exemplarily, the information clustering module uses the information clustering layer of the feature recognition model to perform pose clustering based on the relative position information and the target emotion information, and obtains the target clustering result corresponding to the target student. It combines the relative position information and the target emotion information into a feature vector as the input of the clustering, and then uses a clustering algorithm (such as K-means, DBSCAN, etc.) to cluster the feature vector, classifying similar information into one category, and then generating the target clustering result of the target student. Each clustering cluster represents a pose type and its corresponding emotion state.

[0050] In some embodiments, the information clustering module includes: a center determination module, configured to use the information clustering layer to determine an initial clustering center from the target pose information based on the relative position information and the target emotion information, and obtain the initial position information and the initial emotion information corresponding to the initial clustering center; a distance calculation module, configured to use the information clustering layer to calculate the distance information between the target pose information and the initial clustering center based on the relative position information, the target emotion information, the initial position information, and the initial emotion information; an initial clustering module, configured to cluster the target pose information according to the distance information to obtain an initial clustering result; a fluctuation calculation module, configured to calculate the data distribution fluctuation of each first cluster in the initial clustering result to obtain the target fluctuation value corresponding to the first cluster; a clustering adjustment module, configured to adjust the initial clustering result according to the target fluctuation value to obtain the target clustering result corresponding to the target student; wherein, the distance information is obtained according to the following formula:

[0051] ;

[0052] Wherein, represents the distance information between the i-th target pose information and the j-th initial clustering center; represents weight information; n represents the quantity corresponding to the relative position information; represents the weight information corresponding to the k-th relative position information; represents the k-th relative position information of the i-th target pose information; represents the k-th initial position information of the j-th initial clustering center; represents the target emotion information corresponding to the i-th target pose information; represents the initial emotion information corresponding to the j-th initial clustering center.

[0053] Exemplarily, the information clustering module includes a center determination module, a distance calculation module, an initial clustering module, a fluctuation calculation module, and a clustering adjustment module.

[0054] Exemplarily, the center determination module extracts relative position information and target emotion information from the target pose information using the information clustering layer as input features for clustering. Using an initial center selection method (such as random selection or heuristic-based selection) of a clustering algorithm (such as the K-means algorithm), several initial clustering centers are selected from the target pose information, and then the relative position information and target emotion information of each initial clustering center are extracted to generate initial position information and initial emotion information.

[0055] Exemplarily, the distance calculation module calculates the distance information between each sample in the target pose information and the initial clustering center according to the following formula:

[0056] ;

[0057] where represents the distance information between the i-th target pose information and the j-th initial clustering center; represents weight information; n represents the quantity corresponding to the relative position information; represents the weight information corresponding to the k-th relative position information; represents the k-th relative position information of the i-th target pose information; represents the k-th initial position information of the j-th initial clustering center; represents the target emotion information corresponding to the i-th target pose information; represents the initial emotion information corresponding to the j-th initial clustering center.

[0058] Exemplarily, the formula allows relative position information and target emotion information to be incorporated into the distance calculation simultaneously. By fusing features from multiple dimensions (such as the relative position of the body and the emotional state), the similarity between the target pose information and the initial clustering center can be measured more comprehensively. This fusion of multi-dimensional features can improve the accuracy and robustness of the clustering results because features in a single dimension (such as only considering body position) may not fully reflect the complex relationship between pose and emotion. The formula can comprehensively consider the feature differences in multiple dimensions. Traditional distance metric methods (such as Euclidean distance) may only focus on position information and ignore emotion information or other important features. Through the weighted summation method in the formula, the distance information can more accurately reflect the difference between the target pose information and the clustering center.

[0059] Exemplarily, based on the distance information, the initial clustering module assigns each sample of target pose information to the cluster where the nearest initial clustering center is located, and then sorts the assigned samples by cluster to generate the initial clustering result.

[0060] Exemplarily, the fluctuation calculation module calculates the data distribution fluctuation for each first cluster in the initial clustering result to obtain the target fluctuation value corresponding to the first cluster. For example, the target fluctuation value is used to measure the data distribution within each cluster. The fluctuation value can be based on statistics such as variance and standard deviation of the samples within the cluster, and then the fluctuation calculation is performed on the samples within each first cluster to generate the target fluctuation value corresponding to each cluster.

[0061] Exemplarily, the clustering adjustment module determines whether the distribution of the cluster is reasonable according to the target fluctuation value. If the target fluctuation value of a certain cluster is too large, it indicates that the data distribution within the cluster is uneven and may need to be adjusted. For a cluster with an overly large target fluctuation value, splitting operations (dividing the cluster into multiple sub-clusters) or merging operations (merging the cluster with other clusters) can be taken to optimize the clustering result, and then after adjustment, the target clustering result corresponding to the target student is generated.

[0062] Exemplarily, the implementation of the information clustering module includes center determination, distance calculation, initial clustering, fluctuation calculation, and clustering adjustment. This process helps to cluster the target pose information and adjust the clustering result according to the data distribution fluctuation, and finally obtain the target clustering result corresponding to the target student. This process can effectively identify the common pose types of the target student and their corresponding emotional states, provide targeted intervention suggestions for teachers, and improve the classroom performance and learning effect of students.

[0063] In some embodiments, the behavior prediction module includes: a reconstructed behavior layer of the behavior prediction model that uses an autoencoder to combine the neighboring behavior data for behavior reconstruction to obtain a first predicted behavior; a predicted behavior layer of the behavior prediction model that uses an encoder and a decoder to combine the neighboring behavior data for behavior prediction to obtain a second predicted behavior; and a behavior fusion layer of the behavior prediction model that performs behavior fusion based on the first predicted behavior and the second predicted behavior to obtain the target predicted behavior corresponding to the target student under the first abnormal behavior.

[0064] Exemplarily, the reconstructed behavior layer uses an autoencoder to combine the neighboring behavior data for behavior reconstruction to obtain a first predicted behavior. For example, the neighboring behavior data of the target student is collected, which includes but is not limited to posture information, action sequences, etc. Then, an autoencoder model is constructed. The autoencoder usually consists of an encoder and a decoder. The encoder compresses the input data into a low-dimensional representation, and the decoder reconstructs the low-dimensional representation into the original data. Then, the neighboring behavior data is preprocessed to make it suitable for the input format of the autoencoder, such as normalization, padding, etc. Thus, the autoencoder is used to reconstruct the neighboring behavior data, that is, the neighboring behavior data is compressed into a low-dimensional representation by the encoder, and then the low-dimensional representation is reconstructed into the first predicted behavior by the decoder.

[0065] Exemplarily, the predicted behavior layer of the behavior prediction model uses an encoder and a decoder to combine the neighboring behavior data for behavior prediction. An encoder and a decoder model are constructed. The encoder compresses the input data into a low-dimensional representation, and the decoder converts the low-dimensional representation into a future predicted behavior. The neighboring behavior data is preprocessed to make it suitable for the input format of the encoder and the decoder. The encoder is used to encode the neighboring behavior data to generate a low-dimensional representation, and then the decoder is used to convert the low-dimensional representation into a future second predicted behavior.

[0066] Exemplarily, the behavior fusion layer of the behavior prediction model performs behavior fusion based on the first predicted behavior and the second predicted behavior to obtain the target predicted behavior corresponding to the target student under the first abnormal behavior. For example, if weighted averaging is used, the weights of the first predicted behavior and the second predicted behavior need to be determined. Then, the first predicted behavior and the second predicted behavior are preprocessed to ensure that they have the same dimension and format. Then, according to the fusion rule, the first predicted behavior and the second predicted behavior are fused to generate the target predicted behavior corresponding to the target student under the first abnormal behavior.

[0067] For example, weighted average is selected as the behavior fusion method. The weights of the first predicted behavior and the second predicted behavior are determined. For example, the weight of the first predicted behavior is 0.6 and the weight of the second predicted behavior is 0.4. Then, the first predicted behavior and the second predicted behavior are normalized to ensure that they have the same dimension and format. Thus, the weighted average method is used to fuse the first predicted behavior and the second predicted behavior to generate the target predicted behavior corresponding to the first abnormal behavior of the target student.

[0068] In some embodiments, the abnormal screening module includes: a similarity calculation module for calculating the similarity between the target predicted behavior and the first abnormal behavior; and a screening processing module for screening out the second abnormal behavior corresponding to the target student from the first abnormal behavior according to the similarity.

[0069] Exemplarily, the similarity calculation module represents the target predicted behavior and the first abnormal behavior as comparable feature vectors. These features may include the key point positions of the posture, the sequence of actions, the emotional state, etc. Then, a similarity measurement method, such as cosine similarity, Euclidean distance, correlation coefficient, etc., is used to calculate the similarity between the target predicted behavior and each first abnormal behavior.

[0070] Exemplarily, the screening processing module sets a similarity threshold for determining which first abnormal behaviors are similar enough to the target predicted behavior. Then, according to the set threshold, the first abnormal behaviors with similarity lower than the threshold are screened out, and the screened behaviors are used as the second abnormal behaviors of the target student.

[0071] In some embodiments, the abnormal classification module includes: a first feature extraction module for extracting deflection features from the second abnormal behavior to obtain the first behavior feature corresponding to the second abnormal behavior; a second feature extraction module for extracting gradient features from the second behavior feature by using adjacent information to obtain the second behavior feature corresponding to the second abnormal behavior under instantaneous behavior; a third feature extraction module for performing emotion recognition on the second abnormal behavior to obtain the third behavior feature corresponding to the second abnormal behavior; a feature fusion module for fusing the first behavior feature, the second behavior feature, and the third behavior feature to obtain the target behavior feature corresponding to the second abnormal behavior; a correlation calculation module for determining a preset type and a preset feature variable corresponding to the preset type, and determining a correlation value between the second abnormal behavior and the preset type according to the target behavior feature and the preset feature variable; and a type determination module for determining the abnormal type corresponding to the second abnormal behavior from the preset types according to the correlation value. Wherein, the correlation value is obtained according to the following formula:

[0072]

[0073] Among them, represents the correlation value between the i-th second abnormal behavior and the j-th preset type; represents the preset feature variable corresponding to the j-th preset type, represents the target behavior feature corresponding to the i-th second abnormal behavior; represents a balance parameter; r represents the number of corresponding features in the target behavior feature; err represents an error parameter; represents the similarity between the preset feature variable corresponding to the j-th preset type and the target behavior feature corresponding to the i-th second abnormal behavior; represents an adjustment parameter; represents the preset feature vector corresponding to the w-th adjacent preset type corresponding to the j-th preset type.

[0074] Exemplarily, the first feature extraction module performs deflection feature extraction on the second abnormal behavior to obtain the first behavior feature corresponding to the second abnormal behavior, and analyzes the deflection in the second abnormal behavior (such as sudden changes in body posture, abnormal offsets of actions, etc.). The deflection feature can be captured by the relative change of pose data. The extracted deflection feature is represented as a feature vector, such as the deflection angle, deflection amplitude, etc. For example, the deflection angle of the body posture in the second abnormal behavior (such as sudden elevation of the arm or abnormal bending of the leg) is extracted, and then the deflection angle and amplitude are recorded to form the first behavior feature.

[0075] Exemplarily, the second feature extraction module performs gradient feature extraction on the second behavior feature using adjacent information to obtain the second behavior feature corresponding to the second abnormal behavior under the instantaneous behavior. The front and back frame information of the second abnormal behavior is analyzed to capture the change trend and instantaneous gradient feature of its behavior, and then the gradient feature of the second abnormal behavior is calculated, such as the rate and direction of behavior change. For example, the front and back frames of the second abnormal behavior (such as frames with a time interval of 0.5 seconds) are analyzed, and then the gradient feature of behavior change, such as the speed and direction of action change, is calculated to form the second behavior feature.

[0076] Exemplarily, the third feature extraction module performs emotion recognition on the second abnormal behavior to obtain the third behavior feature corresponding to the second abnormal behavior. For example, an emotion recognition model or rule (such as facial expression, speech intonation, behavior pattern, etc.) is used to recognize the emotion state in the second abnormal behavior. The recognized emotion state is represented as a feature vector, such as emotion category (anxiety, anger, etc.) and emotion intensity, so as to recognize the emotion state (such as anxiety) in the second abnormal behavior through facial expression and behavior pattern, and record the emotion category and intensity to form the third behavior feature.

[0077] Exemplarily, the feature fusion module obtains the target behavior feature corresponding to the second abnormal behavior by a feature fusion method, such as weighted average, splicing, neural network fusion, etc., for the first row behavior feature, the second row behavior feature, and the third row behavior feature.

[0078] Exemplarily, the correlation calculation module determines a preset type and a preset feature variable corresponding to the preset type, and determines a correlation value between the second abnormal behavior and the preset type according to the target behavior feature and the preset feature variable. Define corresponding feature variables for each preset abnormal type (such as anxiety, happiness, etc.), and then calculate the correlation value between the target behavior feature and the feature variable of each preset type according to the following formula:

[0079]

[0080] Wherein, represents the correlation value between the i-th second abnormal behavior and the j-th preset type; represents the preset feature variable corresponding to the j-th preset type, represents the target behavior feature corresponding to the i-th second abnormal behavior; represents a balance parameter; r represents the number of corresponding features in the target behavior feature; err represents an error parameter; represents the similarity between the preset feature variable corresponding to the j-th preset type and the target behavior feature corresponding to the i-th second abnormal behavior; represents a regulation parameter; represents the preset feature vector corresponding to the w-th adjacent preset type corresponding to the j-th preset type.

[0081] Exemplarily, the balance parameter helps to balance the importance of different features, ensuring that no single feature overly dominates the calculation of the correlation value. The error parameter allows the model to tolerate imperfect feature matching to a certain extent, making the correlation calculation more robust. Furthermore, by quantifying the similarity between the preset feature variable and the target behavior feature, the matching degree between the second abnormal behavior and the preset type can be objectively measured. Considering adjacent preset types takes into account the influence of adjacent preset types, which helps to more accurately determine the type of abnormal behavior when there are fuzzy boundaries between types. Thus, through the above comprehensive calculation, the second abnormal behavior can be more precisely classified into the most appropriate preset type, thereby improving the accuracy of abnormal behavior recognition, and further enabling a comprehensive and accurate assessment of the relationship between the second abnormal behavior and the preset type, thus improving the overall performance and reliability of abnormal behavior classification.

[0082] Exemplarily, the type determination module determines the exception type corresponding to the second abnormal behavior from the preset types according to the correlation value. A correlation threshold is set for each preset type, and the type whose correlation value exceeds the threshold can be considered as a candidate type, and then the preset type with the highest correlation value is selected as the exception type of the second abnormal behavior.

[0083] In some embodiments, the preference recognition module includes: a first analysis module configured to perform viewing behavior analysis on the second behavior data to obtain a first viewing behavior sequence and a second viewing behavior sequence corresponding to the target student, wherein the first viewing behavior sequence is used to represent the sequence corresponding to the target student completely viewing the video, and the second viewing behavior sequence is used to represent the sequence corresponding to the target student not completely viewing the video; a second analysis module configured to perform viewing result analysis on the second behavior data to obtain a first viewing score sequence and a second viewing score sequence corresponding to the target student, wherein the first viewing score sequence is used to represent the sequence corresponding to the target student liking to view the video, and the second viewing score sequence is used to represent the sequence corresponding to the target student not liking to view the video; a first embedding processing module configured to perform feature representation on the first viewing behavior sequence based on the first embedding layer of the interest recognition model to obtain a first feature vector and perform feature representation on the first viewing score sequence to obtain a second feature vector, where the interest recognition model is trained by using the first historical behavior sequence, the second historical behavior sequence, the first historical score sequence, and the second historical score sequence corresponding to the historical student as model inputs, and using the historical behavior preference corresponding to the historical student under the first historical behavior sequence, the second historical behavior sequence, the first historical score sequence, and the second historical score sequence as the model output; wherein the first historical behavior sequence is used to represent the sequence corresponding to the historical student completely viewing the video, the second historical behavior sequence is used to represent the sequence corresponding to the historical student not completely viewing the video, the first historical score sequence is used to represent the sequence corresponding to the historical student liking to view the video, and the second historical score sequence is used to represent the sequence corresponding to the historical student not liking to view the video; a second embedding processing module configured to perform feature representation on the second viewing behavior sequence based on the second embedding layer of the interest recognition model to obtain a third feature vector and perform feature representation on the second viewing score sequence to obtain a fourth feature vector; a first denoising module configured to perform denoising processing on the first feature vector based on the first interest denoising layer of the interest recognition model to obtain a first denoised vector and perform denoising processing on the second feature vector to obtain a second denoised vector; a second denoising module configured to perform denoising processing on the third feature vector based on the second interest denoising layer of the interest recognition model to obtain a third denoised vector and perform denoising processing on the fourth feature vector to obtain a fourth denoised vector; a vector fusion module configured to perform feature fusion on the first denoised vector, the second denoised vector, the third denoised vector, and the fourth denoised vector based on the multi-head attention layer of the interest recognition model to obtain a target fusion vector; a preference classification module configured to perform multi-target preference classification based on the target fusion vector according to the preference classification layer of the interest recognition model to obtain the target behavior preference corresponding to the target student.

[0084] Exemplarily, before analyzing the behavior preferences of a target student using an interest recognition model, it is first necessary to construct a comprehensive training dataset. Specifically, it is necessary to collect the first historical behavior sequences and the second historical behavior sequences of multiple historical students. Among them, the first historical behavior sequence records the behavior data of historical students watching videos in full, while the second historical behavior sequence records the behavior data of historical students not watching videos in full. By performing interaction analysis on the first historical behavior sequence and the second historical behavior sequence, such as liking, commenting, sharing, etc., the videos liked by historical students are identified to generate the first historical score sequence, and the videos disliked by historical students are identified to generate the second historical score sequence. Next, through manual annotation, combined with the first historical behavior sequence, the second historical behavior sequence, the first historical score sequence, and the second historical score sequence, the behavior preferences of historical students are preference-annotated to obtain the corresponding historical behavior preferences of historical users. The historical behavior preferences are used to characterize the type of behavior preferences corresponding to historical students under the first historical behavior sequence, the second historical behavior sequence, the first historical score sequence, and the second historical score sequence.

[0085] Exemplarily, the first historical behavior sequence, the second historical behavior sequence, the first historical score sequence, and the second historical score sequence corresponding to the historical user are used as the input of the interest recognition model, and the historical behavior preferences corresponding to the historical user are used as the output of the interest recognition model. The cross-entropy is used as the loss function for multi-round model training, and finally an interest recognition model that can accurately identify the behavior preferences of students is obtained.

[0086] Exemplarily, the first analysis module processes the second behavior data (i.e., the viewing behavior data of the target student) to obtain the videos that the target student has watched in full, which are recorded as the first viewing behavior sequence, and the videos that the target student has not watched in full, which are recorded as the second viewing behavior sequence.

[0087] Exemplarily, the second analysis module performs interaction analysis on the second behavior data to analyze the videos liked by the target student according to the viewing results (such as liking, commenting, sharing, etc.) to form the first viewing score sequence, and the videos disliked by the target student to form the second viewing score sequence.

[0088] Exemplarily, the interest recognition model includes a first embedding layer, a second embedding layer, a first interest denoising layer, a second interest denoising layer, a multi-head attention layer, and a preference classification layer. The first embedding processing module uses the first embedding layer of the interest recognition model to extract features from the first viewing behavior sequence to obtain the first feature vector, and uses the same embedding layer to extract features from the first viewing score sequence to obtain the second feature vector.

[0089] Exemplarily, the second embedding processing module uses the second embedding layer of the interest recognition model to extract features from the second viewing behavior sequence to obtain a third feature vector, and uses the same embedding layer to extract features from the second viewing score sequence to obtain a fourth feature vector.

[0090] Exemplarily, the first denoising module uses the first interest denoising layer of the interest recognition model to denoise the first feature vector to obtain a first denoised vector. The second feature vector is similarly denoised to obtain a second denoised vector.

[0091] Exemplarily, the second denoising module uses the second interest denoising layer of the interest recognition model to denoise the third feature vector to obtain a third denoised vector. The fourth feature vector is similarly denoised to obtain a fourth denoised vector.

[0092] Exemplarily, the multi-head attention layer of the interest recognition model is used to fuse the features of the first denoised vector, the second denoised vector, the third denoised vector, and the fourth denoised vector to obtain a target fusion vector. Then, the preference classification layer of the interest recognition model is used to classify the target fusion vector. For example, the target behavior preference of the target student is determined through a classification algorithm (such as softmax, etc.).

[0093] In some embodiments, the data adjustment module includes: a preference score module for summarizing all the target behavior preferences to obtain an overall behavior preference, and obtaining a target preference score corresponding to the target behavior preference according to the target proportion of the target behavior preference in the overall behavior preference; a preference ranking module for ranking the target behavior preferences according to the target preference score to obtain a target preference ranking corresponding to the target student; an exclusion determination module for determining the target exclusion degree of the target student for the associated teaching text according to the target abnormal behavior; and a text generation module for obtaining the target teaching text corresponding to the target student by using a text generation model in combination with the associated teaching text according to the target preference ranking and the target exclusion degree.

[0094] Exemplarily, the data adjustment module includes a preference score module, a preference ranking module, an exclusion determination module, and a text generation module.

[0095] Exemplarily, the preference score module summarizes all the target behavior preferences to obtain an overall behavior preference, and obtains a target preference score according to the target proportion of the target behavior preference in the overall behavior preference.

[0096] Exemplarily, the preference ranking module ranks the target behavior preferences according to the target preference score to obtain a target preference ranking corresponding to the target student. For example, the target behavior preferences are ranked from high to low according to the target preference score to obtain the target preference ranking corresponding to the target student.

[0097] Exemplarily, the rejection determination module determines the target rejection degree of the target student with respect to the associated teaching text according to the target abnormal behavior. For example, by analyzing the target abnormal behavior, its relevance to the associated teaching text is determined. For example, if the target abnormal behavior is "inattentive", it may indicate that the student is not interested in or has a sense of rejection towards the current teaching content. Thus, according to the severity of the target abnormal behavior, the rejection degree of the target student with respect to the associated teaching text is determined. For example, the rejection degree can be divided into three levels: low, medium, and high.

[0098] Exemplarily, the text generation module combines the associated teaching text using a text generation model according to the target preference ranking and the target rejection degree to obtain the target teaching text corresponding to the target student. For example, according to the target preference ranking, the content that the student prefers more is given priority. Considering the target rejection degree, the content and presentation method of the teaching text are adjusted to reduce the student's sense of rejection. For example, if the student does not like text and has a high rejection degree, the complexity of the text content can be reduced, and simpler language and more illustrations can be used. Thus, by integrating the preference ranking and rejection degree information, using the text generation model, and combining with the associated teaching text, teaching content that conforms to the student's preferences is generated. For example, a natural language generation (NLG) model can be used to generate more personalized and acceptable teaching text according to the student's preferences and rejection degree, so as to obtain the target teaching text corresponding to the target student.

[0099] In some embodiments, the text generation module includes: a knowledge recognition module for recognizing knowledge points of the associated teaching text to obtain the target knowledge points corresponding to the associated teaching text; a path recognition module for determining the incorrect explanation path corresponding to the target knowledge points according to the target rejection degree, and obtaining the first correct explanation path corresponding to the target knowledge points according to the target preference ranking; a path trimming module for trimming the incorrect explanation path involved in the first correct explanation path to obtain the second correct explanation path corresponding to the target knowledge points; a knowledge generation module for using the text generation model to obtain the correct teaching text corresponding to the associated teaching text according to the second correct explanation path and the target knowledge points; a text fusion module for fusing the initial teaching text and the correct teaching text to obtain the target teaching text corresponding to the target student.

[0100] Exemplarily, the text generation module includes a knowledge recognition module, a path recognition module, a path trimming module, a knowledge generation module, and a text fusion module.

[0101] Exemplarily, the knowledge recognition module uses natural language processing techniques (such as named entity recognition, keyword extraction, etc.) to extract key knowledge points from the associated teaching text. For example, if the associated teaching text is about the content of "Newton's second law", the key knowledge points extracted may include "Newton's second law", "force", "mass", "acceleration", etc., so as to obtain the target knowledge points corresponding to the associated teaching text.

[0102] Exemplarily, the path recognition module determines which explanation paths may cause rejection or incomprehension of the target student according to the target rejection degree. For example, if the target rejection degree is high, complex mathematical derivations or formal languages may need to be avoided. Then, according to the target preference ranking, the most suitable explanation path for the student's preference is selected. For example, if the student prefers simple and intuitive explanations, select an explanation path that contains more diagrams and examples. Thus, the incorrect explanation path and the first correct explanation path corresponding to the target knowledge points are obtained.

[0103] Exemplarily, the path trimming module checks whether the first correct explanation path contains any incorrect explanation paths. If it does, trim it to ensure that the explanation path meets the student's preference and rejection degree. For example, if the first correct explanation path contains complex mathematical derivations, they can be replaced with simpler explanations or diagrams, so as to obtain the second correct explanation path corresponding to the target knowledge points.

[0104] Exemplarily, the knowledge generation module uses a text generation model (such as a deep learning model, a natural language generation model, etc.) to generate correct teaching text according to the second correct explanation path and the target knowledge points. For example, an NLG model can be used to generate text containing diagrams, examples and simple explanations to make it more understandable.

[0105] Exemplarily, the text fusion module fuses the correct teaching text with the initial teaching text to ensure the coherence and integrity of the teaching content. For example, the correct explanation path can be inserted or replaced in the initial teaching text to make it more in line with the needs of the student, so as to obtain the target teaching text corresponding to the target student.

[0106] The embodiment of the present invention also provides a storage medium for computer-readable storage. The storage medium stores one or more programs, and the one or more programs can be executed by one or more processors to implement the steps of any intelligent education system provided in the specification of the embodiment of the present invention.

[0107] Among them, the storage medium may be an internal storage unit of the terminal device in the foregoing embodiments, such as the hard disk or memory of the terminal device. The storage medium may also be an external storage device of the terminal device, such as a plug-in hard disk equipped on the terminal device, a Smart Media Card (SMC), a Secure Digital (SD) card, a Flash Card, etc.

[0108] Those of ordinary skill in the art can understand that all or some of the steps in the methods disclosed above, and the functional modules / units in the systems and devices, can be implemented as software, firmware, hardware, and their appropriate combinations. In the hardware embodiment, the division between the functional modules / units mentioned above does not necessarily correspond to the division of physical components; for example, a physical component may have multiple functions, or a function or step may be executed by several physical components in cooperation. Some or all of the physical components may be implemented as software executed by a processor, such as a central processing unit, a digital signal processor, or a microprocessor, or may be implemented as hardware, or may be implemented as an integrated circuit, such as an application-specific integrated circuit. Such software can be distributed on a computer-readable medium, which may include a computer storage medium (or non-transitory medium) and a communication medium (or transitory medium). As is well known to those of ordinary skill in the art, the term computer storage medium includes volatile and non-volatile, removable and non-removable media implemented in any method or technology for storing information, such as computer-readable instructions, data structures, program modules, or other data. Computer storage media includes, but is not limited to, RAM, ROM, EEPROM, flash memory or other memory technologies, CD-ROM, digital versatile disks (DVDs) or other optical disk storage, magnetic cassettes, tapes, magnetic disk storage or other magnetic storage devices, or any other medium that can be used to store the desired information and can be accessed by a computer. In addition, as is well known to those of ordinary skill in the art, a communication medium typically includes computer-readable instructions, data structures, program modules, or other data in a modulated data signal such as a carrier wave or other transmission mechanism, and may include any information delivery medium.

[0109] It should be understood that the term "and / or" used in the specification and appended claims of the present invention refers to any combination and all possible combinations of one or more of the associated listed items, and includes these combinations. It should be noted that in this text, the term "comprising", "including" or any other variant thereof is intended to cover non-exclusive inclusion, so that a process, method, article or system comprising a series of elements not only includes those elements, but also includes other elements not expressly listed, or further includes elements inherent in such process, method, article or system. Without further limitation, an element defined by the statement "comprising an..." does not exclude the existence of additional identical elements in the process, method, article or system comprising the element.

[0110] The serial numbers of the above embodiments of the present invention are only for description and do not represent the advantages or disadvantages of the embodiments. The above are only specific embodiments of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art within the technical scope disclosed by the present invention can easily think of various equivalent modifications or substitutions, and these modifications or substitutions should be covered within the protection scope of the present invention. Therefore, the protection scope of the present invention shall be subject to the protection scope of the claims.

Claims

1. An intelligent education system, characterized in that, The system includes: A data acquisition module, configured to obtain first behavior data corresponding to a target student under an initial teaching text and second behavior data corresponding to the target student watching a teaching video; An anomaly analysis module, configured to perform anomaly behavior analysis on the first behavior data to obtain a first anomaly behavior corresponding to the target student and a first anomaly position corresponding to the first anomaly behavior; A data acquisition module, configured to obtain adjacent behavior data corresponding to the target student from the first behavior data according to the first anomaly position; A behavior prediction module, configured to perform behavior reconstruction according to the adjacent behavior data to obtain a target prediction behavior corresponding to the target student under the first anomaly behavior; An anomaly screening module, configured to screen out a second anomaly behavior corresponding to the target student according to the target prediction behavior and the first anomaly behavior; An anomaly classification module, configured to perform anomaly classification according to the second anomaly behavior to obtain an anomaly type corresponding to the second anomaly behavior; An anomaly determination module, configured to determine a target anomaly behavior corresponding to the target student according to the anomaly type, and obtain an associated teaching text corresponding to the target anomaly behavior from the initial teaching text; A preference recognition module, configured to perform behavior analysis on the second behavior data to obtain a target behavior preference corresponding to the target student; A data adjustment module, configured to adjust the associated teaching text according to the target anomaly behavior and the target behavior preference to obtain a target teaching text corresponding to the target student; Among them, the behavior prediction module includes: using a reconstruction behavior layer of a behavior prediction model to perform behavior reconstruction by combining an autoencoder with the adjacent behavior data to obtain a first prediction behavior; using a prediction behavior layer of the behavior prediction model to perform behavior prediction by combining an encoder and a decoder with the adjacent behavior data to obtain a second prediction behavior; using a behavior fusion layer of the behavior prediction model to perform behavior fusion according to the first prediction behavior and the second prediction behavior to obtain the target prediction behavior corresponding to the target student under the first anomaly behavior.

2. The system according to claim 1, characterized in that, The anomaly analysis module includes: A target recognition module, configured to perform target recognition on the first behavior data to obtain target pose information corresponding to the target student in each frame; A pose clustering module, configured to perform pose clustering on the target pose information to obtain a target clustering result corresponding to the target student; A key pose determination module, configured to determine key pose information corresponding to the first behavior data according to the target clustering result; An anomaly scoring module, configured to perform anomaly behavior scoring according to the key pose information to obtain a target anomaly score corresponding to the key pose information; An anomaly behavior determination module, configured to determine the first anomaly behavior corresponding to the target student and the first anomaly position corresponding to the first anomaly behavior according to the target anomaly score; 3. The system according to claim 2, wherein The pose clustering module includes: A joint recognition module, configured to use a joint recognition layer of a feature recognition model to recognize target joints in the target pose information to obtain joint positions corresponding to the target joints of the target student in the target pose information; An information calculation module, configured to calculate the relative position information corresponding to the target joint in the target pose information according to the joint positions by using the information calculation layer of the feature recognition model; An emotion recognition module, configured to determine the target emotion information corresponding to the target student according to the relative position information and the corresponding facial information in the target pose information by using the emotion recognition layer of the feature recognition model; An information clustering module, configured to perform pose clustering according to the relative position information and the target emotion information by using the information clustering layer of the feature recognition model to obtain the target clustering result corresponding to the target student.

4. The system according to claim 3, wherein The information clustering module includes: A center determination module, configured to determine an initial clustering center from the target pose information according to the relative position information and the target emotion information by using the information clustering layer, and obtain the initial position information and the initial emotion information corresponding to the initial clustering center; A distance calculation module, configured to calculate the distance information between the target pose information and the initial clustering center according to the relative position information, the target emotion information, the initial position information, and the initial emotion information by using the information clustering layer; An initial clustering module, configured to cluster the target pose information according to the distance information to obtain an initial clustering result; A fluctuation calculation module, configured to calculate the data distribution fluctuation of each first type of cluster in the initial clustering result to obtain the target fluctuation value corresponding to the first type of cluster; A clustering adjustment module, configured to adjust the initial clustering result according to the target fluctuation value to obtain the target clustering result corresponding to the target student; Wherein, the distance information is obtained according to the following formula: the distance information therebetween; α represents weight information; n represents the quantity corresponding to the relative position information; β k represents the weight information corresponding to the k-th relative position information; pos ik represents the k-th relative position information of the i-th target pose information; pos jk represents the k-th initial position information of the j-th initial clustering center; emo i represents the target emotion information corresponding to the i-th target pose information; emo j represents the initial emotion information corresponding to the j-th initial clustering center.

5. The system according to claim 1, characterized in that, The abnormal screening module includes: A similarity calculation module, configured to calculate the similarity between the target predicted behavior and the first abnormal behavior; A screening processing module, configured to screen out the second abnormal behavior corresponding to the target student from the first abnormal behavior according to the similarity.

6. The system according to claim 1, wherein The abnormal classification module includes: A first feature extraction module, configured to extract the deflection feature of the second abnormal behavior to obtain the first behavior feature corresponding to the second abnormal behavior; A second feature extraction module, configured to extract the gradient feature of the second abnormal behavior by using adjacent information to obtain the second behavior feature corresponding to the second abnormal behavior in the instantaneous behavior; A third feature extraction module, configured to perform emotion recognition on the second abnormal behavior to obtain the third behavior feature corresponding to the second abnormal behavior; A feature fusion module, configured to fuse the first behavior feature, the second behavior feature, and the third behavior feature to obtain the target behavior feature corresponding to the second abnormal behavior; A correlation calculation module, configured to determine a preset type and a preset feature variable corresponding to the preset type, and determine the correlation value between the second abnormal behavior and the preset type according to the target behavior feature and the preset feature variable; A type determination module, configured to determine the abnormal type corresponding to the second abnormal behavior from the preset types according to the correlation value; Wherein, the correlation value is obtained according to the following formula: Among them, rel ij represents the correlation value between the i-th second abnormal behavior and the j-th preset type; v j represents the preset feature variable corresponding to the j-th preset type, actor i represents the target behavior feature corresponding to the i-th second abnormal behavior; δ represents the balance parameter; r represents the number of corresponding features in the target behavior feature; err represents the error parameter; sim(v j , actor i ) represents the similarity between the preset feature variable corresponding to the j-th preset type and the target behavior feature corresponding to the i-th second abnormal behavior; μ represents the adjustment parameter; v w represents the preset feature vector corresponding to the w-th adjacent preset type corresponding to the j-th preset type.

7. The system according to claim 1, wherein The preference recognition module includes: A first analysis module, configured to perform viewing behavior analysis on the second behavior data to obtain a first viewing behavior sequence and a second viewing behavior sequence corresponding to the target student, where the first viewing behavior sequence is used to represent the sequence corresponding to the target student's complete viewing of the video, and the second viewing behavior sequence is used to represent the sequence corresponding to the target student's incomplete viewing of the video; A second analysis module, configured to perform viewing result analysis on the second behavior data to obtain a first viewing score sequence and a second viewing score sequence corresponding to the target student, where the first viewing score sequence is used to represent the sequence corresponding to the target student's liking to watch the video, and the second viewing score sequence is used to represent the sequence corresponding to the target student's disliking to watch the video; A first embedding processing module, configured to perform feature representation on the first viewing behavior sequence based on the first embedding layer of the interest recognition model to obtain a first feature vector and perform feature representation on the first viewing score sequence to obtain a second feature vector. The interest recognition model is trained by using the first historical behavior sequence, the second historical behavior sequence, the first historical score sequence, and the second historical score sequence corresponding to the historical students as model inputs, and using the historical behavior preferences corresponding to the historical students under the first historical behavior sequence, the second historical behavior sequence, the first historical score sequence, and the second historical score sequence as model outputs; where the first historical behavior sequence is used to represent the sequence corresponding to the historical student's complete viewing of the video, the second historical behavior sequence is used to represent the sequence corresponding to the historical student's incomplete viewing of the video, the first historical score sequence is used to represent the sequence corresponding to the historical student's liking to watch the video, and the second historical score sequence is used to represent the sequence corresponding to the historical student's disliking to watch the video; A second embedding processing module, configured to perform feature representation on the second viewing behavior sequence based on the second embedding layer of the interest recognition model to obtain a third feature vector and perform feature representation on the second viewing score sequence to obtain a fourth feature vector; A first denoising module, configured to perform denoising processing on the first feature vector based on the first interest denoising layer of the interest recognition model to obtain a first denoised vector and perform denoising processing on the second feature vector to obtain a second denoised vector; A second denoising module, configured to perform denoising processing on the third feature vector based on the second interest denoising layer of the interest recognition model to obtain a third denoised vector and perform denoising processing on the fourth feature vector to obtain a fourth denoised vector; A vector fusion module, configured to perform feature fusion on the first denoised vector, the second denoised vector, the third denoised vector, and the fourth denoised vector based on the multi-head attention layer of the interest recognition model to obtain a target fusion vector; A preference classification module, configured to perform multi-target preference classification based on the target fusion vector according to the preference classification layer of the interest recognition model to obtain the target behavior preference corresponding to the target student.

8. The system according to claim 1, characterized in that, The data adjustment module includes: A preference score module, which is used to summarize all the target behavior preferences to obtain an overall behavior preference, and obtain a target preference score corresponding to the target behavior preference according to the target proportion of the target behavior preference in the overall behavior preference; A preference ranking module, which is used to rank the target behavior preferences according to the target preference scores to obtain a target preference ranking corresponding to the target student; An exclusion determination module, which is used to determine the target exclusion degree of the target student for the associated teaching text according to the target abnormal behavior; A text generation module, which is used to obtain the target teaching text corresponding to the target student by using a text generation model according to the target preference ranking and the target exclusion degree in combination with the associated teaching text.

9. The system according to claim 8, characterized in that, The text generation module includes: A knowledge identification module, which is used to identify knowledge points of the associated teaching text to obtain target knowledge points corresponding to the associated teaching text; A path identification module, which is used to determine an incorrect explanation path corresponding to the target knowledge point according to the target exclusion degree, and obtain a first correct explanation path corresponding to the target knowledge point according to the target preference ranking; A path trimming module, which is used to trim the incorrect explanation path involved in the first correct explanation path to obtain a second correct explanation path corresponding to the target knowledge point; A knowledge generation module, which is used to obtain a correct teaching text corresponding to the associated teaching text by using the text generation model according to the second correct explanation path and the target knowledge point; A text fusion module, which is used to fuse the initial teaching text and the correct teaching text to obtain the target teaching text corresponding to the target student.

Images