A method for detecting social relationships

Abstract

This disclosure reveals a method for detecting social relationships, including: counting the number of times all students have been seatmates and constructing a student feature matrix; calculating the intimacy of social relationships among all students based on the number of times they have been seatmates and constructing a student social relationship matrix; adding labels to some students to obtain a label set for some students; updating the student feature matrix using the label set for some students to obtain an updated student feature matrix; training a social relationship detection model using the updated student feature matrix and the student social relationship matrix; predicting labels for the remaining students using the trained social relationship detection model and outputting the labels for the remaining students; and classifying the labels of all students to complete the social relationship detection.

Description

Technical Field

[0001] This disclosure belongs to the field of social relationship detection technology, specifically relating to a social relationship detection method. Background Technology

[0002] Classroom social relationship detection technology identifies potential social groups among students based on the strength of their interpersonal relationships, grouping students with strong relationships into a single social group. Existing graph convolutional network (GCN)-based social relationship detection methods learn node features from social network topology and node attributes using GCNs, then use these learned features for clustering to ultimately detect social relationships. However, in a classroom setting, applying these methods requires collecting a large amount of student information to construct a supervised learning model, which reduces the efficiency of social relationship detection. Furthermore, these methods do not fully utilize the strength of interpersonal relationships among students, resulting in low detection accuracy. Summary of the Invention

[0003] To address the shortcomings of existing technologies, the purpose of this disclosure is to provide a social relationship detection method that effectively improves the accuracy of classroom social relationship detection by incorporating the degree of intimacy of students' social relationships.

[0004] To achieve the above objectives, this disclosure provides the following technical solutions:

[0005] A method for detecting social relationships includes the following steps:

[0006] S100: Count the number of times all students are deskmates and construct a student feature matrix;

[0007] S200: Calculate the degree of intimacy of social relationships among all students based on the number of times each student sits next to the other, and construct a student social relationship matrix;

[0008] S300: Add tags to some students to obtain a tag set for some students;

[0009] S400: Update the student feature matrix using the label set of the aforementioned students to obtain the updated student feature matrix;

[0010] S500: The social relationship detection model is trained using the updated student feature matrix and student social relationship matrix;

[0011] S600: Use the trained social relationship detection model to predict the labels of the remaining students and output the labels of the remaining students;

[0012] S700: Categorize all students' tags, thus completing the social relationship detection.

[0013] Preferably, in step S100, the number of times all students are seatmates is calculated using the following formula:

[0014]

[0015] Where k represents the number of classes, and i and j represent student i and student j, respectively. This represents the number of times student i and student j are seatmates in the k-th lesson. This represents the number of times student i and student j are seatmates in the (k-1)th lesson. This indicates the number of times student i and student j were seatmates in the first lesson.

[0016] Preferably, in step S100, the student feature matrix is ​​represented as follows:

[0017]

[0018] Here, the feature matrix H represents all students, and its size is n×f, where n represents the number of students, f represents the dimension of the student's vector representation, and H... 11 H represents the first element in the vector representation of the first student. 1f H represents the f-th element in the vector representation of the first student. n1 H represents the first element in the vector representation of the nth student. nf The vector representation of the nth student is the f-th element.

[0019] Preferably, in step S200, the degree of intimacy of social relationships among all students is calculated using the following formula:

[0020]

[0021] Where t represents the total number of times this course is taken in the semester, and m represents the coefficient for judging the degree of intimacy. The smaller the coefficient, the stronger the ability to eliminate random factors. This represents the number of times student i and student j sat together after a course ends in a semester. ij Let represent the degree of intimacy between student i and student j, and tanh(·) represent the hyperbolic tangent function.

[0022] Preferably, in step S200, the student social relationship matrix is ​​represented as follows:

[0023]

[0024] Among them, l 11 This indicates the closeness of student 1's social relationship with themselves; nn This indicates the closeness of student n's social relationship with themselves; l1n This indicates the degree of intimacy in the social relationship between student 1 and student n; l n1 This indicates the degree of intimacy in the social relationship between student n and student 1.

[0025] Preferably, in step S400, the updated student feature matrix is ​​represented as follows:

[0026] Stu=[Stu1,Stu2,Stu3,…,Stu n ] T

[0027] For any student s, Stu s Obtained from the following formula:

[0028]

[0029] Among them, H s The vector representation of student s without labels; l a This indicates the added tag. The student's tag is l a The vector representation after that.

[0030] Preferably, in step S500, the social relationship detection model includes: an input layer, a graph convolutional layer 1, a graph convolutional layer 2, and an output layer.

[0031] Preferably, in step S500, the training process of the social relationship detection model includes the following steps:

[0032] Step 1: Use the updated student feature matrix and student social relationship matrix to perform forward propagation on the social relationship detection model to obtain the model's label prediction results for the labeled students;

[0033] Step 2: Calculate the loss using the label prediction results and the added labels, and then use the calculated loss to backpropagate the model.

[0034] Step 3: Repeat steps 1 and 2. When the social relationship detection model converges, the training is complete; if it does not converge, increase the number of training iterations until it converges.

[0035] Compared with the prior art, the beneficial effects of this disclosure are as follows:

[0036] 1. In the method described in this disclosure, student tags are automatically generated without the need to collect students' personal information in advance, which effectively improves the efficiency of social relationship detection in the classroom;

[0037] 2. By incorporating the strength of students' social relationships into the method described in this disclosure, the method becomes more realistic and effectively improves the accuracy of classroom social relationship assessment. Attached Figure Description

[0038] Figure 1 This is a flowchart of a social relationship detection method provided in one embodiment of this disclosure;

[0039] Figure 2 This is a schematic diagram of the structure of a social relationship detection model provided in another embodiment of this disclosure;

[0040] Figure 3 This is a student-social relationship intimacy graph with added tags, provided in another embodiment of this disclosure;

[0041] Figure 4 This is a graph showing the intimacy of social relationships among all students, provided in another embodiment of this disclosure. Detailed Implementation

[0042] The following will refer to the appendix. Figures 1 to 4 Specific embodiments of this disclosure are described in detail. While specific embodiments of this disclosure are shown in the accompanying drawings, it should be understood that this disclosure may be implemented in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided so that this disclosure will be thorough and complete, and will fully convey the scope of this disclosure to those skilled in the art.

[0043] It should be noted that certain terms are used in the specification and claims to refer to specific components. Those skilled in the art will understand that different terms may be used to refer to the same component. This specification and claims do not distinguish components based on differences in terminology, but rather on differences in function. The terms "comprising" or "including" used throughout the specification and claims are open-ended and should be interpreted as "comprising but not limited to." The following descriptions are preferred embodiments for carrying out this disclosure; however, these descriptions are for the purpose of understanding the general principles of the specification and are not intended to limit the scope of this disclosure. The scope of protection of this disclosure is determined by the appended claims.

[0044] To facilitate understanding of the embodiments of this disclosure, further explanations and descriptions will be provided below with reference to the accompanying drawings and specific embodiments. The accompanying drawings do not constitute a limitation on the embodiments of this disclosure.

[0045] In one embodiment, such as Figure 1 As shown, this disclosure provides a method for detecting social relationships, including the following steps:

[0046] S100: Count the number of times each student sits next to another student, and construct a student feature matrix based on the number of times.

[0047] In this step, the number of times each student sits next to another student is calculated using the following formula:

[0048]

[0049] Where k represents the number of classes, and i and j represent student i and student j, respectively. This represents the number of times student i and student j are seatmates in the k-th lesson. This represents the number of times student i and student j are seatmates in the (k-1)th lesson. This indicates the number of times student i and student j were seatmates in the first lesson.

[0050] For example, after the second lesson, student A and student B sat together once, that is... In the third lesson, student A and student B were deskmates. After the third lesson,

[0051] After obtaining the number of times each student sat next to another student, to facilitate training the social relationship detection model in subsequent steps, a student is described by a vector of length f. This vector is the student's vector representation, which is obtained by the following formula:

[0052] H p =[H p1 H p2 H p3 H pf (2)

[0053] Among them, H p Let p be the vector representation of student p.

[0054] The feature matrix composed of the vector representations of all students is represented as follows:

[0055]

[0056] Here, the feature matrix H represents all students, and its size is n×f, where n represents the number of students, f represents the dimension of the student's vector representation, and H... 11 H represents the first element in the vector representation of the first student. 1f H represents the f-th element in the vector representation of the first student. n1 H represents the first element in the vector representation of the nth student. nf The vector representation of the nth student is the f-th element.

[0057] Each element in the feature matrix H is obtained by the following formula:

[0058]

[0059] Among them, H pq Let H be the element in the p-th row and q-th column of the matrix. For example, the vector representation of student 1 is H1 = [1, 0, 0, 0, ..., 0], with length f, the first element being 1, and the rest being 0; the vector representation of student 15 is H... 15 = [0, 0, ..., 0, 1, 0, ..., 0], with a length of f, the 15th element has a value of 1, and the rest of the elements have a value of 0.

[0060] S200: Calculate the social intimacy among all students based on the number of times each student sits next to the other, and construct a student social relationship matrix based on the social intimacy.

[0061] In this step, the level of social intimacy among all students is calculated using the following formula:

[0062]

[0063] Where t represents the total number of times this course is taken in the semester, and m represents the coefficient for judging the degree of intimacy. The smaller the coefficient, the stronger the ability to eliminate random factors. minr represents the smallest non-zero value. This represents the number of times student i and student j sat together after a course ends in a semester. ij Let represent the degree of intimacy between student i and student j, and tanh(·) represent the hyperbolic tangent function.

[0064] For example, suppose a course has 15 sessions in a semester, and the only students participating in this course are student e, student f, student g, and student h. Student e and student f are seatmates 7 times, student g and student h are seatmates 9 times, student e and student g are seatmates 1 time, student e and student h are seatmates 0 times, student f and student g are seatmates 2 times, and student f and student h are seatmates 0 times. By definition of minr, if minr = 1, then m = 1.

[0065] For example, suppose there are 12 classes in a semester. If m = 0.2 and student a and student b sit together 6 times, then... l a，b =0.1803, when student a and student b sit together no more than 5 times, l a，b =0, meaning that if two students sit together no more than 5 times, they will not be considered to have a social relationship; when m = 0.5 and student a and student b sit together 6 times, then... l a，b =0.8000, when student a and student b sit together no more than twice, la，b =0, meaning that if two students sit together no more than twice, they will not be considered to have a social relationship.

[0066] For example, suppose there are 12 classes in a semester, and student A and student B sit together 8 times. Let m = 1, then the calculated l a，b =0.9692; If student a and student b sit together 9 times, then Then the calculated l a，b =0.9756; If student a and student b sit together 10 times, then Then the calculated l a，b =0.9801.

[0067] It is evident that the number of times student a and student b sit together is directly proportional to the degree of intimacy between them.

[0068] The student social relationship matrix X0, constructed based on the degree of social intimacy, is represented as follows:

[0069]

[0070] Among them, l 11 This indicates the closeness of student 1's social relationship with themselves; nn This indicates the closeness of student n's social relationship with themselves; l 1n This indicates the degree of intimacy in the social relationship between student 1 and student n; l n1 This indicates the degree of intimacy in the social relationship between student n and student 1.

[0071] S300: Add tags to some students to obtain a tag set for some students;

[0072] In this step, we can construct a student-social relationship intimacy graph by using all students as nodes and the degree of social relationship intimacy among all students as edges; find key nodes in the student-social relationship intimacy graph and add labels to the key nodes to obtain a labeled student-social relationship intimacy graph, thereby obtaining a set of labels for some students corresponding to the key nodes;

[0073] Since the social relationship detection model is based on GCN, training this model requires adding labels to some students. The vector representation of these labeled students is obtained by the following formula:

[0074]

[0075] in, Let la represent the vector representation of student 'a' with the label 'la', where la represents the label and is obtained by the following formula:

[0076]

[0077] Where Avg(a) represents the average level of intimacy in the social relationships of student a, which is obtained by the following formula:

[0078]

[0079] Global represents the average level of intimacy in all students' social relationships, which is obtained by the following formula:

[0080]

[0081] Here, card(·) represents the number of elements in the set.

[0082] Nei1(a) represents the set of students who have a direct social relationship with student a and have been tagged, which is obtained by the following formula:

[0083] Nei1(a)={b|l ab ≠0 and b∈Label} (11)

[0084] Where Label represents the set of all students who have been labeled, l ab This indicates the degree of intimacy in the social relationship between student A and student B.

[0085] Nei2(a) represents the set of students who have an indirect social relationship with student b and have been tagged, which is obtained by the following formula:

[0086] Nei2(a)={c|l cd ≠0, d∈Nei1(a)andc∈Label} (12)

[0087] Among them, l cd This indicates the degree of intimacy in the social relationship between student A and student B.

[0088] S400: Update the student feature matrix using the label set of the aforementioned students to obtain the updated student feature matrix Stu;

[0089] In this step, Stu = [Stu1, Stu2, Stu3, ..., Stu... n ] T This is the updated student feature matrix.

[0090] For any student s, Stu s Obtained from the following formula:

[0091]

[0092] Among them, H sThe vector representation of student s without labels; l a This indicates the added tag. The student's tag is l a The vector representation after that.

[0093] S500: The social relationship detection model is trained using the updated student feature matrix and student social relationship matrix;

[0094] S600: Use the trained social relationship detection model to predict the labels of the remaining students and output the labels of the remaining students;

[0095] In this step, the weight matrix in the trained social relationship detection model is updated to... and The prediction result of using the trained social relationship detection model to predict the labels of the remaining students is obtained by the following formula:

[0096]

[0097] in, The weight matrices after training are n×n and n×g, respectively, where g represents the number of elements in the Label.

[0098] S700: Categorize all students' tags, thus completing the social relationship detection.

[0099] In this step, after obtaining the labels of the remaining students, students with the same label are placed into the same group, indicating that students in the group have closer social relationships than students outside the group. Additionally, students who do not have social relationships with any other students are given the same label and placed into the same group, thus completing the social relationship detection. It is important to note that placing students who do not have social relationships with any other students into the same group does not imply that students in this group have closer social relationships than students outside the group; it only indicates that students in this group have social relationships with all other students.

[0100] The above embodiments constitute the complete technical solution of this disclosure. Social relationships play an important role in students' campus life. Students' social relationships have a significant impact on their academic performance and even personality development. Therefore, understanding students' social relationships through social relationship detection methods has very practical guiding significance. This disclosure improves the efficiency and accuracy of social relationship detection by automatically generating tags and introducing methods such as the strength of social relationship intimacy, and has good application value in student psychological counseling and analysis of abnormal student behavior.

[0101] In another embodiment, such as Figure 2As shown, the social relationship detection model includes:

[0102] Input layer: Input the student's feature matrix Stu and the student's social relationship matrix X0, and output the product of Stu and (X0+I) Layer0 as the input to the graph convolutional layer, where I is an identity matrix of the same size as X0, and Layer0 is obtained by the following formula:

[0103] Layer0 = Stu(X0 + I)

[0104] Convolutional layer 1: Input Layer0, output Layer1 is the product of Layer0 and W1 calculated using the ReLU function. The calculation method for Layer1 is as follows:

[0105] Layer1 = ReLU(Layer0W1)

[0106] W1 is a trainable matrix.

[0107] Graph Convolutional Layer 2: After obtaining the output layer 1 (Layer 1), the output layer 2 is the product of layer 1, W2, and (X0+I) calculated using ReLU. Layer 2 is obtained from the following formula:

[0108] Layer2 = ReLU((X0+I)Layer1W2)

[0109] W2 is a trainable matrix.

[0110] Output layer: After obtaining the output of Layer 2 of the graph convolutional layer, the student's label prediction result X. pred We obtain it from the following formula:

[0111] X pred =Softmax(Layer2)

[0112] Here, softmax() represents the sofftmax function.

[0113] In another embodiment, the training process of the social relationship detection model includes the following steps:

[0114] Step 1: Perform forward propagation on the social relationship detection model using the updated student feature matrix and student social relationship matrix to obtain the model's predicted label X for the labeled students. pred ;

[0115] In this step, the label set of the aforementioned students is used for forward propagation of the social relationship detection model using the following formula:

[0116] X GCN=ReLU(ReLU(Stu(X0+I)W1)(X0+I)W2) (15)

[0117] Among them, X GCN X0 represents the input student social relationship matrix, I represents the n×n identity matrix, n represents the number of students, and W1 and W2 represent trainable weight matrices with sizes of n×n and n×g, respectively, where g represents the number of elements in the Label.

[0118] Step 2: Calculate the loss using the model's prediction results of the added labels for the students and the labels added in step S300, and use the calculated loss to backpropagate the model.

[0119] In this step, the output student social relationship matrix X GCN The predicted results X for all students obtained after applying the Softmax function pred From X pred Find the labeled students in the set and calculate the loss function Loss using the labels of these students and the labels of the students in the set Label. Loss is obtained by the following formula:

[0120]

[0121] Where, N all This represents the number of students who added the label, where 's' represents the number of students who added the label, and 'Loss' represents the number of students who added the label. s This represents the loss of the students who added the label, and Loss s Obtained from the following formula:

[0122]

[0123] Where tr represents the student's actual label in the Label, L w Y represents the number of labels added to the Label. s，tr Indicate whether the label for student s is tr, P s，tr X represents pred The probability that the label of student s is predicted to be tr.

[0124] Step 3: Repeat steps 1 and 2. When the social relationship detection model converges, the training is complete; if it does not converge, increase the number of training iterations until it converges.

[0125] The present disclosure will now provide a detailed description of the above solution in conjunction with specific embodiments.

[0126] Assuming a classroom has 9 rows and 8 columns of seats and a total of 39 students, taking 10 attendance records as an example, the social relationship detection among the 39 students consists of the following steps:

[0127] 1. Count the number of times all students are deskmates according to formula (1), and construct a student feature matrix of size 39×39 based on the statistical results. A partial example of this feature matrix is ​​as follows:

[0128]

[0129] 2. Calculate the social intimacy among students and construct a 39×39 student social relationship matrix based on the intimacy of student social relationships. A partial example of this matrix is ​​shown below:

[0130]

[0131] In this matrix, 0.561 represents the degree of intimacy of the social relationship between student 2 and student 1, and 0.981 represents the degree of intimacy of the social relationship between student 2 and student 39, which is 0.981.

[0132] 3. Following the method described in S300, we obtained 11 qualified students. Using these 39 students as nodes and the degree of intimacy of their social relationships as edges, we added labels to these students and constructed a system as follows: Figure 3 The student-social relationship intimacy graph shown constitutes the label set. The 11 students are student 3, student 7, student 10, student 2, student 19, student 24, student 34, student 37, student 29, student 25, student 28, and student 4. The added labels are 3, 7, 10, 2, 19, 24, 34, 37, 29, 25, 28, and 4, respectively. The student feature matrix is ​​updated using these labels, and the updated feature matrix is ​​as follows:

[0133]

[0134] In this matrix, 3 indicates that student 3 has added the label 3.

[0135] 4. Optimize the model using the updated student feature matrix Stu, student social relationship matrix X0, and Adam optimizer. Set the number of training iterations to 300, the learning rate to 0.001, and the trainable matrices W1 and W2 in the model to 39×39 and 39×12, respectively.

[0136] 5. Use the model trained in step 4 to predict the labels of other students. After the prediction is complete, X pred= [2, 3, 2, 3, 4, 10, 7, 7, 8, 7, 10, 10, 7, 2, 7, 12, 7, 2, 6, 19, 25, 28, 3, 25, 24, 25, 24, 28, 28, 29, 29, 6, 19, 34, 34, 37, 9, 37, 12], the corresponding label prediction results for students 1 to 39 are 2, 3, 2, 3, 4, 10, 7, 7, 8, 7, 10, 10, 7, 2, 7, 12, 7, 2, 6, 19, 25, 28, 3, 25, 24, 25, 24, 28, 28, 29, 29, 6, 19, 34, 34, 37, 9, 37, 12. Students 5, 9, 16, 19, 37, 23, 32, and 39 are students who have no social relationships with any other students. These students are assigned the same label 4. The predicted final labels for students 1 through 39 are 2, 3, 2, 3, 4, 10, 7, 7, 4, 7, 10, 10, 7, 2, 7, 4, 7, 2, 4, 19, 25, 28, 4, 25, 24, 25, 24, 28, 28, 29, 29, 4, 19, 34, 34, 37, 4, 37, 4. The student-social relationship intimacy graph is shown below. Figure 4 As shown, students with the same label are placed into the same group, forming the following groups:

[0137] Students in Group 1: Student 3, Student 18, Student 1, Student 14;

[0138] Group 2 students: Student 4, Student 2;

[0139] Group 3 students: Student 11, Student 12, Student 6;

[0140] Students in group 4: Student 8, Student 7, Student 10, Student 13, Student 15, Student 17;

[0141] Group 5 students: Student 20, Student 33;

[0142] Group 6 students: Student 38, Student 36;

[0143] Group 7 students: Student 26, Student 21, Student 24;

[0144] Group 8 students: Student 29, Student 28, Student 22;

[0145] Group 9 students: Student 25, Student 27;

[0146] Group 10 students: Student 30, Student 31;

[0147] Group 11 students: Student 35, Student 34;

[0148] Group 12 students: Student 5, Student 9, Student 16, Student 19, Student 37

[0149] Student 23, Student 32, Student 39;

[0150] The above indicates that students within a group have closer social relationships than students outside the group, thus completing the social relationship test. It should be noted that the students in group 12 are all students who do not have any social relationships with any other students; this does not necessarily mean that these students have closer social relationships than students outside the group.

[0151] 6. The detection results of this method are compared and explained below.

[0152] This disclosure uses an Intel i5-7500 CPU as the computing platform. Ten attendance records from 39 individuals in an attendance system were selected as the dataset for testing and compared with traditional algorithms and deep learning-based algorithms. We use the NMI metric to evaluate the results of this invention; the NMI calculation formula is as follows:

[0153]

[0154] Suppose that a certain algorithm clusters the samples as X = [X1, X2, ..., X...]. n The standard clustering result is Y = [Y1, Y2, ..., Y]. n ], then the joint probability distribution probability distribution function

[0155] Traditional methods were compared using the Louvain method, Girvain-Newman method, and AP method, with NMI used as the evaluation metric. The accuracy of the Louvain method was 0.7819, the Girvain-Newman method was 0.6840, and the AP method was 0.7754. Among deep learning-based algorithms, the SGCN method was compared, with NMI used as the evaluation metric. The SGCN method achieved an accuracy of 0.8498. Using NMI as the evaluation metric, the accuracy of our proposed method was 0.9351.

[0156] The results show that, under the same dataset, Louvain performs best among traditional methods, while SGCN improves detection performance by 8.68% compared to Louvain among deep learning-based methods, and our method improves detection accuracy by 10.04% compared to SGCN.

[0157] The foregoing general description of the invention and its specific embodiments should not be construed as a limitation on the technical solution of the invention. Those skilled in the art, based on the disclosure of this application, can add, reduce, or combine the disclosed technical features in the foregoing general description and / or specific embodiments (including examples) without departing from the constituent elements of the invention, to form other technical solutions within the scope of protection of this application.

Claims

1. A method for detecting social relationships, comprising the following steps: S100: Count the number of times all students are deskmates and construct a student feature matrix; S200: Calculate the degree of intimacy of social relationships among all students based on the number of times each student sits next to the other, and construct a student social relationship matrix; S300: Add tags to some students to obtain a tag set for some students; S400: Update the student feature matrix using the label set of the aforementioned students to obtain the updated student feature matrix; S500: The social relationship detection model is trained using the updated student feature matrix and student social relationship matrix; S600: Use the trained social relationship detection model to predict the labels of the remaining students and output the labels of the remaining students; S700: Categorize all students' tags, thus completing the social relationship detection; in, In step S100, the number of times each student is seated with another student is calculated using the following formula: ， Where k represents the number of classes, and i and j represent student i and student j, respectively. This represents the number of times student i and student j are seatmates in the k-th lesson. Representing the The number of times student i and student j were seatmates in this course. This indicates the number of times student i and student j were seatmates in the first lesson; In step S200, the degree of intimacy of social relationships among all students is calculated using the following formula: ， Where t represents the total number of times this course is taken during the semester. The coefficient represents the degree of intimacy; the smaller the coefficient, the stronger the ability to eliminate random factors. This refers to students after the end of a certain course in a certain semester. With students The number of times we sat together This indicates the degree of intimacy between student i and student j. Represents the hyperbolic tangent function; In step S400, the updated student feature matrix is ​​represented as follows: ， For any student , Obtained from the following formula: ， Among them, H s The vector representation of student s without labels; l a This indicates the added tag. The student's tag is l a The vector representation after that.

2. The method according to claim 1, wherein, In step S100, the student feature matrix is ​​represented as follows: ， Among them, the feature matrix Represents all students, with a size of , Let f represent the number of students, and let f represent the dimension of the student's vector representation. This represents the first element in the vector representation of the first student. This represents the f-th element in the vector representation of the first student. Indicates the first The first element in the vector representation of each student. Indicates the first The f-th element in the vector representation of each student.

3. The method according to claim 1, wherein, In step S200, the student social relationship matrix is ​​represented as follows: ， Among them, l 11 This indicates the closeness of student 1's social relationship with themselves; nn This indicates the closeness of student n's social relationship with themselves; l 1n This indicates the degree of intimacy in the social relationship between student 1 and student n; l n1 This indicates the degree of intimacy in the social relationship between student n and student 1.

4. The method according to claim 1, wherein, In step S500, the social relationship detection model includes: an input layer, a graph convolutional layer 1, a graph convolutional layer 2, and an output layer.

5. The method according to claim 1, wherein, In step S500, the training process of the social relationship detection model includes the following steps: Step 1: Use the updated student feature matrix and student social relationship matrix to perform forward propagation on the social relationship detection model to obtain the model's label prediction results for the labeled students; Step 2: Calculate the loss using the label prediction results and the added labels, and then use the calculated loss to backpropagate the model. Step 3: Repeat steps 1 and 2. When the social relationship detection model converges, the training is complete; if it does not converge, increase the number of training iterations until it converges.

Images