Education recommendation method and system based on similarity alignment and uniqueness uniformity

Abstract

The invention provides an education recommendation method based on similarity alignment and uniqueness uniformity, and belongs to the field of education recommendation. The method comprises the following steps: S1, obtaining individual features of students and content features of courses, and constructing a student-course interaction diagram through interaction behaviors; s2, aggregating the multi-dimensional information to generate an embedded representation of each student and course, and constructing an embedded representation space; s3, by means of a similarity alignment loss function, after the embedding representation of the student is aligned with the embedding representation of the learned course, similar courses are searched for the student, and further alignment is carried out; s4, realizing embedding representation uniform distribution of students and courses in the embedding representation space by using a unique uniform loss function; s5, carrying out multi-task training, optimizing a similar alignment-unique uniformity index, and obtaining a course recommendation scheme; and S6, if the learning state of the student changes, repeatedly executing the steps S3-S5, and optimizing the course recommendation scheme. According to the invention, accurate alignment of students and courses and balanced distribution of resources are ensured, and recommended contents can be intelligently adjusted.

Description

Technical Field

[0001] The present invention relates to the field of education recommendation, and in particular to an education recommendation method and system based on similarity alignment and uniqueness uniformity. Background Art

[0002] With the rapid development of information technology, the information resources stored on the Internet platform are growing exponentially, and users often need to spend a lot of time to find the information resources they need. In recent years, recommendation algorithms have been widely used in many fields. For example, they have achieved remarkable results in video recommendations and e-commerce recommendations, greatly saving users' time costs. In the intelligent education platform, students' learning behavior will generate massive data. By analyzing these historical behavior data and their inherent characteristics, we can provide students with course recommendations.

[0003] For example, CN114154839B discloses a course recommendation method based on online education platform data, which utilizes embedding technology and graph neural network models, based on students' basic information and historical data generated by students during the learning process, taking into account the impact of interaction data between different students, mining the relationship between students, and taking into account the similarity between courses and historical interaction data between students and courses. It can provide students with recommended courses, but it still has the following problems: it only provides course recommendations to students based on interaction data between students and students, students and courses, and similarity between courses, and it is difficult to ensure the matching degree between students and courses, that is, it is difficult to ensure students' learning efficiency and learning quality, and it is also difficult to ensure balanced distribution of course resources. In the recommendation process, there may be a tendency for course resources to be centralized. Summary of the invention

[0004] The technical problem to be solved by the present invention is how to improve the matching degree between students and courses when recommending courses to students to ensure the learning quality of students, and how to achieve reasonable allocation of learning resources when recommending.

[0005] The present invention solves the above-mentioned problem by the following technical means: an education recommendation method based on similarity alignment and uniqueness uniformity, comprising the following steps:

[0006] S1: Obtain the individual characteristics of students and the knowledge structure and difficulty of the course content, and construct a student-course interaction diagram through interactive behaviors;

[0007] S2: Based on the individual characteristics of each student and the knowledge structure and difficulty of the course content, the embedding representation of each student and course is generated by aggregating multi-dimensional information to construct an embedding representation space;

[0008] S3: Using the similarity alignment loss function, after the student’s embedding representation is aligned with the embedding representation of the learned course, similar courses are found for the student and further aligned;

[0009] S4: Use the unique uniform loss function to achieve uniform distribution of embedding representations of students and courses in the embedding representation space;

[0010] S5: Perform multi-task training, optimize the similarity alignment-unique uniformity index, and obtain the course recommendation scheme through similarity calculation;

[0011] S6: If the student’s learning status changes, recalculate the embedded representation of the student and the course, and execute S3-S5 again to optimize the course recommendation plan.

[0012] Preferably, in step S1, the individual characteristics of the students include knowledge mastery, learning speed and cognitive level; the interactive behavior includes implicit feedback and explicit feedback, wherein implicit feedback indicates that there is an interactive record between the student and the course, and explicit feedback indicates that the student has a clear preference behavior for the course, including likes and high ratings.

[0013] Preferably, in step S1, the student-course interaction graph is represented by G = (N, E), wherein N = {S∪C} represents the set of all students S and courses C, and E represents the courses that the students have studied.

[0014] Preferably, in step S3, the process of capturing similar sample pairs by cosine similarity is: using the formula Calculate the cosine similarity between each student or course and all courses or students to obtain the corresponding similarity score, select similar students or courses from them, and generate similar sample pairs, including student-similar courses and course-similar students.

[0015] Preferably, in step S3, the similarity alignment loss function is Where P pos are sample pairs of students and courses they have learned, are the student embedding representation and course embedding representation after l2 regularization in the sample pairs of students and the courses they have learned, respectively; P sim_pos is a pair of similar samples captured by cosine similarity, They are respectively the student embedding representation and course embedding representation after l2 regularization in similar sample pairs.

[0016] Since the interaction data between students and courses is generally sparse, it is difficult for students to effectively align the courses they are potentially good at or interested in. By obtaining similar sample pairs, more course options can be provided to students, which can help students discover the courses they are potentially good at or interested in, and is conducive to the personalized development of students.

[0017] Preferably, in step S4, the unique uniform loss function is: Where f(s) and f(s') are the embedding representations of students and similar students, respectively; f(c) and f(c') are the embedding representations of courses and similar courses, respectively; Unique represents the uniqueness strategy; P student and P course They are represented as the student set and the course set respectively.

[0018] Preferably, the uniqueness strategy is: E'=Unique(E), which means that duplicate samples are removed from the sample set E to obtain a new sample set E'.

[0019] After applying the uniqueness strategy to the sampled samples, duplicate samples are effectively filtered out. Specifically, for duplicate students or courses generated during the sampling process, a uniqueness strategy is introduced to filter duplicate samples, thereby ensuring that when calculating the uniformity of the student and course embedding representation space, it can effectively avoid the lack of uniformity caused by the sample sampling during training that may cause the sample to be calculated multiple times, and further avoid the problem of serious deviation in uniformity.

[0020] Preferably, the similarity alignment-unique uniformity index in step S5 is: Loss = L similar_align +αL unique_uniform , where α is used to adjust the weights between different tasks.

[0021] Preferably, in step S6, the student's learning status changes, including the student completing a course or the student's knowledge mastery improving.

[0022] Corresponding to the above method, the present invention also provides an education recommendation system based on similarity alignment and uniqueness uniformity, comprising the following modules:

[0023] Mapping module: used to obtain students’ individual characteristics and the knowledge structure and difficulty of course content, and to construct a student-course interaction map through interactive behaviors;

[0024] Generate Embedding Representation Module: This module is used to generate the embedding representation of each student and course by aggregating multi-dimensional information based on the individual characteristics of each student and the knowledge structure and difficulty of the course content, and construct the embedding representation space;

[0025] Similarity alignment module: It is used to find similar courses for students and further align them after the embedding representation of the students is aligned with the embedding representation of the learned courses using the similarity alignment loss function;

[0026] Unique uniform module: used to achieve uniform distribution of embedding representations of students and courses in the embedding representation space using the unique uniform loss function;

[0027] Obtaining recommendation scheme module: used for multi-task training, optimizing similarity alignment-unique uniformity index and obtaining course recommendation schemes through similarity calculation;

[0028] Optimize recommendation scheme module: It is used to recalculate the embedding representation of students and courses when the student's learning status changes. The similarity alignment module, unique uniform module and recommendation scheme acquisition module are run again to optimize the course recommendation scheme.

[0029] The advantages of the present invention are:

[0030] (1) Through the similarity alignment method, similar sample pairs can be obtained to provide students with more course options, which can help students discover courses that they are potentially good at or interested in, and is conducive to the personalized development of students. Through further alignment, the embedding of students and courses is accurately aligned in terms of cognitive needs, and the learning content of each student is matched with their knowledge mastery and learning progress, thereby ensuring learning quality and improving learning efficiency.

[0031] (2) The unique uniform method is used in combination with the unique strategy to filter out repeated samples in the embedding representation and then calculate the uniformity, which avoids the problem of uniformity loss caused by multiple sample calculations during training. After removing the repeated samples, the uniformity calculation is performed, which can balance the distribution of student embedding and course embedding in the representation space, avoid the tendency of course resources to be centralized in the recommendation, and realize the reasonable allocation of personalized learning resources.

[0032] (3) The present invention can capture students’ mastery status and changing trends in the learning process in real time, and intelligently adjust the recommended content, thus truly achieving teaching in accordance with students’ aptitude and time. BRIEF DESCRIPTION OF THE DRAWINGS

[0033] Figure 1 This is a flow chart of Embodiment 1 of the present invention;

[0034] Figure 2 The student-course interaction diagram in Embodiment 1 of the present invention;

[0035] Figure 3 This is the course recommendation scheme after similarity alignment and uniqueness uniform optimization in implementation 1 of the present invention. DETAILED DESCRIPTION

[0036] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described below in combination with the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0037] Example 1

[0038] Combination Figure 1 , specifically introduce the implementation steps of this embodiment, the present invention provides an education recommendation method based on similarity alignment and uniqueness uniformity, including the following steps:

[0039] S1: Obtain the individual characteristics of students and the knowledge structure and difficulty of the course content, and construct a student-course interaction diagram through interactive behaviors. The specific process is as follows:

[0040] S11: Obtain different individual characteristics of students, including knowledge mastery, learning speed, cognitive level, etc., expressed as feature set F s ,These characteristic information is used to represent the learning status of students;

[0041] S12: Get detailed information about the course. The course information includes content features such as the structure and difficulty of knowledge points, which are represented as feature set F c ,These feature information is used to represent the knowledge requirements of the course;

[0042] S13: Extract the courses that students have interacted with from their individual characteristics. The interactive behaviors are divided into two types: implicit feedback and explicit feedback. Implicit feedback indicates that there is an interactive record between students and courses, and explicit feedback indicates that students have clear preferences for courses, such as liking or giving high ratings.

[0043] S14: Figure 2 As shown in the figure, the student-course interaction graph G (interaction Figure 1 Generally refers to a graph containing interactive behaviors), in this embodiment, G = (N, E), where N = {S∪C} represents the set of all students S and courses C, and E represents the courses that the students have learned (it can also be described as a "student-course" direct term).

[0044] S2: Generate embedded representations of students and courses by aggregating multi-dimensional individual characteristics such as students’ knowledge mastery, learning speed, and cognitive level, as well as the structure and difficulty of the course’s knowledge points;

[0045] In the embedding generation process, the individual characteristics of students F s Content characteristics of the coursec Integrate so that the generated embedding representation contains both the student’s learning characteristics and the course’s knowledge points, so as to reflect the matching degree of learning needs in the subsequent processing;

[0046] The generated embedding representation combines the multi-dimensional characteristics of students and courses to construct an adaptive embedding representation space to measure the potential correlation between student characteristics and the structure of course knowledge points.

[0047] After constructing the embedded representation space, alignment and uniformity, as important indicators for measuring the representation space, can effectively reflect the Euclidean distance between embedded representations and their distribution within the space;

[0048] The calculation formula of the alignment is: Where f(x) and f(y) are the embedded representations of x and y respectively;

[0049] The calculation formula for the uniformity is:

[0050] However, directly optimizing the representation of students and courses through the above two key indicators to improve the model's understanding of student cognition has obvious flaws: first, the interaction data between students and courses is too sparse, which makes it difficult for students to effectively align their potential courses of interest; second, due to the large sample size of student and course data, sample sampling during batch training may cause repeated samples to be calculated multiple times, resulting in a lack of uniformity and further causing serious uniformity deviation.

[0051] To solve the above problems, it is necessary to optimize alignment and uniformity:

[0052] S3: Use the similarity alignment loss function to optimize the alignment. After the student’s embedding representation is aligned with the embedding representation of the learned course, similar courses are found for the student and further aligned. The specific process is as follows:

[0053] S31: Using formula Align students with embedded representations of learned courses;

[0054] S32: Using formula Calculate the cosine similarity between each student or course and all courses or students to obtain the corresponding similarity score, select similar students or courses from them, and generate similar sample pairs, including student-similar courses and course-similar students;

[0055] S33: Using formula Further optimize the alignment of students and course embedding, where P pos are sample pairs of students and courses they have learned, are the student embedding representation and course embedding representation after l2 regularization in the sample pairs of students and the courses they have learned, respectively; P sim_pos is a pair of similar samples captured by cosine similarity, They are respectively the student embedding representation and course embedding representation after l2 regularization in similar sample pairs.

[0056] S4: Use the unique uniform loss function to optimize uniformity and achieve uniform distribution of embedded representations of students and courses in the embedded representation space. The specific process is as follows:

[0057] S41: using the uniqueness strategy, the implementation principle is: E'=Unique(E), which means removing duplicate samples from the sample set E to obtain a new sample set E';

[0058] After applying the uniqueness strategy to the sampled samples, duplicate samples are effectively filtered out. Specifically, for duplicate students or courses generated during the sampling process, a uniqueness strategy is introduced to filter the duplicate samples, thereby ensuring that when calculating the uniformity of the overall representation space of students and courses, the uniformity deviation problem can be effectively avoided;

[0059] S42: Utilize the formula Further optimize the uniformity of student and course embeddings, where f(s) and f(s') are the embedding representations of students and similar students, f(c) and f(c') are the embedding representations of courses and similar courses, Unique represents the uniqueness strategy, and P student and P course They are represented as the student set and the course set respectively.

[0060] After steps S3 and S4, the generated embedding representation can not only accurately reflect students’ cognitive needs but also ensure the uniformity of the embedding space and prevent imbalanced resource distribution.

[0061] S5: Perform multi-task training and optimize similarity alignment-unique uniformity index Loss = L similar_align +αL unique_uniform Ensure that alignment and uniformity are optimized simultaneously, where α is used to adjust the weights between different tasks. Finally, the course recommendation scheme is obtained by calculating the similarity between the student embedding and all course embeddings and sorting the scores in descending order. The example course recommendation scheme is as follows Figure 3 shown.

[0062] S6: If the student’s learning status changes, such as completing a course or improving knowledge mastery, recalculate the embedding representation of the student and the course to reflect the latest learning status, and then execute steps S3-S5 to optimize the course recommendation plan to ensure that the new embedding representation can accurately reflect the current cognitive needs, realign the updated student embedding with the embedding of the relevant courses, and maintain the diversity and balance of the recommended resources, so that the recommended content meets the student’s latest learning needs and preferences.

[0063] This embodiment uses the similarity alignment loss function. In the similarity alignment process, by optimizing the distance between the embeddings of students and courses, the embeddings of students with similar needs and corresponding courses are closer, and the cognitive needs of students can be accurately matched with the required courses, ensuring that the learning needs of students can be better met and the potential of students can be tapped; using the unique uniform loss function, the repeated samples in the embedding representation are filtered out, the distribution of the embedding space of students and courses is balanced, the centralization tendency of course resources in the recommendation is avoided, and the reasonable allocation of personalized learning resources is achieved; and by dynamically updating the embedding representation of students and the relationship between courses, the recommended content is continuously optimized to adapt to the changes in students' learning progress, ensuring the timeliness and accuracy of personalized recommendations.

[0064] Example 2

[0065] Corresponding to the above-mentioned embodiment 1, this embodiment 2 further provides an education recommendation system based on similarity alignment and uniqueness uniformity, including the following modules:

[0066] Mapping module: used to obtain individual characteristics of students and the knowledge structure and difficulty of course content, and to construct a student-course interaction map through interactive behaviors. Specifically, it includes the following units:

[0067] Obtain student characteristic information unit: used to obtain different individual characteristics of students, including knowledge mastery, learning speed, cognitive level, etc., expressed as a characteristic set F s ,These characteristic information is used to represent the learning status of students;

[0068] Get course feature information unit: used to obtain detailed information about the course. Course information includes content features such as the structure and difficulty of knowledge points, expressed as a feature set F c ,These feature information is used to represent the knowledge requirements of the course;

[0069] Finding interactive behavior units: This is used to extract the courses that students have interacted with from their individual characteristics. Interactive behaviors are divided into two types: implicit feedback and explicit feedback. Implicit feedback indicates that there is an interactive record between students and courses, while explicit feedback indicates that students have a clear preference for courses, such as liking or giving high ratings.

[0070] Generate interaction graph unit: used to generate student-course interaction graph G (interaction Figure 1 Generally refers to a graph containing interactive behaviors). In this embodiment, G = (N, E), where N = {S∪C} represents the set of all students S and courses C, and E represents the courses that the student has learned (it can also be described as a "student-course" positive term).

[0071] Generate embedded representation module: used to aggregate the individual characteristics of each student and the knowledge point structure and difficulty of the course content, generate embedded representation, and construct an embedding space of the embedded representation, which includes student information and course information;

[0072] Similarity alignment module: It is used to use the similarity alignment loss function to find similar courses for students and further align them after the embedding representation of the students is aligned with the embedding representation of the learned courses. It specifically includes the following units:

[0073] Preliminary alignment unit: used to utilize the formula Align students with embedded representations of learned courses;

[0074] Find similar sample pairs: Use the formula Calculate the cosine similarity between each student or course and all courses or students to obtain the corresponding similarity score, select similar students or courses from them, and generate similar sample pairs, including student-similar courses and course-similar students;

[0075] Optimize alignment unit: used to utilize formula Further optimize the alignment of students and course embedding, where P pos are sample pairs of students and courses they have learned, are the student embedding representation and course embedding representation after l2 regularization in the sample pairs of students and the courses they have learned, respectively; P sim_pos is a pair of similar samples captured by cosine similarity, They are respectively the student embedding representation and course embedding representation after l2 regularization in similar sample pairs.

[0076] Unique uniform module: It is used to use the unique uniform loss function to achieve uniform distribution of embedded representations of students and courses in the embedded representation space. It specifically includes the following units:

[0077] Removing duplicate sample units: used to utilize the uniqueness strategy, the implementation principle is: E'=Unique(E), which means removing duplicate samples from the sample set E to obtain a new sample set E';

[0078] Optimized uniformity unit: used to utilize the formula Further optimize the uniformity of student and course embeddings, where f(s) and f(s') are the embedding representations of students and similar students, f(c) and f(c') are the embedding representations of courses and similar courses, Unique represents the uniqueness strategy, and P student and P course They are represented as the student set and the course set respectively.

[0079] Get recommendation solution module: used for multi-task training, based on similar alignment-unique uniform index Loss = L similar_align +αL unique_uniform Ensure that alignment and uniformity are optimized simultaneously, where α is used to adjust the weights between different tasks. Finally, the course recommendation solution is obtained by calculating the similarity between the student embedding and all course embeddings and arranging the scores in descending order.

[0080] Optimize recommendation scheme module: It is used to recalculate the embedding representation of students and courses to reflect the latest learning status when the student's learning status changes, such as completing a course or improving knowledge mastery. The similarity alignment module, unique uniform module and acquisition recommendation scheme module are run again to optimize the course recommendation scheme.

[0081] This embodiment first obtains the individual characteristics of students and the knowledge point structure and difficulty of course content through the mapping module, and constructs a student-course interaction map through interactive behavior; then generates the embedded representation of each student and course through the embedding representation generation module to construct the embedding representation space; then uses the similarity alignment module to find similar courses for students and further align them after the embedded representation of the student is aligned with the embedded representation of the learned course; also uses the unique uniform module to use the unique uniform loss function to achieve uniform distribution of the embedded representation of students and courses in the embedding representation space; uses the acquisition recommendation plan module to perform multi-task training, optimizes the similarity alignment-unique uniform index to optimize the embedded representation and obtain the course recommendation plan; if the student's learning status changes, the optimization recommendation plan module is started to recalculate the embedded representation of the student and the course, and the similarity alignment module, the unique uniform module and the acquisition recommendation plan module are run again to optimize the course recommendation plan.

[0082] This embodiment uses a similar alignment module and a unique uniform module to achieve precise alignment of student embedding and course embedding, improve the matching degree between students and courses, ensure the quality of students' learning, and ensure the diversity and balance of recommended resources, so that the recommended content meets the students' learning needs and preferences; this embodiment also uses an optimized recommendation plan module to dynamically optimize the course recommendation plan to achieve teaching according to students' aptitude and time.

[0083] The above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that the technical solutions described in the aforementioned embodiments may still be modified, or some of the technical features may be replaced by equivalents. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. An education recommendation method based on similarity alignment and uniqueness uniformity, characterized in that: The following steps are involved: S1: Obtain the individual characteristics of students and the knowledge structure and difficulty of the course content, and construct a student-course interaction diagram through interactive behaviors; S2: Based on the individual characteristics of each student and the knowledge structure and difficulty of the course content, the embedding representation of each student and course is generated by aggregating multi-dimensional information to construct an embedding representation space; S3: Using the similarity alignment loss function, after the student’s embedding representation is aligned with the embedding representation of the learned course, similar courses are found for the student and further aligned; S4: Use the unique uniform loss function to achieve uniform distribution of embedding representations of students and courses in the embedding representation space; S5: Perform multi-task training, optimize the similarity alignment-unique uniformity index, and obtain the course recommendation scheme through similarity calculation; S6: If the student’s learning status changes, recalculate the embedded representation of the student and the course, and execute S3-S5 again to optimize the course recommendation plan.

2. The educational recommendation method based on similarity alignment and uniqueness uniformity according to claim 1, characterized in that: In step S1, the individual characteristics of the students include knowledge mastery, learning speed and cognitive level; the interactive behavior includes implicit feedback and explicit feedback, wherein implicit feedback indicates that there is an interactive record between the student and the course, and explicit feedback indicates that the student has a clear preference behavior for the course, including likes and high ratings.

3. The education recommendation method based on similarity alignment and uniqueness uniformity according to claim 1, characterized in that: In step S1, the student-course interaction graph is represented as G = (N, E), where N = {S∪C} represents the set of all students S and courses C, and E represents the courses that the students have learned.

4. The educational recommendation method based on similarity alignment and uniqueness uniformity according to claim 1, characterized in that: In step S3, the process of capturing similar sample pairs by cosine similarity is as follows: Calculate the cosine similarity between each student or course and all courses or students to obtain the corresponding similarity score, select similar students or courses from them, and generate similar sample pairs, including student-similar courses and course-similar students.

5. The educational recommendation method based on similarity alignment and uniqueness uniformity according to claim 4, characterized in that: In step S3, the similarity alignment loss function is Where P pos are sample pairs of students and courses they have learned, are the student embedding representation and course embedding representation after l2 regularization in the sample pairs of students and the courses they have learned; P sim_pos is a pair of similar samples captured by cosine similarity, They are respectively the student embedding representation and course embedding representation after l2 regularization in similar sample pairs.

6. The educational recommendation method based on similarity alignment and uniqueness uniformity according to claim 1, characterized in that: In step S4, the unique uniform loss function is: Where f(s) and f(s') are the embedding representations of students and similar students, respectively; f(c) and f(c') are the embedding representations of courses and similar courses, respectively; Unique represents the uniqueness strategy; P student and P course They are represented as the student set and the course set respectively.

7. The educational recommendation method based on similarity alignment and uniqueness uniformity according to claim 6, characterized in that: The uniqueness strategy is: E'=Unique(E), which means that duplicate samples are removed from the sample set E to obtain a new sample set E'.

8. The educational recommendation method based on similarity alignment and uniqueness uniformity according to claim 1, characterized in that: The similarity alignment-unique uniformity index in step S5 is: Loss = L similar_align +αL unique_uniform , where α is used to adjust the weights between different tasks.

9. The educational recommendation method based on similarity alignment and uniqueness uniformity according to claim 1, characterized in that: In step S6, the student's learning status changes, including the student completing a course or the student's knowledge mastery improving.

10. An educational recommendation system based on similarity alignment and uniqueness uniformity, characterized in that: Includes the following modules: Mapping module: used to obtain students’ individual characteristics and the knowledge structure and difficulty of course content, and to construct a student-course interaction map through interactive behaviors; Generate Embedding Representation Module: This module is used to generate the embedding representation of each student and course by aggregating multi-dimensional information based on the individual characteristics of each student and the knowledge structure and difficulty of the course content, and construct the embedding representation space; Similarity alignment module: It is used to find similar courses for students and further align them after the embedding representation of the students is aligned with the embedding representation of the learned courses using the similarity alignment loss function; Unique uniform module: used to achieve uniform distribution of embedding representations of students and courses in the embedding representation space using the unique uniform loss function; Obtaining recommendation scheme module: used for multi-task training, optimizing similarity alignment-unique uniformity index and obtaining course recommendation schemes through similarity calculation; Optimize recommendation scheme module: It is used to recalculate the embedding representation of students and courses when the student's learning status changes. The similarity alignment module, unique uniform module and recommendation scheme acquisition module are run again to optimize the course recommendation scheme.

Images