Intelligent course recommendation method based on digital human interaction

Abstract

The invention discloses an intelligent course recommendation method based on digital human interaction. The method comprises a course scoring analysis model building module, a course matching analysis model building module and a course recommendation degree calculation and recommendation module. The method comprises the following steps: firstly, constructing a course scoring analysis model through an analytic hierarchy process, and establishing a student-course scoring matrix; secondly, comprehensively utilizing the student-curriculum scoring matrix and the student similarity calculated by the student feature information, calculating the fitness of the curriculums to the students, and generating a candidate curriculum list according to the size of the fitness; then, the matching degree between the courses is calculated according to the candidate course list and the selected courses extracted from the course database; and finally, weighting the matching degree between the courses and the fitness of the courses to the students to obtain comprehensive course recommendation degrees, and generating a final recommendation course sequence according to the sorting of the course recommendation degrees. The method meets the course selection requirements of the students, so that the course selection experience of the students is improved.

Description

Technical Field

[0001] The present invention relates to the fields of recommendation and artificial intelligence, and in particular to a method for intelligently recommending campus courses based on digital human interaction. Background Art

[0002] With the popularity of online education, students are faced with more and more course choices. How to quickly and effectively choose the right courses has become the focus of attention for students and educational institutions. Existing course recommendation systems mostly rely on simple keyword matching or rule-based recommendations, lacking personalization and intelligence, and cannot fully reflect students' course selection preferences. Therefore, the accuracy and success rate of recommendations are low. In contrast, digital human interaction technology, which has not yet been widely used, has the potential to improve the accuracy and personalization of recommendations. These technologies can better understand students' needs by identifying their interests and hobbies, and make intelligent recommendations based on real-time rating information, thereby reducing the duplication rate of recommendations and improving the success rate of recommendations. Summary of the invention

[0003] The purpose of the present invention is to address the defects and shortcomings of the above-mentioned prior art and provide an intelligent recommendation method based on digital human interactive recommendation, which realizes interactive course recommendation for different students and improves the accuracy of recommendation.

[0004] The technical solution adopted by the present invention to solve the technical problem is: an intelligent course recommendation method based on digital human interaction, comprising:

[0005] Extract students' ratings of courses from the rating database and build a course rating analysis model to calculate the suitability P of the course for the target students;

[0006] Obtain relevant information about the course from the course database and build a course matching analysis model to calculate the matching degree Q between the candidate courses and the selected courses of the target students;

[0007] The suitability P of candidate courses for target students and the matching degree Q between courses are weighted respectively to obtain the course recommendation degree R, and the candidate courses are ranked and recommended according to the course recommendation degree R.

[0008] Method flow:

[0009] Step 1: Establish a student-course rating matrix as a basis for calculating the suitability of courses for students;

[0010] Step 2: Communicate with students through digital humans, collect and extract their characteristic information and transmit it to the background;

[0011] Step 3: Calculate the similarity W between the student and other students based on the feature information;

[0012] Step 4: Based on the similarity calculated in step 3, combine the student-course rating matrix and calculate the suitability P of the course for the student.

[0013] Furthermore, the specific steps of establishing the student-course rating matrix in step 1 above include:

[0014] Step 1-1: Divide the course evaluation index system into the target layer, the criterion layer and the scheme layer, where the target layer is k, the criterion layer includes the teacher's teaching quality, the course practicality, the course assessment difficulty and the course attractiveness, which are recorded as S1, S2, S3, S4 respectively, and the scheme layer is the various courses entered in the database;

[0015] Step 1-2: Integrate and normalize the data of each course at the program level, perform z-SCORE standardization on the course-indicator matrix, and obtain a normalized table of course-indicator data;

[0016] Step 1-3: Solve the indicator weights of the criterion layer, compare the importance of related indicators on the criterion layer, and use the 1-9 scaling method to obtain the judgment matrix;

[0017] Step 1-4: Quantify and pass the consistency test to obtain the weight w of each indicator i ;

[0018] Step 1-5: Calculate the comprehensive score of the course. According to the weighted and normalized dish index data matrix, the hierarchical analysis model is used to calculate the comprehensive score of the course. The calculation formula is:

[0019]

[0020] Among them, C is the course name, C Ai is the indicator data corresponding to the course, w i is the weight of the indicator, S C Give a comprehensive rating to a course;

[0021] Step 1-6: Use the comprehensive score of the course as the student's score for the course and establish a student-course rating matrix.

[0022] Furthermore, the specific steps of the above step 2 of digital human communicating with students and extracting student feature information include:

[0023] Step 2-1: After removing the names of the conversation users and other information, connect all the conversation contents with spaces;

[0024] Step 2-2: A complete text sequence is formed for each student, and this is used as the original text description representing each student's interests;

[0025] Step 2-3: Transmit student interest feature information to the backend server.

[0026] Furthermore, the specific steps of calculating the similarity W between the student and other students in step 3 above include:

[0027] Step 3-1: Use the word vector model to embed the target student u tar The characteristic information and student u i The characteristic information is mapped into information vector I tar and I i ,

[0028] Step 3-2: Calculate the target student u using cosine similarity calculation tar The information vector I tar With students i The information vector I i The similarity between tar ,u i ), the calculation formula is:

[0029]

[0030] Furthermore, the specific steps of calculating the suitability P of the course to the student in step 4 above include:

[0031] Step 4-1: Sort all students according to the similarity W between the student and other students calculated in claim 5 and select the student with the target student u tar The top n similar students are taken as the candidate student table U;

[0032] Step 4-2: Calculate the candidate student u in the candidate student table U j Courses that have been learned but the target students have not learned c i For target students tar The fitness P(u tar , c i ), the calculation formula is:

[0033]

[0034] Among them, N(u tar ,u j ) represents student u tar Uneducated but candidate students j A collection of courses you have taken, represents the candidate student u j For Course C i The rating of W(u tar u j ) is the target student u tar With students j The similarity between

[0035] Step 4-3: Sort the student's fitness P according to the candidate courses and construct a candidate course table T as the target student u. tar Candidate courses.

[0036] Furthermore, the specific steps of constructing the course collocation analysis model of the present invention include:

[0037] Step 1: Get the target student u from the course database tar Information about selected courses;

[0038] Step 2: Students can choose courses c i and course c j The case is regarded as a course pair, and the frequency of occurrence of the course pair is the support (c i , c j ), calculate the support between courses, the calculation formula is:

[0039]

[0040] Among them, |N(c i )∩N(c j )| means to select course c at the same time i and course c j , |N(U)| represents the number of students who have selected courses;

[0041] Step 3: Filter out all candidate course pairs whose support is higher than a certain threshold.

[0042] Furthermore, the specific steps of calculating the collocation degree Q between courses in the present invention include:

[0043] Step 1: Calculate the confidence of each course in the course pair and determine the course c j In the course c i The frequency of occurrence in the course pair of c, that is, in the course c i Select c under the condition j The probability of , and vice versa, is calculated as:

[0044]

[0045] Among them, |N(c i )∩N(c j )| means to select course c at the same time i and course c j The number of students, |N(c i )| means select course c i The number of students, |N(c j)| means select course c j Number of students;

[0046] Step 4: Construct a course-course bipartite graph. If a student also chooses course c i and course c j , then in course c i and course c j Add an edge between them, and the weight of the edge is the confidence calculated in step 3;

[0047] Step 5: Calculate the matching degree Q for each candidate course pair. The calculation formula is:

[0048]

[0049] in, In the tth iteration, course c i and course c j The matching degree between them, α is an adjustment factor, which controls the balance between the new matching degree and the initial matching degree, and its value is in the range of [0,1]. i ) indicates that the course c i A collection of directly connected courses, w ij It is a course c i and course c j The initial edge weight between ;

[0050] Step 6: Construct the course matching table G based on the course support and confidence, and use Q as the matching degree between the two courses in each course pair.

[0051] Furthermore, the steps of calculating the course recommendation degree R and recommending courses in the present invention specifically include:

[0052] Step 1: Find course c in the candidate course table T i and its effect on target students tar The fitness P,

[0053] Step 2: Find the course c that appears in the course matching table G. j and candidate courses c i The matching scheme is obtained to obtain the matching degree Q; if it does not exist, the matching degree is 0, and the course suitability P for students and the matching degree Q between courses are weighted respectively to obtain the course recommendation degree R, which is calculated as follows:

[0054] R(c i )=w1×P+w2×Q

[0055] Among them, R(c i ) represents the target student u tar Study Coursec iw1 and w2 are the weights of the course’s suitability for students P and the matching degree between courses Q, respectively. The values ​​are in the range of [0,1], and the sum of the two is 1.

[0056] Step 3: Sort the candidate courses according to the course recommendation degree R, and select the top k courses to recommend to the target student u tar .

[0057] Beneficial effects: The present invention includes constructing a course scoring analysis model module, constructing a course matching analysis model module and a course recommendation calculation and recommendation module. First, a course scoring analysis model is constructed by the hierarchical analysis method to establish a student-course scoring matrix. At the same time, the student feature information is collected through digital human interaction and extracted using the pre-training model and stored in the background. Secondly, the student-course scoring matrix and the student similarity calculated from the student feature information are comprehensively used to calculate the suitability of the course for the student, and a candidate course list is generated according to the size of the suitability. Then, the matching degree between courses is calculated based on the candidate course list and the selected courses extracted from the course database. Finally, by weighting the matching degree between courses and the suitability of courses for students, a comprehensive course recommendation degree is obtained, and the final recommended course sequence is generated according to the size of the course recommendation degree. This method can provide reasonable, efficient and personalized course recommendations based on differences between students, course suitability and course matching, which meets the students' course selection needs, thereby improving the students' course selection experience. BRIEF DESCRIPTION OF THE DRAWINGS

[0058] Figure 1 A flowchart of an intelligent course recommendation method based on digital human interaction provided by an embodiment of the present invention.

[0059] Figure 2 A schematic diagram of a hierarchical analysis model of a course evaluation index system for an intelligent course recommendation method based on digital human interaction provided in an embodiment of the present invention.

[0060] Figure 3 A structural diagram of an intelligent course recommendation method based on digital human interaction provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0061] This specific embodiment is merely an explanation of the present application and is not a limitation of the present application. After reading this specification, those skilled in the art may make modifications to the present embodiment without any creative contribution as needed, but such modifications are protected by the patent law as long as they are within the scope of the claims of the present application.

[0062] In order to make the purpose, technical solutions and advantages of the embodiments of the present application clearer, the technical solutions in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of this application.

[0063] The embodiments of the present application are further described in detail below in conjunction with the drawings in the specification.

[0064] Embodiment 1

[0065] In one embodiment of the present invention, an intelligent course recommendation method based on digital human interaction is provided. Figure 1 As shown, the main steps are described as follows:

[0066] (I) Construct a course scoring analysis model to calculate the suitability of the course for the target students:

[0067] By reading relevant literature and analyzing specific course data, the present invention divides the course comprehensive scoring index system into three levels: target level, criterion level and scheme level. The target level is the comprehensive score S of a course, and the criterion level includes four indicators: teacher teaching quality, course practicality, course assessment difficulty and course attractiveness, which are represented as S1, S2, S3 and S4.

[0068] Among them, the scoring basis for the teacher teaching quality indicators at the criterion level is: extracting the users' evaluation data on the teacher's teaching quality in the course rating, five-star praise corresponds to 100 points, four stars corresponds to 80 points, three stars corresponds to 60 points, two stars corresponds to 40 points, and one star corresponds to 20 points. The final teacher teaching quality rating of the course is equal to the score corresponding to each star level multiplied by the percentage of the number of people.

[0069] The scoring methods for course practicality, course assessment difficulty and course attractiveness are consistent with the scoring methods for teacher teaching quality.

[0070] The program layer is the various courses entered into the database.

[0071] According to the relationship between various factors in the system, the overall goal is determined and an operational hierarchical analysis model is established, such as Figure 2 shown.

[0072] First, the data of each course at the program level are integrated and normalized. Here, we perform zSCORE standardization on the course indicator matrix, normalize each data point according to its difference from the overall data mean, so that the final data mean is 0 and the standard deviation is 1, and obtain the sample course indicator data normalization table. Next, the indicator weights at the criterion level are calculated. The importance of the relevant indicators at the criterion level is compared, and the results are quantified using the 19-scale method. According to the factors that affect the comprehensive score of the course, the importance of the indicators is set, and the judgment matrix table 1 at the criterion level is obtained:

[0073] Table 1 Criteria layer judgment matrix

[0074]

[0075] The judgment matrix needs to pass the consistency test to confirm the hierarchical single ordering. If the consistency index CR < 0.1, it passes the consistency test. The CR calculation formula is: CR = CI / RI.

[0076] The comprehensive score S of the course is the sum of the product of the weight of each indicator and the indicator data matrix corresponding to the course. According to the weight and the normalized dish indicator data matrix, the comprehensive score of the course is calculated using the hierarchical analysis model. The calculation formula is:

[0077]

[0078] Among them, C is the course name, C Ai is the indicator data corresponding to the course, w i is the weight of the indicator, S C Give a comprehensive rating to a course.

[0079] Through digital humans communicating with students and extracting student feature information, the pre-trained model and fine-tuning mechanism are used to understand the semantics of the text input by students during the communication. The text input by students during the communication is marked and cleaned, redundant information such as stop words and modal particles are removed, and the text format is standardized, such as the use of standardized terms, processing spelling errors, etc., as the student's feature information, and the extracted student personal information is transmitted to the backend server.

[0080] Use the word vector model to embed the target student u tar The characteristic information and student u i The characteristic information is mapped into information vector I tar and I i .

[0081] Calculate the target student u using cosine similarity calculation tar The information vector I tar With students i The information vector I i The similarity betweentar ,u i ).

[0082] The information vectorization and similarity calculation utilize the BERT model to implement text vectorization and similarity calculation, and its construction method includes the following steps:

[0083] Step 1: Environment construction and data preparation:

[0084] Clean the collected text data, including removing stop words and punctuation marks, and performing stemming or word form restoration.

[0085] Step 2: Model training and fine-tuning:

[0086] Divide the dataset into training set, validation set, and test set. Select a pre-trained BERT model suitable for the task, use the training set to fine-tune the BERT model, and complete the model construction to adapt to the course recommendation task.

[0087] Step 3: Text vectorization

[0088] Input the student’s feature information into the BERT model. Use the output of the [CLS] token of the BERT model as the vector representation of the feature information.

[0089] Step 4: Calculate text similarity

[0090] Use the cosine similarity formula to calculate the target student u tar The information vector I tar With students i The information vector I i The similarity between tar ,u i ), as the target student u tar With candidate students i The similarity between them is calculated as:

[0091]

[0092] Among them, I tar For target students tar The information vector, I i For students i information vector.

[0093] According to the similarity W, select the student with the target u. tar The top n similar students are taken as the candidate student table U.

[0094] Calculate the candidate student u in the candidate student table U j Courses that have been learned but the target students have not learned c i For target studentstar The fitness P(u tar , c j ). The calculation formula is:

[0095]

[0096] Among them, N(u tar ,u j ) represents student u tar Uneducated but candidate students j A collection of courses you have taken, represents the candidate student u j For Course C i Rating of W(u tar ,u j ) is the target student u tar With candidate students j The similarity between

[0097] Sort the student's fitness P value according to the candidate courses, and generate an ordered candidate course table T, in which the courses will be given to student u. tar The candidate courses will serve as the basis for the subsequent calculation of course recommendation.

[0098] (II) Construct a course matching analysis model to calculate the matching degree between the candidate courses and the selected courses of the target students:

[0099] 1. Get the target student u from the course database tar The selected course information of the student. i and course c j The situation is regarded as a course pair, and the frequency of occurrence of the course pair is the support (c i , c j ), calculate the support between courses and filter out all candidate course pairs whose support is higher than a certain threshold. The calculation formula is:

[0100]

[0101] Among them, |N(c i )∩N(c j )| means to select course c at the same time i and course c j and |N(U)| represents the number of students who have selected the course.

[0102] 2. Calculate the confidence of each course in the course pair and determine the course c j In the course c i The frequency of occurrence in the course pair of c, that is, in the course c iSelect c under the condition j The probability of , and vice versa. The calculation formula is:

[0103]

[0104] Among them, |N(c i )∩N(c j )| means to select course c at the same time i and course c j The number of students, |N(c i )| means select course c i The number of students, |N(c j )| means select course c j the number of students.

[0105] 3. Construct a course-course bipartite graph. If a student chooses course c i and course c j , then in course c i and course c j Add an edge between them, the weight of the edge is the confidence, and calculate the matching degree Q for each candidate course pair. The calculation formula is:

[0106]

[0107] in, In the tth iteration, course c i and course c j α is a regulating factor that controls the balance between the new matching degree and the initial matching degree, and its value is in the range of [0,1]. i ) indicates that the course c i A collection of directly connected courses. ij It is a course c i and course c j The initial edge weights between .

[0108] 4. Construct the course matching table G based on the course support and confidence, and regard Q as the matching degree between two courses.

[0109] (III) Calculate the course recommendation degree and rank and recommend candidate courses according to the course recommendation degree:

[0110] 1. Find course c in candidate course list T i and its effect on target students tar The fitness P.

[0111] 2. Find the course c that appears in the course matching table G. j and candidate courses c i, thereby obtaining the matching degree Q; if it does not exist, the matching degree is 0.

[0112] 3. Combine the course suitability P for students and the matching degree Q between courses to obtain the course recommendation degree R (c i ), indicating that the target students are learning course c i The final course recommendation R(c i ) is obtained by weighting the suitability of the course to the student and the matching degree between courses, where the weights of the suitability of the course to the student and the matching degree between courses are w1 and w2 respectively. The calculation formula is:

[0113] R(c i )=w1×P+w2×Q

[0114] Among them, w1 and w2 are the weights of the course’s suitability P for students and the matching degree Q between courses, respectively. Their values ​​are in the range of [0,1], and their sum is 1.

[0115] 4. According to the course recommendation R(c i ) Sort the candidate courses and select the top k courses to recommend to the target student u tar .

[0116] Through the application of these technologies, our intelligent course recommendation model based on digital human interaction can better understand students' course selection needs and provide them with personalized course recommendation services, thereby improving student experience and satisfaction.

[0117] Embodiment 2

[0118] The present invention provides an intelligent course recommendation method based on digital human interaction, the method comprising:

[0119] Step 1: Construct a course rating analysis model to calculate the suitability of the course for the target students.

[0120] 1. The course evaluation index system is divided into the target layer, the criterion layer and the program layer. The target layer is the comprehensive score of a course, and the criterion layer includes the teacher's teaching quality, course practicality, course assessment difficulty and course attractiveness, which are recorded as S1, S2, S3 and S4 respectively. The program layer is the various courses entered in the database.

[0121] 2. Integrate and normalize the data of each course at the program level, perform z-SCORE standardization on the course-indicator matrix, and obtain a normalized table of course-indicator data.

[0122] 3. The 19-scale method is used to quantify the importance of indicators, and a criterion layer judgment matrix is ​​constructed. The consistency of the judgment matrix is ​​confirmed through consistency testing, and the weight of each indicator is obtained based on the judgment matrix and the constructed hierarchical analysis model.

[0123] 4. Calculate the comprehensive score of the course. According to the weighted and normalized course indicator data matrix, use the hierarchical analysis model to calculate the comprehensive score of each course. Use the comprehensive score of the course as the student's score for the course to establish a student-course score matrix.

[0124] 5. Through digital humans communicating with students and extracting student feature information, the pre-trained model and fine-tuning mechanism are used to understand the semantics of the text input by students during the communication. The text input by students during the communication is marked and cleaned, redundant information such as stop words and modal particles are removed, and the text format is standardized, such as the use of standardized terminology and the processing of spelling errors, etc., as the student's feature information.

[0125] The specific method is to remove the name of the conversation user and other information from the question and answer content between the digital human and the students, and then use spaces to connect all the conversation content to form a complete text sequence for each student, which is used as the original text description to characterize the student's interests.

[0126] 6. Use the word vector model to embed the target student u tar The characteristic information and student u i The characteristic information is mapped into information vector I tar and I i .

[0127] 7. Use cosine similarity to calculate the target student u tar The information vector I tar With students i The information vector I i The similarity between tar ,u i ).

[0128] 8. Select the target student u according to the similarity tar The top n similar students are taken as the candidate student table U.

[0129] 9. Calculate the candidate student u in the candidate student table U j Courses that have been learned but the target students have not learned c i For target students tar The fitness P(u tar , c i ).

[0130] Step 2: Construct a course matching analysis model and calculate the target student u tar The compatibility between candidate courses and selected courses.

[0131] 1. Get the target student u from the course database tar Selected course information, students will select courses at the same time ci and course c j The case is regarded as a course pair, and the frequency of occurrence of the course pair is the support (c i , c j ), calculate the support between courses, and filter out all candidate course pairs whose support is higher than a certain threshold.

[0132] 2. Calculate the confidence of each course in the course pair and determine the course c j In the course c i The frequency of occurrence in the course pair of c, that is, in the course c i Select c under the condition j and vice versa.

[0133] 3. Construct a course-course bipartite graph. If a student chooses course c i and course c j , then in course c i and course c j Add an edge between them, the weight of the edge is the confidence, and calculate the matching degree Q for each candidate course pair. Construct the course matching table G according to the course support and confidence, and regard Q as the matching degree between two courses.

[0134] Step 3: Weight the suitability of candidate courses for target students and the compatibility between courses to obtain the course recommendation degree, and rank and recommend candidate courses based on the course recommendation degree.

[0135] 1. Find course c in candidate course list T i and its effect on target students tar The fitness P.

[0136] 2. Find the course c that appears in the course matching table G. j and candidate courses c i , thereby obtaining the matching degree Q; if it does not exist, the matching degree is 0.

[0137] 3. Weight the course suitability P for students and the matching degree Q between courses respectively to obtain the course recommendation degree R(c i ), which means for the target student u tar Study Coursec i The final course recommendation R(c i ) is obtained by weighting the suitability of the course for the student and the compatibility between courses.

[0138] 4. Sort the candidate courses according to the course recommendation degree R, and select the top k courses to recommend to the target student u tar .

[0139] Those skilled in the art can clearly understand that for the convenience and simplicity of description, only the division of the above-mentioned functional units and modules is used as an example. In actual applications, the above-mentioned functions can be distributed and completed by different functional units and modules as needed, that is, the internal structure of the system described in the present application is divided into different functional units or modules to complete all or part of the functions described above.

Claims

1. An intelligent course recommendation method based on digital human interaction, characterized in that: include: Extract students' ratings of courses from the rating database and build a course rating analysis model to calculate the suitability P of the course for the target students; Obtain relevant information about the course from the course database and build a course matching analysis model to calculate the matching degree Q between the candidate courses and the selected courses of the target students; The suitability P of candidate courses for target students and the matching degree Q between courses are weighted respectively to obtain the course recommendation degree R, and the candidate courses are ranked and recommended according to the course recommendation degree R.

2. According to claim 1, the intelligent course recommendation method based on digital human interaction is characterized in that: The specific steps include: Step 1: Establish a student-course rating matrix as a basis for calculating the suitability of courses for students; Step 2: Communicate with students through digital humans, collect and extract their characteristic information and transmit it to the background; Step 3: Calculate the similarity W between the student and other students based on the feature information; Step 4: Based on the similarity calculated in step 3, combine the student-course rating matrix and calculate the suitability P of the course for the student.

3. The intelligent course recommendation method based on digital human interaction according to claim 2 is characterized in that: The specific steps of establishing the student-course rating matrix in step 1 include: Step 1-1: Divide the course evaluation index system into the target layer, the criterion layer and the scheme layer, where the target layer is k, the criterion layer includes the teacher's teaching quality, the course practicality, the course assessment difficulty and the course attractiveness, which are recorded as S1, S2, S3, S4 respectively, and the scheme layer is the various courses entered in the database; Step 1-2: Integrate and normalize the data of each course at the program level, perform z-SCORE standardization on the course-indicator matrix, and obtain a normalized table of course-indicator data; Step 1-3: Solve the indicator weights of the criterion layer, compare the importance of related indicators on the criterion layer, and use the 1-9 scaling method to obtain the judgment matrix; Step 1-4: Quantify and pass the consistency test to obtain the weight w of each indicator i ; Step 1-5: Calculate the comprehensive score of the course. According to the weighted and normalized dish index data matrix, the hierarchical analysis model is used to calculate the comprehensive score of the course. The calculation formula is: Among them, C is the course name, C Ai is the indicator data corresponding to the course, w i is the weight of the indicator, S c Give a comprehensive rating to a course; Step 1-6: Use the comprehensive score of the course as the student's score for the course and establish a student-course rating matrix.

4. The intelligent course recommendation method based on digital human interaction according to claim 2 is characterized in that: The specific steps of step 2 in which the digital human communicates with the student and extracts the student's characteristic information include: Step 2-1: After removing the names of the conversation users and other information, connect all the conversation contents with spaces; Step 2-2: A complete text sequence is formed for each student, and this is used as the original text description representing each student's interests; Step 2-3: Transmit student interest feature information to the backend server.

5. The intelligent course recommendation method based on digital human interaction according to claim 2 is characterized in that: The specific steps of step 3 for calculating the similarity W between the student and other students include: Step 3-1: Use the word vector model to embed the target student u tar The characteristic information and student u i The characteristic information is mapped into information vector I tar and I i , Step 3-2: Calculate the target student u using cosine similarity calculation tar The information vector I tar With students i The information vector I i The similarity between tar ,u i ), the calculation formula is:

6. The intelligent course recommendation method based on digital human interaction according to claim 2 is characterized in that: The specific steps of calculating the suitability P of the course to the student in step 4 include: Step 4-1: Sort all students according to the similarity W between the student and other students calculated in claim 5 and select the student with the target student u tar The top n similar students are taken as the candidate student table U; Step 4-2: Calculate the candidate student u in the candidate student table U j Courses that have been learned but the target students have not learned c i For target students tar The fitness P(u tar , c i ), the calculation formula is: Among them, N(u tar ,u j ) represents student u tar Uneducated but candidate students j A collection of courses you have taken, represents the candidate student u j For Course C i The rating of W(u tar u j ) is the target student u tar With students j The similarity between Step 4-3: Sort the student's fitness P according to the candidate courses and construct a candidate course table T as the target student u tar Candidate courses.

7. The intelligent course recommendation method based on digital human interaction according to claim 1 is characterized in that: The specific steps of constructing a course matching analysis model include: Step 1: Get the target student u from the course database tar Information about selected courses; Step 2: Students can choose courses c i and course c j The situation is regarded as a course pair, and the frequency of occurrence of the course pair is the support (c i , c j ), calculate the support between courses, the calculation formula is: Among them, |N(c i )∩N(c j )| means to select course c at the same time i and course c j , |N(U)| represents the number of students who have selected courses; Step 3: Filter out all candidate course pairs whose support is higher than a certain threshold.

8. The intelligent course recommendation method based on digital human interaction according to claim 1, characterized in that: The specific steps for calculating the collocation degree Q between courses include: Step 1: Calculate the confidence of each course in the course pair and determine the course c j In the course c i The frequency of occurrence in the course pair of c, that is, in the course c i Select c under the condition j The probability of , and vice versa, is calculated as: Among them, |N(c i )∩N(c j )| means to select course c at the same time i and course c j The number of students, |N(c i )| means select course c i The number of students, |N(c j )| means select course c j Number of students; Step 4: Construct a course-course bipartite graph. If a student also chooses course c i and course c j , then in course c i and course c j Add an edge between them, and the weight of the edge is the confidence calculated in step 3; Step 5: Calculate the matching degree Q for each candidate course pair. The calculation formula is: in, In the tth iteration, course c i and course c j The matching degree between them, α is an adjustment factor, which controls the balance between the new matching degree and the initial matching degree, and its value is in the range of [0, 1]. i ) indicates that the course c i A collection of directly connected courses, w ij It is a course c i and course c j The initial edge weights between Step 6: Construct the course matching table G based on the course support and confidence, and use Q as the matching degree between the two courses in each course pair.

9. The intelligent course recommendation method based on digital human interaction according to claim 1, characterized in that: The steps of calculating the course recommendation R and recommending courses specifically include: Step 1: Find course c in the candidate course table T i and its effect on target students tar The fitness P, Step 2: Find the course c that appears in the course matching table G. j and candidate courses c i The matching scheme is obtained to obtain the matching degree Q; if it does not exist, the matching degree is 0, and the course suitability P for students and the matching degree Q between courses are weighted respectively to obtain the course recommendation degree R, which is calculated as follows: R(c i )=w1×P+w2×Q Among them, R(c i ) represents the target student u tar Study Coursec i w1 and w2 are the weights of the course’s suitability for students P and the matching degree between courses Q, respectively. The values ​​are in the range of [0, 1], and the sum of the two is 1. Step 3: Sort the candidate courses according to the course recommendation degree R, and select the top k courses to recommend to the target student u tar .

Images