A method and system for after-class evaluation of students based on AI

By combining the multi-head attention mechanism and interest matrix decomposition technology of the Transformer model and LFM model, the problems of low efficiency and low accuracy of post-class comments in the existing technology are solved, and personalized and multi-dimensional automated post-class comments are realized. Adapting to the needs of different learning scenarios, the accuracy and humanization of comments have been improved.

CN118378956BActive Publication Date: 2024-11-15HANGZHOU JIEJING SCI & TECH
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202410811999.1
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2024-06-21
Publication Date
2024-11-15
Estimated Expiration
2044-06-21

AI Technical Summary

Technical Problem

In the existing technology, after-class review methods have problems such as low efficiency, high cost, unstable quality and lack of personalization. Manual reviews take a long time and are easily affected by subjective factors, while machine reviews are not accurate and credible, which is difficult to Adapt to different learning scenarios and needs.

Method used

The multi-head attention mechanism and LFM model of the Transformer model are adopted to analyze the students' learning data and decompose the interest matrix, and personalized multi-dimensional after-class comments are generated to explore the relationship comments related to learning habits, preferences and abilities.

Benefits of technology

It realizes the personalization and multi-dimensional adaptability of automated after-class comments, improves the accuracy and humanization of comments, adapts to the needs of different learning scenarios, and reduces labor costs and time consumption.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118378956B_ABST
    Figure CN118378956B_ABST
Patent Text Reader

Abstract

The present invention provides a method and system for after-class review based on AI students, the method comprising: obtaining student learning data, and performing data preprocessing on the learning data, wherein the learning data includes interest data, constructing multi-dimensional review data based on the learning data, and constructing a student learning interest matrix according to the student learning data; determining latent data, and decomposing the interest matrix into a latent relationship matrix according to an LFM model algorithm and the latent data, wherein the latent relationship matrix includes multi-dimensional review data; inputting the latent relationship matrix into a multi-head attention mechanism layer of a Transformer model, and adjusting the weights of a network layer through a fully connected feedforward neural network connected to the multi-head attention mechanism layer; and outputting the output review data of the corresponding student using an activation function in the output layer of the Transformer model, so as to provide personalized reviews on the learning data of the corresponding student.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the field of educational technology, and in particular to a method and system for after-class evaluation of students based on AI. Background Art

[0002] At present, there are two main ways of post-class comments: one is manual comments, that is, teachers or teaching assistants manually score and comment on the content submitted by students' homework, tests, videos, etc. The advantage of this method is that it can give personalized and professional comments based on the specific situation of the students, but it also has some disadvantages, such as low efficiency, long time consumption, high cost, unstable quality, and easy to be affected by subjective factors. The other is machine comments, that is, computer programs automatically score and comment on students' content according to preset standards and models. The advantage of this method is that it can improve the efficiency and consistency of comments, reduce the cost and manpower requirements of comments, but it also has some disadvantages, such as the low accuracy and credibility of machine comments, lack of humanized and personalized design, and difficulty in adapting to different learning scenarios and needs. Summary of the invention

[0003] One of the invention purposes of the present invention is to provide a method and system for after-class evaluation of students based on AI. The method and system utilize the multi-head attention mechanism of the Transformer model to perform attention analysis on students' learning-related pictures, texts, videos, and audios, and conduct a comprehensive classification analysis of students' learning styles, learning problems, learning interests, and learning effects during the learning process, and then generate after-class learning evaluations of students in different dimensions based on the classification analysis results. The after-class evaluation in the present invention has the characteristics of personalized multi-dimensional automatic evaluation, so the automated after-class evaluation of the present invention can adapt to the needs of different learning scenarios.

[0004] Another invention object of the present invention is to provide a method and system for after-class evaluation of students based on AI. The method and system also obtain interest data and latent data related to student learning, and construct an interest matrix based on the student's learning data using the LFM model. The present invention uses the LFM model to decompose the interest matrix into two latent relationship matrices, wherein the latent relationship matrix includes corresponding interest data and corresponding comment data, and subsequent data processing is performed based on the two latent relationship matrices. Therefore, the present invention can obtain the latent relationship between interest data and comment data based on the LFM model, so that the LFM model can be used to mine the relationship comments related to students' learning habits, learning preferences, and learning abilities during the learning process.

[0005] Another inventive object of the present invention is to provide a method and system for after-class evaluation of AI students. The method and system input the two latent class relationship matrices obtained by decomposing the student interest data into the multi-head attention mechanism layer of the corresponding Transformer model, and use the multi-head attention mechanism of the Transformer model to deeply understand the implicit learning data of the students in the learning process, and output personalized evaluation data based on the implicit learning data.

[0006] In order to achieve at least one of the above-mentioned invention objects, the present invention further provides a method for post-class evaluation of students based on AI, the method comprising:

[0007] Acquire student learning data and perform data preprocessing on the learning data, wherein the learning data includes interest data, construct multi-dimensional review data based on the learning data, and construct a student learning interest matrix based on the student learning data;

[0008] Determine latent class data, and decompose the interest matrix into a latent class relationship matrix according to the LFM model algorithm and the latent class data, wherein the latent class relationship matrix includes multi-dimensional review data;

[0009] Inputting the latent class relationship matrix into the multi-head attention mechanism layer of the Transformer model, and adjusting the weight of the network layer through the fully connected feedforward neural network connected to the multi-head attention mechanism layer;

[0010] The output layer of the Transformer model uses an activation function to output the output review data of the corresponding student, which is used to personalize the review of the learning data of the corresponding student.

[0011] According to one of the preferred embodiments of the present invention, the method for acquiring student learning data includes: acquiring the student's learning video, picture, audio and text learning objects, and acquiring the browsing data of the learning objects as interest data, wherein the browsing data includes browsing learning objects, browsing time, click data during browsing and browsing progress, annotating each learning object, wherein the annotation content includes learning object type, learning object attributes, learning object channel, and annotating the acquired browsing data, converting the annotated non-numerical data into numerical data, and obtaining the student's learning data.

[0012] According to another preferred embodiment of the present invention, the data preprocessing method includes: after acquiring the numerical learning data, further performing the following data cleaning operations based on the numerical data: deleting invalid data and erroneous data in the learning data, and interpolating the missing data using a linear interpolation algorithm, wherein the linear interpolation formula is: y=y0+α(y1-y0), wherein y is the insertion value, y1 and y0 are the values ​​at both ends of the insertion value y respectively, and α is the linear coefficient, and further obtaining clean learning data; constructing an interest matrix based on the clean learning data, and normalizing the elements of the interest matrix.

[0013] According to another preferred embodiment of the present invention, the method for constructing the multidimensional review data includes: pre-configuring the review data of each learning object and the review data of the browsing data, constructing the mapping relationship between the learning object data, the browsing data and the corresponding review data, and defining the learning object review data as M=(m1, m2, m3, m4...m n ), where m n Represents the review data of the nth learning object; the review data of the browsed data type is defined as V=(v1,v2,v3,v4.....v s ),v s Represents the review data of the sth browsing data type. The review data of the learning object and the review data of the browsing data type are spliced ​​forward and backward using a splicing algorithm to obtain the following complete multi-dimensional review data: E = (m1-v2, m1-v3, m1-v4, ... m1-v s ...m2-v s ..m n -v s ), where - represents the concatenation character, m n -v s Represents a review data, the multidimensional review data includes n×s review data, and the constructed multidimensional review data E is used to construct an interest matrix.

[0014] According to another preferred embodiment of the present invention, the method for constructing the interest matrix includes: pre-establishing a correspondence, constructing a first matrix F1 based on preprocessed clean learning data, wherein a second matrix F2 of review data is established according to the mapping relationship between the learning object data, browsing data and corresponding review data, performing a dot multiplication on the mapping relationship between the first matrix F1 and the second matrix F2 to obtain the interest matrix F=F1·F2, and further calculating the latent class relationship matrix of the students' learning process according to the interest matrix.

[0015] According to another preferred embodiment of the present invention, the method for obtaining the latent class relationship matrix includes: predetermining the latent class relationship type and dimension, using the LFM model and decomposing the interest matrix F into the following latent class relationship matrix W= = , where u represents the student interest data, k represents the predetermined latent class data, and T represents the transpose symbol of the matrix. represents the relationship matrix between the student interest data u and the latent data k, i represents the corresponding review data, It represents the relationship matrix between the students' latent data and the review data.

[0016] According to another preferred embodiment of the present invention, when the interest matrix is ​​decomposed to obtain the latent class relationship matrix W= After that, the latent class relationship matrix is ​​further input into the multi-head attention mechanism layer of the Transformer model, and the following result is output in the multi-head attention mechanism layer: , where head represents the attention head, head i represents the head attention of the i-type review data in the latent class relationship matrix. In the multi-head attention mechanism layer formula, Q, K and V are the query matrix, key matrix and value matrix respectively, and W is the latent class relationship matrix, where the head i The latent class relationship matrix W is obtained by performing h linear transformations of Q, K and V.

[0017] According to another preferred embodiment of the present invention, the following formula is used to perform dot product scaling in the Transformer model: , where d k The output is the scaling factor, including the student learning review data related to the latent class relationship.

[0018] In order to achieve at least one of the above-mentioned invention purposes, the present invention further provides a system for after-class review based on AI students, and the system executes the above-mentioned method for after-class review based on AI students.

[0019] The present invention further provides a computer-readable storage medium, which stores a computer program, and the computer program is executed by a processor to implement the above-mentioned AI-based after-class evaluation method for students. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] Figure 1 Shown is a flow chart of an after-class review method based on AI students according to the present invention. DETAILED DESCRIPTION

[0021] The following description is used to disclose the present invention so that those skilled in the art can implement the present invention. The preferred embodiments described below are only examples, and those skilled in the art can think of other obvious variations. The basic principles of the present invention defined in the following description can be applied to other embodiments, variations, improvements, equivalents, and other technical solutions that do not deviate from the spirit and scope of the present invention.

[0022] It is to be understood that the term "one" should be understood as "at least one" or "one or more", that is, in one embodiment, the number of an element may be one, while in another embodiment, the number of the element may be multiple, and the term "one" should not be understood as a limitation on the quantity.

[0023] Please combine Figure 1 , the present invention discloses a method and system for after-class review based on AI students, wherein the method mainly includes the following parts: first, it is necessary to obtain the course learning data related to the students, wherein the course learning data of the students includes but is not limited to learning videos, pictures, audio and text as learning objects, and the course learning data in the present invention can be obtained from the corresponding learning platform, such as the course English learning platform. Since each student has his own account as a unique ID to log in to the learning platform, the present invention monitors the learning process and learning results of the corresponding student account on the school platform after the account authorization tracking of the corresponding learning platform to obtain the learning record of the student on the platform. The present invention constructs an interest matrix based on the learning data of the students, and further uses the algorithm of the LFM model to decompose the interest matrix of the students into a latent class relationship matrix, and inputs the latent class relationship matrix into the multi-head attention mechanism layer in the Transformer model for attention, so that the student review data finally outputted pays more attention to the key learning data of the students. It is worth mentioning that the construction of the latent class relationship matrix in the present invention can effectively mine the implicit data in the students' own learning process, such as the students' attention to a certain course, the learning efficiency of a certain course, and the students' own learning habits, learning preferences, behavioral preferences, etc. Thereby, personalized after-class comments can be obtained. Compared with traditional mechanical after-class comments, the present invention can realize personalized in-depth analysis of different influencing factors in the learning process of students, thereby improving the accuracy of after-class comments.

[0024] Specifically, the present invention processes the learning data of the students, wherein the data processing process includes: data labeling, data cleaning and data conversion of the learning data, wherein the data labeling method mainly includes: labeling of the students' learning objects and browsing data, wherein the learning objects are objects with physical text such as videos, pictures, audio and texts that the students browse. For example, English cartoons, English electronic novels, English songs and English magazines, etc., the above learning objects of the present invention are only examples, and the present invention will not repeat them in detail. The labeling of the learning objects of the present invention includes but is not limited to the learning object type, learning object attributes and learning object channels, wherein the learning object type can be the form of its display, such as audio, video and text, etc.; the object attribute is an explanation of the content part of the learning object. It should be noted that the attributes of the learning objects in the present invention can be obtained by using methods including but not limited to keyword recognition and extraction analysis. For example, for electronic novels, the text content can be extracted for keyword analysis to determine the type of the novel. Of course, in another preferred embodiment of the present invention, the learning object attributes can obtain the attribute information of the corresponding learning object according to the attributes pre-defined by the learning platform. For example, the corresponding learning audio and learning text will be pre-stored in the general learning platform, and the learning audio and learning text will be classified according to the content. The present invention can extract keywords according to the results of the pre-classification to obtain corresponding attribute information. The learning object channel represents the source of the learning course, such as some famous teacher courses can be obtained from external channels. Further, the present invention also marks the browsing data, wherein the browsing data in the present invention includes but is not limited to the students' browsing learning objects, browsing time, click data and browsing progress during the browsing process for a certain course. Since some of the annotated data may be non-numerical data, the present invention further converts the non-numerical data into numerical data, wherein the method for converting the non-numerical data into numerical data can be implemented by including but not limited to one-hot encoding and map function, and the above implementation method is a prior art, and the present invention will not be described in detail. It should be noted that the browsing data in the present invention is the student's interest data for a certain course, and some implicit data can be obtained from the browsing data, and the implicit data can be further mined to obtain a more personalized and refined and accurate course learning review during the student's learning process.

[0025] The present invention also requires data preprocessing of the collected student data, wherein the data preprocessing method includes data cleaning and supplementation, and the specific method includes: after obtaining the numerical learning data, further performing the following data cleaning operations according to the numerical data: deleting invalid data and erroneous data in the learning data, and interpolating the missing data using a linear interpolation algorithm, wherein the linear interpolation formula is: y=y0+α(y1-y0), wherein y is the insertion value, y1 and y0 are respectively the values ​​at both ends of the insertion value y, and α is the linear coefficient, and further obtaining clean learning data; constructing an interest matrix based on the clean learning data, and normalizing the elements of the interest matrix. In the above data preprocessing process, the availability of the data can be effectively improved. Avoid the influence of impurity data on the accuracy and training of the model.

[0026] It is worth mentioning that in one of the core preferred embodiments of the present invention, the present invention needs to pre-configure multi-dimensional review data for realizing personalized review of student keyword learning data under the multi-head attention mechanism of the Transformer model. The method for constructing the multi-dimensional review data includes the following steps: pre-constructing the mapping relationship between the learning object data, browsing data and corresponding review data, defining the learning object review data as M=(m1, m2, m3, m4...m n ), where m n Represents the review data of the nth learning object; the review data of the browsed data type is defined as V=(v1,v2,v3,v4.....v s ),v s Represents the review data of the sth browsing data type. The review data of the learning object and the review data of the browsing data type are spliced ​​forward and backward using a splicing algorithm to obtain the following complete multi-dimensional review data: E = (m1-v2, m1-v3, m1-v4, ... m1-v s ...m2-v s ..m n -v s ), where - represents the concatenation character, m n -v sRepresents a comment data, the multidimensional comment data includes n×s comment data, and the constructed multidimensional comment data E is used to construct an interest matrix. In the above core preferred embodiment, the present invention constructs comment data for the learning object data comment data and the browsing data respectively, and constructs the corresponding comment data by splicing the comment data. Through the construction method of the above comment data, more abundant comment results can be obtained. For example, the comment data of the learning object includes the English reading comprehension of a certain video, and the comment data of the browsing data can be set according to the number of students' browsing times and browsing time to include but not limited to long-term learning, short-term browsing, note learning, repeated learning, etc., then according to the above splicing algorithm, the comment data of the students' English reading comprehension of a certain video long-term learning / short-term browsing / note learning / repeated learning can be obtained. It should be noted that the long-term learning and short-term browsing described in the present invention can be judged by setting the upper and lower thresholds of the browsing time, and the corresponding comment data are extracted. For the comment data of other browsing data, different rules can be set according to the form of the browsing data, and the present invention will not be described in detail.

[0027] Further, the present invention constructs the interest matrix after obtaining clean learning data and multidimensional comment data, wherein the construction method of the interest matrix includes: constructing a first matrix F1 based on the preprocessed clean learning data, wherein the second matrix F2 of comment data is established according to the mapping relationship between the learning object data, browsing data and corresponding comment data, and the mapping relationship between the first matrix F1 and the second matrix F2 is multiplied to obtain the interest matrix F=F1·F2, and further calculating the hidden class relationship matrix in the student learning process according to the interest matrix. It should be noted that the second matrix of comment data is implemented according to the corresponding mapping relationship list. For example, when comment data is established for English video learning, when the student clicks and browses the English video, the comment data for the English video is constructed according to the mapping relationship list. At this time, the elements at the corresponding positions in the first matrix F1 have the learning object and attribute data of the English video, and the corresponding positions in the second matrix F2 also have the comment data corresponding to the mapping of the learning object of the English video. When there is no corresponding learning object in the first matrix F1, there is no corresponding comment data in the second matrix F2.

[0028] Furthermore, the method for obtaining the latent class relationship matrix of the present invention includes: predetermining the latent class relationship type and dimension, using the LFM model and decomposing the interest matrix F into the following latent class relationship matrix W= = , where u represents the student interest data, k represents the predetermined latent class data, and T represents the transpose symbol of the matrix. represents the relationship matrix between the student interest data u and the latent data k, i represents the corresponding review data, It represents the relationship matrix between the student's latent data and the comment data. It should be noted that the latent relationship dimension can be set according to actual needs. For example, data of multiple dimensions such as "number of clicks", "login time", "browsing time", "learning object attributes" can be set as latent data, and the interest matrix F can be further disassembled according to the latent data dimension to obtain the above-mentioned latent relationship matrix. Through the latent relationship matrix, data such as the user's implicit preferences in the learning process can be obtained, so that personalized comments can be made. It should be noted that since the interest matrix in the present invention includes comment data, the student's hidden learning preference data and the hidden comment data corresponding to the hidden learning preference data can also be obtained during the calculation of the above-mentioned latent relationship matrix.

[0029] Furthermore, when the interest matrix is ​​decomposed to obtain the latent class relationship matrix W= After that, the latent class relationship matrix is ​​further input into the multi-head attention mechanism layer of the Transformer model, and the following result is output in the multi-head attention mechanism layer: , where head represents the attention head, head i represents the head attention of the i-type review data in the latent class relationship matrix. In the multi-head attention mechanism layer formula, Q, K and V are the query matrix, key matrix and value matrix respectively, and W is the latent class relationship matrix, where the head i The latent class relationship matrix W is obtained by performing h linear transformations of Q, K and V. The multi-head attention mechanism layer of the Transformer model in the present invention can realize the output of key review data based on the user's learning implicit preference data, which is more humane than the traditional mechanized review. In order to accelerate the training and output of the Transformer model, the following formula is used in the Transformer model for dot product scaling: , where d k The output is the scaling factor, including the student learning review data related to the latent class relationship.

[0030] The embodiments disclosed in the present invention, the process described above with reference to the flowchart can be implemented as a computer software program. The embodiments disclosed in the present invention include a computer program product, which includes a computer program carried on a computer-readable medium, and the computer program contains a program code for executing the method shown in the flowchart. In such an embodiment, the computer program can be downloaded and installed from the network through the communication part, and / or installed from a removable medium. When the computer program is executed by the central processing unit (CPU), the above-mentioned functions defined in the method of the present application are executed. It should be noted that the above-mentioned computer-readable medium of the present application can be a computer-readable signal medium or a computer-readable storage medium or any combination of the above two. The computer-readable storage medium can be, for example, but not limited to, a system, device or device of an electrical, magnetic, optical, electromagnetic, infrared segment, or semiconductor, or any combination of the above. More specific examples of computer-readable storage media may include, but are not limited to, an electrical connection with one or more wire segments, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above. In the present application, a computer-readable storage medium may be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, apparatus, or device. In the present application, a computer-readable signal medium may include a data signal propagated in baseband or as part of a carrier wave, which carries a computer-readable program code. Such propagated data signals may take a variety of forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination of the above. A computer-readable signal medium may also be any computer-readable medium other than a computer-readable storage medium, which may send, propagate, or transmit a program for use by or in conjunction with an instruction execution system, apparatus, or device. The program code embodied on the computer readable medium may be transmitted using any appropriate medium, including but not limited to wireless, electrical wire, optical cable, RF, etc., or any suitable combination of the foregoing.

[0031] The flow chart and block diagram in the accompanying drawings illustrate the possible architecture, function and operation of the system, method and computer program product according to various embodiments of the present invention. In this regard, each square box in the flow chart or block diagram can represent a module, a program segment or a part of a code, and the module, the program segment or a part of the code contains one or more executable instructions for realizing the specified logical function. It should also be noted that in some alternative implementations, the functions marked in the square box can also occur in a sequence different from that marked in the accompanying drawings. For example, two square boxes represented in succession can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each square box in the block diagram and / or flow chart, and the combination of the square boxes in the block diagram and / or flow chart can be implemented with a dedicated hardware-based system that performs the specified function or operation, or can be implemented with a combination of dedicated hardware and computer instructions.

[0032] Those skilled in the art should understand that the embodiments of the present invention described above and shown in the accompanying drawings are only examples and do not limit the present invention. The objectives of the present invention have been fully and effectively achieved. The functional and structural principles of the present invention have been demonstrated and explained in the embodiments. Without departing from the principles, the implementation methods of the present invention may be deformed or modified in any way.

Claims

1. A post-class review method based on AI students, characterized in that: The method comprises: Acquire student learning data and perform data preprocessing on the learning data, wherein the learning data includes interest data, construct multi-dimensional review data based on the learning data, and construct a student learning interest matrix based on the student learning data; Determine latent class data, and decompose the interest matrix into a latent class relationship matrix according to the LFM model algorithm and the latent class data, wherein the latent class relationship matrix includes multi-dimensional review data; Inputting the latent class relationship matrix into the multi-head attention mechanism layer of the Transformer model, and adjusting the weight of the network layer through the fully connected feedforward neural network connected to the multi-head attention mechanism layer; Outputting the output review data of the corresponding student using an activation function at the output layer of the Transformer model, so as to provide a personalized review of the learning data of the corresponding student; The method for constructing multidimensional review data includes: pre-configuring review data of each learning object and review data of browsing data, constructing a mapping relationship between the learning object data, browsing data and corresponding review data, defining the learning object review data as M=(m1, m2, m3, m4...m n ), where m n Represents the review data of the nth learning object; the review data of the browsed data type is defined as V= (v1, v2, v3, v4.....v s ), v s Represents the review data of the sth browsing data type. The review data of the learning object and the review data of the browsing data type are spliced ​​forward and backward using a splicing algorithm to obtain the following complete multi-dimensional review data: E = (m1-v2, m1-v3, m1-v4, ... m1-v s ...m2-v s ..m n -v s ), where - represents the concatenation character, m n -v s represents a review data, the multidimensional review data includes n×s review data, and the constructed multidimensional review data E is used to construct an interest matrix; The method for constructing the interest matrix includes: pre-establishing a first matrix F1 corresponding to the pre-processed clean learning data, wherein a second matrix F2 of the review data is established according to the mapping relationship between the learning object data, the browsing data and the corresponding review data, performing a point multiplication on the mapping relationship between the first matrix F1 and the second matrix F2 to obtain the interest matrix F=F1·F2, and further calculating the hidden class relationship matrix of the learner in the learning process according to the interest matrix; The method for obtaining the latent class relationship matrix includes: predetermining the latent class relationship type and dimension, using the LFM model and decomposing the interest matrix F into the following latent class relationship matrix W= = , where u represents the student interest data, k represents the predetermined latent class data, and T represents the transpose symbol of the matrix. represents the relationship matrix between the student interest data u and the latent data k, i represents the corresponding review data, It represents the relationship matrix between the students' latent data and the review data; The method for acquiring student learning data includes: acquiring the student's learning video, picture, audio and text learning objects, and acquiring the browsing data of the learning objects as interest data, wherein the browsing data includes browsing learning objects, browsing time, click data during browsing and browsing progress, annotating each learning object, wherein the annotation content includes learning object type, learning object attributes, learning object channel, and annotating the acquired browsing data, converting the annotated non-numerical data into numerical data, and obtaining the student's learning data.

2. According to claim 1, a method for post-class review based on AI students is characterized in that: The data preprocessing method includes: after obtaining the numerical data, further performing the following data cleaning operations based on the numerical data: deleting invalid data and erroneous data in the learning data, and interpolating the missing data using a linear interpolation algorithm, wherein the linear interpolation formula is: y=y0+α(y1-y0), wherein y is the insertion value, y1 and y0 are respectively the values ​​at both ends of the insertion value y, and α is the linear coefficient, and further obtaining clean learning data; constructing an interest matrix based on the clean learning data, and normalizing the elements of the interest matrix.

3. According to the AI-based post-class review method of claim 1, it is characterized in that: When the interest matrix is ​​decomposed to obtain the latent class relationship matrix W= After that, the latent class relationship matrix is ​​further input into the multi-head attention mechanism layer of the Transformer model, and the following result is output in the multi-head attention mechanism layer: , where head represents the attention head, head i represents the attention of the i-type review data in the latent class relationship matrix. In the multi-head attention mechanism layer formula, Q, K and V are the query matrix, key matrix and value matrix respectively, and W is the latent class relationship matrix, where the head i The latent class relationship matrix W is obtained by performing h linear transformations of Q, K and V.

4. According to claim 3, a method for post-class review based on AI students is characterized in that: The following formula is used for dot product scaling in the Transformer model: , where d k The output is the scaling factor, including the student learning review data related to the latent class relationship.

5. An AI-based after-class review system for students, characterized in that: The system executes an AI-based after-class evaluation method for students as described in any one of claims 1 to 4 above.

6. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, and the computer program is executed by a processor to implement the AI-based after-class evaluation method for students as described in any one of claims 1 to 4.