Recommendation method based on machine learning and big data analysis
Patent Information
- Application Number
- CN202611384365.8
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-09-08
- Publication Date
- 2026-10-13
AI Technical Summary
[0003]已有的课程推荐方法大多针对在线课程，基于课程内容进行相关课程推荐，或者基于用户经验进行课程推荐，这些推荐方法仅专注于用户兴趣，鲜有从实际学习结果角度入手，专门适用于高等教育硬性线下选修课的课程推荐方法
[0035]本发明将课程的成绩与用户的学习能力进行强耦合，以用户的学习能力入手，通过为各课程赋予能够表征其学习能力需求的向量，同时为各用户均赋予能够表征其学习能力的向量，并基于此进行课程推荐，大大提高了课程推荐的精准度，最大程度确保向用户推荐其擅长学习的课程。
Smart Images
Figure QLYQS_2
Abstract
Description
Technical Field
[0001] This invention relates to the fields of deep reinforcement learning and big data analytics, specifically to recommendation methods based on machine learning and big data analytics. Background Technology
[0002] With the popularization of the concept of comprehensive quality education, elective courses have become a necessary course in higher education. There are no special professional restrictions on elective courses, and they are available to students from all institutions. For students, the initial intention of choosing elective courses is often to obtain a good grade with the highest cost performance. This is especially true for working postgraduate students and students in adult education institutions. While balancing work and studies, they especially need to select courses that they are good at and that are easy to get good grades from the vast number of elective courses.
[0003] Existing course recommendation methods are mostly designed for online courses, recommending courses based on course content or user experience. These methods focus only on user interests and rarely consider actual learning outcomes. There are few course recommendation methods specifically suitable for mandatory offline elective courses in higher education. Summary of the Invention
[0004] To address the shortcomings of existing technologies, this invention provides a recommendation method based on machine learning and big data analysis, comprising the following steps:
[0005] S1: Based on each user's learning ability attributes, obtain the attribute vector of each user. Based on the attribute vector of the group of users, cluster the group of users to obtain multiple user classes;
[0006] S2: For each user, identify them as the target user and their user class as the target class. Iterate through the recommended courses that users in the target class have studied but the target user has not. Identify the currently iterated recommended course as the target course and determine the comprehensive recommendation factor of the target course using the following method:
[0007] S21: Determine the matching factor between the target course and the target class based on the total average score of the target course and the average score of the target course category; determine the matching factor between the target user and the target class based on the attribute vector of the target user and the attribute vector of the group users of the target class; determine the macro recommendation factor of the target course based on the matching factor between the target course and the target class, and the matching factor between the target user and the target class.
[0008] S22: Determine the micro-recommendation factor of the target course based on the difficulty factor of the target course, the attribute vector of the target user, the attribute vector of the target class group users, and the performance achieved in the target course;
[0009] S23: Determine the neutral recommendation factor for the target course based on the difficulty factor of the target course;
[0010] The difficulty factor of the target course is determined by the total average score of the target course and the average score of all courses studied by all users;
[0011] S24: The comprehensive recommendation factor for the target course is obtained by weighting and summing the macro recommendation factors, micro recommendation factors, and neutral recommendation factors using the recommendation model.
[0012] S3: Sort the target courses in reverse order according to the comprehensive recommendation factors and recommend the top-ranked courses to the target users.
[0013] In S21, the method for determining the fit factor between the target course and the target class is as follows:
[0014] Q1 = LP / KP;
[0015] Where Q1 is the fit factor between the target course and the target class, KP is the total average score of the target course, and LP is the average score of the target course class;
[0016] The overall average score of the target course is the average score achieved by the group of users learning the target course across all user categories;
[0017] The average score for the target course category is the average score achieved by the group of users in the target category who are learning the target course.
[0018] The method for determining the fit factor between the target user and the target class is as follows: take the mean of the attribute vectors of the target class group users to obtain the target class group attribute vector, and take the inner product of the target user attribute vector and the target class group attribute vector to obtain the fit factor Q2 between the target user and the target class.
[0019] Target Course Macro Recommendation Factor T1=Q1 Q2.
[0020] The method for determining the difficulty factor of the target course is as follows:
[0021] N = AP / KP;
[0022] Where KP is the overall average score of the target course, and AP is the overall average score;
[0023] The overall average score of the target course is the average score achieved by the group of users learning the target course across all user categories;
[0024] The overall average score is the average score of all courses taken by all users.
[0025] In S22, the method for determining the micro-recommendation factors of the target course is as follows:
[0026] ;
[0027] Among them, T2 is the micro-recommendation factor for the target course;
[0028] Let m be the target user attribute vector, where m is the number of users in the target class who have studied the target course, and j is the index of the user group in the target class. Let j be the attribute vector of the j-th group of users in the target class. It is the inner product of the target user attribute vector and the attribute vector of the j-th group of users in the target class;
[0029] FM represents the score achieved by the j-th group of users in the target class on the target course, with FM being the full score.
[0030] In S23, the method for determining the neutral recommendation factor for the target course is as follows:
[0031] T3 = 1 / N;
[0032] Among them, T3 is the neutral recommendation factor for the target course.
[0033] This also includes S4: feeding back the learning results of the target user on the recommended courses to the recommendation model, so that the recommendation model can learn and update the weight values of the macro recommendation factor, micro recommendation factor and neutral recommendation factor of the target course.
[0034] Among them, the user's learning ability attributes include memory, attention, comprehension, logical reasoning ability, and abstract thinking ability.
[0035] This invention strongly couples course grades with users' learning abilities. Starting with users' learning abilities, it assigns a vector to each course that represents their learning ability needs, and assigns a vector to each user that represents their learning abilities. Based on this, course recommendations are made, which greatly improves the accuracy of course recommendations and ensures that courses that users are good at learning are recommended to the greatest extent possible. Detailed Implementation
[0036] To gain a better understanding of the technical solution and beneficial effects of the present invention, the technical solution of the present invention and its beneficial effects are described in detail below.
[0037] This invention provides a recommendation method based on machine learning and big data analysis, specifically applicable to university elective courses. The system is connected to the academic affairs office, maintains the grades of each course for each user, and makes course recommendations based on the user's learning ability and course grades. A preferred embodiment mainly includes the following steps S1 to S5.
[0038] S1: Based on each user's learning ability attributes, obtain the attribute vector of each user. Based on the attribute vector of the group of users, cluster the group of users to obtain multiple user classes.
[0039] A user's learning ability attributes are closely related to their learning abilities in different subjects, such as memory, attention, comprehension, logical reasoning, and abstract thinking. Different attributes determine the types of subjects a user excels at: for example, users with strong memory are good at subjects with dense knowledge points that are not strongly related to each other and rely on memorization; users with strong attention are good at subjects involving complex formulas and reasoning, such as advanced mathematics and quantum mechanics, which require a high degree of concentration; users with strong comprehension and logical reasoning abilities are good at subjects that require strong thinking skills, such as applied mathematics, philosophy, and economics; and users with strong abstract thinking abilities are good at geometry.
[0040] Therefore, acquiring users' learning ability attributes can lay the foundation for accurate recommendations of courses that match users' strengths in the future.
[0041] In this invention, when a user first enters the system, their attribute vector is determined through a questionnaire. To avoid inaccurate attribute vector acquisition due to subjective judgment, a standard questionnaire can be set up. Based on the user's responses to the standard questionnaire, the user's learning ability in each dimension is mapped to a specific value, thereby obtaining the user's attribute vector. For example, regarding memory ability, the following questions and answers can be provided:
[0042] (1) For a humanities subject knowledge point that requires 20 class hours, how many times do you need to memorize it to the point that you can answer any question after understanding it?
[0043] A: 3-5 times B: 5-10 times C: 10-15 times D: More than 15 times;
[0044] (2) How long does it take to memorize a seven-character quatrain by repeatedly reading it aloud?
[0045] A: 10 minutes B: 20 minutes C: 30 minutes D: More than 30 minutes;
[0046] …………
[0047] Based on the user's questionnaire results, a total score reflecting memory ability is obtained. This total score is then mapped to a specific value between 1 and 10. The magnitude of this value is positively correlated with memory ability; the higher the value, the better the memory.
[0048] Similarly, based on similar questionnaires, the user's attention, comprehension, logical reasoning, and abstract thinking abilities are obtained. The values obtained for each are mapped to the values of specific dimensions in the user attribute vector, resulting in the user attribute vector. For example, if a user scores 3, 4, 5, 6, and 7 for memory, attention, comprehension, logical reasoning, and abstract thinking abilities respectively after completing a standard questionnaire, then the user's attribute vector is (3, 4, 5, 6, 7).
[0049] In this invention, clustering of users based on attribute vectors can be achieved using any existing technology. A common method is to cluster users using the K-means method. This groups users with similar learning abilities together. When recommending courses to users later, the system selects courses with higher scores from the courses studied by other users in the same group and recommends them to that user. This is equivalent to recommending courses that users are good at with a high probability.
[0050] S2: For each user, identify them as the target user and the user class to which the target user belongs as the target class. Iterate through the recommended courses that the target class users have studied but the target user has not studied. Identify the currently iterated recommended courses as target courses. Determine the comprehensive recommendation factor of the target courses through the following methods S21 to S24.
[0051] S21: Determine the fit factor between target courses and target classes based on the overall average score of the target courses and the average score of the target classes:
[0052] Q1 = LP / KP; (1)
[0053] Where Q1 is the fit factor between the target course and the target class, KP is the total average score of the target course, and LP is the average score of the target course class;
[0054] The overall average score of the target course is the average score achieved by the group of users learning the target course across all user categories;
[0055] The average score for the target course category is the average score achieved by the group of users in the target category who are learning the target course.
[0056] If users are divided into 5 classes using K-means, with users 1-10, 11-20, 21-30, 31-40, and 41-50 in each class, and users who have already studied the target course in each class being users 2-10, 11-20, 21-30, 31-40, and 41-50 respectively, and user 1 is currently the target user, and users 1-10 belong to the target class, then the overall average score of the target course is the overall average score of the target course obtained by users 2-50, while the average score of the target course class is only the average score of the target course obtained by users 2-10.
[0057] The ratio of the average score of the target course category to the overall average score of the target courses directly reflects the level of proficiency of users in the target category in the target courses. The higher the ratio, the better the users in the target category have achieved in the course compared to users in other categories. This also indirectly indicates that the target courses are a better fit for the target category.
[0058] The mean of the attribute vectors of the target class group users is taken to obtain the target class group attribute vector. The inner product of the target user attribute vector and the target class group attribute vector is taken to obtain the target user and target class fit factor Q2.
[0059] Macro-level recommendation factors for target courses are determined based on the matching factors between target courses and target categories, as well as the matching factors between target users and target categories.
[0060] T1=Q1 Q2. (2)
[0061] The macro-recommendation factor is based on the matching procedure between the target user and the target class, and the matching degree between the target course and the target class to obtain the matching degree between the target user and the target course. The underlying calculation logic takes into account the performance of the target course in the target class and the performance of the entire group of users, as well as the similarity of the learning ability between the target user and the group of users in the target class. From a macro perspective, it reflects the potential proficiency of the target user in the target course.
[0062] S22: Determine the micro-recommendation factor for the target course based on the difficulty factor of the target course, the attribute vector of the target user, the attribute vector of the target class's group users, and their performance on the target course:
[0063] (3)
[0064] Among them, T2 is the micro-recommendation factor for the target course;
[0065] Let m be the target user attribute vector, where m is the number of users in the target class who have studied the target course, and j is the index of the user group in the target class. Let j be the attribute vector of the j-th group of users in the target class. It is the inner product of the target user attribute vector and the attribute vector of the j-th group of users in the target class;
[0066] FM represents the score achieved by the j-th group of users in the target class on the target course, where FM is the full score.
[0067] N is the difficulty factor of the target course, and its value is positively correlated with the learning difficulty of the target course. The method for determining it is as follows:
[0068] N = AP / KP; (4)
[0069] Where KP is the overall average score of the target course, and AP is the overall average score;
[0070] The overall average score of the target course is the average score achieved by the group of users learning the target course across all user categories;
[0071] The overall average score is the average score of all courses taken by all users.
[0072] In other words, when determining a target user's potential proficiency in a target course, the micro-recommendation factor calculates the similarity of the target user's attribute vector with each group of users in the target class. The logic is equivalent to using the similarity between each group of users and the target user as a weighted value, weighting the ratio of the group user's score in the target course to a perfect score, and using the difficulty factor of the target course as a correction. Unlike the macro-level analysis that considers both the user's attribute vector and the target class's group attribute vector as a whole, the micro-recommendation factor calculates the similarity of the target user's attribute vector with each group of users in the target class. This strongly couples the target course score with one or more specific attributes in the attribute vector, strengthening the contribution of specific attributes closely related to the target course and improving recommendation accuracy.
[0073] S23: Determine the neutral recommendation factor for the target course based on the difficulty factor of the target course;
[0074] T3 = 1 / N. (5)
[0075] The role of the neutral recommendation factor is that some courses may not have an intuitive relationship with the user's learning ability. In other words, whether it is for users with strong memory or users with strong comprehension, their learning performance in the target course is directly related to the amount of time they contribute to the learning. That is, the performance in such courses mainly depends on effort and has no direct correlation with personal talent.
[0076] S24: The comprehensive recommendation factor for the target course is obtained by weighting and summing the macro recommendation factors, micro recommendation factors, and neutral recommendation factors using the recommendation model.
[0077] (6)
[0078] in, The comprehensive recommendation factors for the target course, , , These are the weight values for the macro recommendation factor, micro recommendation factor, and neutral recommendation factor, respectively.
[0079] It is important to note that the sum of the weights of the macro-level recommendation factor, micro-level recommendation factor, and neutral recommendation factor is 1. However, the individual weights of these factors vary for different courses.
[0080] For courses that are not strongly correlated with personal talent but whose academic performance is highly correlated with actual effort, the neutral recommendation factor has a higher weight value, so that courses with higher average scores are recommended to the target users, making it easier for them to achieve high scores.
[0081] For courses highly correlated with individual talents, macro-level and micro-level recommendation factors are given higher weights, while neutral recommendation factors are given lower weights, in order to recommend courses that target users are good at. These courses are further divided into two categories:
[0082] If the learning ability required by the target course is similar to the learning ability represented by the group attribute vector of the target class, since the group attribute vector represents the average attribute vector of all users in the target class, the learning ability required by the target course is often more generalized. For example, if the group attribute vector has values of (6, 6, 5, 2, 1), it means that learning the target course requires memory, attention, and comprehension, but the requirements of the target course for these three abilities are relatively balanced. In this case, a higher weight is given to the macro recommendation factor to recommend such courses to users who possess all three abilities.
[0083] If the learning ability required by the target course is clearly concentrated in one aspect, such as the target course focusing on the need for extremely strong memory, then users with high values in the memory dimension of the attribute vector are more likely to get high scores in the target course. In this case, a higher weight is given to the micro-recommendation factor. Thus, target users with strong memory have a high similarity in attribute vectors with every other group of users who get high scores in the target course. In this case, the micro-recommendation factor value is high, so that such courses are recommended to users with extremely strong memory.
[0084] In the initial stage of this invention, the weights of macro-recommendation factors, micro-recommendation factors, and neutral recommendation factors can be assigned by the instructors of each course based on their experience. Subsequently, the model performs reinforcement learning based on each user's attribute vector and learning performance in each course, so as to update the weights of macro-recommendation factors, micro-recommendation factors, and neutral recommendation factors of each course from time to time.
[0085] S3: Sort the target courses in reverse order according to the comprehensive recommendation factors and recommend the top-ranked courses to the target users.
[0086] S4: Feed back the learning results of the target users on the recommended courses to the recommendation model, so that the recommendation model can learn and update the weight values of the macro recommendation factor, micro recommendation factor and neutral recommendation factor of the target courses.
[0087] Specifically, if the target user's final academic performance in the recommended course is less than or equal to the course's "overall average score"... If the percentage is "(1 - predetermined percentage)," it indicates that the target user is not good at learning the recommended course. Such target users are identified as negative feedback target users, and negative feedback needs to be provided to the recommendation model. Conversely, if the target user's final learning score on the recommended course is greater than or equal to the course's "overall average score," then... If the value is "(1 + pre-booking percentage)", it means that the target user is good at learning the recommended courses. This type of target user is identified as a positive feedback target user, and positive feedback needs to be given to the recommendation model.
[0088] For each target course, based on its positive feedback target users and negative feedback target users, the loss function is constructed as follows:
[0089] (7)
[0090] Where L is the value of the loss function, l is the number of target users for negative feedback, n is the number of target users for positive feedback, p is the index of the target user for negative feedback, and q is the index of the target user for positive feedback;
[0091] These are the macro-recommendation factor and micro-recommendation factor for the target course determined based on the target class to which the p-th negative feedback target user belongs, respectively. These are the macro-recommendation factor and micro-recommendation factor for the target course, determined based on the target class to which the q-th positive feedback target user belongs. For a given target course, since different negative and positive feedback target users may belong to different target classes, the macro-recommendation factor and micro-recommendation factor of the target course will be different when constructing the loss function based on the comprehensive recommendation factor for different target users. However, the neutral recommendation factor depends only on the difficulty level of the target course and is unrelated to the target class to which the target user belongs, so it remains unchanged.
[0092] This invention uses the user's learning ability as the benchmark for all recommendations. Therefore, the assessment of the user's learning ability should be sufficiently accurate, and a strong correlation between learning ability and academic performance should be ensured. In practice, it is difficult to fully achieve this by simply relying on questionnaires. Furthermore, the user's various abilities may change during continuous learning and training. Therefore, in the preferred embodiment of this invention, step S5 is also included: for each user, update their attribute vector according to a preset period based on their acquired learning performance. The specific method includes the following steps S51-S54.
[0093] S51: Determine the attribute vector of the user based on the survey document as the initial attribute vector, and determine the courses that the user has completed as historical learning courses.
[0094] S52: Determine the demand attribute vector of the history learning course based on the group of history learning users behind the history learning course; where the group of history learning users is all users who have studied the corresponding history learning course (not limited to their category).
[0095] (8)
[0096] In the above formula, XC is the demand attribute vector for the corresponding history learning course, S is the number of history learning users in the group who have studied the corresponding history learning course, and r is the index of the history learning users in the group. Let represent the scores achieved by the r-th group of history learners in the corresponding history learning course. Let r be the attribute vector of the r-th group of historical learning users; therefore, the technical essence of the above formula is to take the attribute vectors of the group users who have studied the corresponding historical learning course, and sum them up by weighting according to the learning performance of each group user to obtain the demand attribute vector that can represent the learning ability required to learn the corresponding historical learning course.
[0097] S53: Determine the user's performance representation attribute vector based on the user's grades in each history learning course, the demand attribute vector of each history learning course, and the difficulty factor.
[0098] (9)
[0099] In the above formula, PC is the attribute vector representing the user's performance, t is the number of the user's historical learning courses, and e is the index of the historical learning courses. This refers to the user's score in the e-th history learning course. Let e be the difficulty factor of the e-th history learning course. Let be the demand attribute vector of the e-th history learning course; therefore, the technical essence of the above formula is: to sum the demand attribute vectors of the history learning courses that the user has studied according to the corrected grades obtained by the user in each history learning course, so as to obtain the performance representation attribute vector that can represent the user's learning ability.
[0100] S54: Obtain the updated attribute vector for the user by using the performance representation attribute vector and the initial attribute vector.
[0101] (10)
[0102] In the above formula, CC is the user's initial attribute vector, η is a fixed constant value less than 1, and T is the total number of courses currently included (regardless of category); through " This ensures that as the t-value gradually increases, the weight values of the performance representation attribute vector can still increase to a certain extent, thus avoiding the long-tail effect that occurs when the performance representation attribute vector represents the user's learning ability during the process of increasing the t-value.
[0103] As can be seen from the above process, a user's attribute vector needs to refer to the attribute vectors of other users during the update process. Therefore, in order to avoid "reference fluctuations" that occur during the simultaneous update of multiple user attribute vectors, this invention updates only one user's attribute vector within a predetermined period or at a predetermined time point.
[0104] In this invention, when determining the grades of each course and performing calculations based on those grades, the grades and full marks of each course can be mapped onto the same data dimension. For example, if the full mark for most courses is 100, then for some courses with a full mark of 60 or 150, before implementing the method of this invention, their full marks are mapped to 100 using a ratio mapping method, and the corresponding grades are mapped to the range of 1-100. All scores mentioned throughout this invention are scores mapped onto the same data dimension.
[0105] As described above, the recommendation method based on machine learning and big data analysis of this invention is mainly applicable to the recommendation of university elective courses. When recommending courses, target courses are determined from the set of elective courses. However, to ensure sufficient data volume, the determination of the course requirement attribute vector and the user attribute vector can be based on the grades of all courses taken by all users. Therefore, it needs to be further clarified that:
[0106] In step S2, the "recommended courses" in "iterate through the recommended courses that the target users have studied but the target users have not studied" can refer to the elective courses that the target users have studied but the target users have not studied.
[0107] In step S51, the so-called "courses that the user has completed" includes both elective courses that the user has completed and compulsory courses that the user has completed.
[0108] Similarly, in step S54, the total number of courses T currently included also includes all elective and compulsory courses currently included.
[0109] In summary, this invention strongly couples course performance with users' learning abilities. Starting with users' learning abilities, it assigns a vector representing the learning ability requirements of each course, and assigns a vector representing the learning ability of each user. Based on this, course recommendations are made, which greatly improves the accuracy of course recommendations and ensures that courses that users are good at learning are recommended to the greatest extent possible.
[0110] Although the present invention has been described using the above preferred embodiments, it is not intended to limit the scope of protection of the present invention. Any changes and modifications made by those skilled in the art to the above embodiments without departing from the spirit and scope of the present invention shall still fall within the scope of protection of the present invention. Therefore, the scope of protection of the present invention shall be defined by the claims.
Claims
1. A recommendation method based on machine learning and big data analysis, characterized in that, Includes the following steps: S1: Based on each user's learning ability attributes, obtain the attribute vector of each user. Based on the attribute vector of the group of users, cluster the group of users to obtain multiple user classes; S2: For each user, identify them as the target user and their user class as the target class. Iterate through the recommended courses that users in the target class have studied but the target user has not. Identify the currently iterated recommended course as the target course and determine the comprehensive recommendation factor of the target course using the following method: S21: Determine the fit factor between the target course and the target class based on the total average score of the target course and the average score of the target course category; determine the fit factor between the target user and the target class based on the attribute vector of the target user and the attribute vector of the group users of the target class; Macro-recommendation factors for target courses are determined based on the matching factors between target courses and target categories, as well as the matching factors between target users and target categories. S22: Determine the micro-recommendation factor of the target course based on the difficulty factor of the target course, the attribute vector of the target user, the attribute vector of the target class group users, and the performance achieved in the target course; S23: Determine the neutral recommendation factor for the target course based on the difficulty factor of the target course; The difficulty factor of the target course is determined by the total average score of the target course and the average score of all courses studied by all users; S24: The comprehensive recommendation factor for the target course is obtained by weighting and summing the macro recommendation factors, micro recommendation factors, and neutral recommendation factors using the recommendation model. S3: Sort the target courses in reverse order according to the comprehensive recommendation factors and recommend the top-ranked courses to the target users.
2. The recommendation method based on machine learning and big data analysis as described in claim 1, characterized in that, In S21, the method for determining the fit factor between the target course and the target class is as follows: Q1 = LP / KP; Where Q1 is the fit factor between the target course and the target class, KP is the total average score of the target course, and LP is the average score of the target course class; The overall average score of the target course is the average score achieved by the group of users learning the target course across all user categories; The average score for the target course category is the average score achieved by the group of users in the target category who are learning the target course. The method for determining the fit factor between the target user and the target class is as follows: take the mean of the attribute vectors of the target class group users to obtain the target class group attribute vector, and take the inner product of the target user attribute vector and the target class group attribute vector to obtain the fit factor Q2 between the target user and the target class. Target Course Macro Recommendation Factor T1=Q1 Q2.
3. The recommendation method based on machine learning and big data analysis as described in claim 1, characterized in that, The method for determining the difficulty factor of the target course is as follows: N = AP / KP; Where KP is the overall average score of the target course, and AP is the overall average score; The overall average score of the target course is the average score achieved by the group of users learning the target course across all user categories; The overall average score is the average score of all courses taken by all users.
4. The recommendation method based on machine learning and big data analysis as described in claim 3, characterized in that, In S22, the method for determining the micro-recommendation factors of the target course is as follows: ； Among them, T2 is the micro-recommendation factor for the target course; Let m be the target user attribute vector, where m is the number of users in the target class who have studied the target course, and j is the index of the user group in the target class. Let j be the attribute vector of the j-th group of users in the target class. It is the inner product of the target user attribute vector and the attribute vector of the j-th group of users in the target class; FM represents the score achieved by the j-th group of users in the target class on the target course, with FM being the full score.
5. The recommendation method based on machine learning and big data analysis as described in claim 3, characterized in that, In S23, the method for determining the neutral recommendation factor for the target course is as follows: T3 = 1 / N; Among them, T3 is the neutral recommendation factor for the target course.
6. The recommendation method based on machine learning and big data analysis as described in claim 1, characterized in that, It also includes S4: feeding back the learning results of the target user on the recommended courses to the recommendation model, so that the recommendation model can learn and update the weight values of the macro recommendation factor, micro recommendation factor and neutral recommendation factor of the target course.
7. The recommendation method based on machine learning and big data analysis as described in claim 1, characterized in that, A user's learning ability attributes include memory, attention, comprehension, logical reasoning, and abstract thinking.