A hybrid recommendation method for MOOC courses based on fuzzy logic and teaching plan

Abstract

The invention discloses a hybrid recommendation method for MOOC courses based on fuzzy logic and teaching plan, comprising the following steps: step 1. according to MOOC open course data, the intermediate fuzzy logic is introduced into the ML_BPNN neural network, the loss function is optimized, and finally a hybrid score matrix containing MOOC courses and online student features is obtained; step 2. the global similarity between online students is introduced into the point-wise user decomposition hybrid similarity model, and the hybrid similarity of the interactive relationship between online students and MOOC courses is composed of a predefined similarity calculated by using the Pearson similarity and a learning similarity calculated according to the inner product of a potential vector; step 3. the PoiU-FHSM hybrid similarity calculation result is combined with a course mapping relationship established according to a course arrangement plan of a college or university syllabus, and finally the Top recommended MOOC courses and the pre-requisite and post-requisite recommended MOOC courses associated therewith are output. The invention mainly realizes the simultaneous recommendation of MOOC courses and the pre-requisite and post-requisite MOOC courses associated therewith.

Description

Technical Field

[0001] The present invention relates to a MOOC course recommendation method, and in particular to a MOOC course hybrid recommendation method based on fuzzy logic and teaching plan. Background Art

[0002] Online education has effectively promoted the efficient sharing and dissemination of learning resources, and is one of the signs of the diversified development of the education system. It has prompted a significant and rapid increase in the application demand of recommendation systems on the MOOC (Massive Open Online Courses) platform. The MOOC platform is open, flexible, and autonomous, and the data it contains has the characteristics of high knowledge relevance and high personalized needs. However, the MOOC learning model has certain limitations. For example, the MOOC platform cannot formulate a unified learning plan for online learners, the needs analysis of some online learners is unclear, and some of the knowledge recommended to online learners does not meet the requirements of online learners.

[0003] The teaching plan of colleges and universities is essentially a course study plan, which is an overall plan for setting up an effective course study sequence for the courses that need to be studied in a specified subject. According to the order of opening courses, some of these courses can be studied independently and have no relationship with other courses; some courses have prerequisite course requirements, and all of their prerequisite courses must be completed before studying these courses. The requirements of prerequisite courses define the priority relationship between courses. However, the current MOOC platform and related course recommendation system lack the setting and arrangement of course study plans for online learners based on their learning situation.

[0004] Fuzzy logic uses membership to represent the binary measurement state, and a proposition is no longer either true or false. Intermediate fuzzy logic not only distinguishes the clarity and fuzziness of knowledge, but also distinguishes the opposite negation relationship and contradictory negation relationship of knowledge. On the basis of traditional logical predicates, the concept and formal symbols of fuzzy predicates are defined to represent fuzzy concepts such as "partially", "opposite to" and "non". However, there are few studies on the application of fuzzy logic in recommendation systems. In particular, there is no research on the application of intermediate fuzzy logic in recommendation systems.

[0005] Therefore, the course recommendation system based on the MOOC platform needs to explore the relationship between online learners, between online learners and online courses, and between online courses, to meet the effective match between online learners' flexible learning needs and shared learning resources, provide online learners with flexible and effective learning plans, and improve online learners' learning efficiency and learning resource utilization. Summary of the invention

[0006] In order to solve the above problems, the present invention discloses a hybrid recommendation method for MOOC courses based on fuzzy logic and teaching plan. According to MOOC public course data, intermediate fuzzy logic is introduced into ML_BPNN neural network, and the loss function is optimized to finally obtain a hybrid score matrix containing MOOC courses and online student features; the global similarity between online students is introduced into the PoiU-FHSM hybrid similarity calculation model to obtain more relationships between student courses; the PoiU-FHSM hybrid similarity calculation result is combined with the course mapping relationship established according to the course arrangement plan of the syllabus to obtain the Top recommended MOOC courses and the prerequisite and subsequent recommended MOOC courses associated with them; the present invention combines MOOC course recommendation with the teaching plan of colleges and universities, and realizes the simultaneous recommendation of MOOC courses and the prerequisite and subsequent MOOC courses associated with them.

[0007] The technical solution of the present invention is:

[0008] A hybrid recommendation method for MOOC courses based on fuzzy logic and teaching plan includes the following steps performed in sequence:

[0009] Step 1. Based on the MOOC open course data, the intermediate fuzzy logic is introduced into the ML_BPNN neural network to optimize the loss function, and finally a mixed score matrix containing the characteristics of MOOC courses and online students is obtained;

[0010] Step 2. The global similarity between online students is introduced into the Pointwise User pair factored heterogeneous similarity model (PoiU-FHSM). The hybrid similarity of the interaction relationship between online students and MOOC courses is composed of the predefined similarity calculated by Pearson similarity and the learning similarity calculated by the inner product of the latent vector.

[0011] Step 3. Combine the PoiU-FHSM hybrid similarity calculation results with the course mapping relationship established according to the course schedule plan of the college syllabus, and finally output the top recommended MOOC courses and their associated prerequisite and post-recommended MOOC courses.

[0012] Preferably, step 1 of the present invention is:

[0013] Step 11. Use FCM (Fuzzy C-Means) clustering algorithm to cluster online student information and obtain the target student cluster. Let S = {s1, s2, ..., s m} is the set of student clusters after clustering, C = {c1, c2, ..., c n} is a set of courses. In the student cluster, an initial rating matrix is ​​constructed based on the students’ ratings of the courses, where r ij is the initial score of student i for course j, m is the number of students in the student cluster where the target student belongs, and n is the number of courses.

[0014] Step 12. By counting the sentiment keywords in the MOOC course review information, the min-max normalization method (as shown in Formula 1) is used to weight the initial score matrix of Step 11, and finally the student-course score matrix (as shown in Formula 2) can be obtained.

[0015]

[0016] Among them, x ij is the weighted value of student i’s review information on course j, f ij is the keyword frequency of student i for course j, min is the minimum frequency, and max is the maximum frequency.

[0017]

[0018] Among them, r′ ij =r ij +(x ij +a) / (1+a), x ij is the weighted variable of student i’s review information on course j, a is the smoothing coefficient, r ij is the initial rating matrix of student i’s rating of course j, r′ ij is the weighted score of student i on course j.

[0019] Step 13. For the student-course score matrix R, let the row vector set constitute the student matrix R*i (i∈{1,2,3,...,n}) and the column vector set constitute the course matrix R*j (j∈{1,2,3,...,m}).

[0020] In the ML_BPNN neural network model, the student-course score matrix R as the input is decomposed into the student vector R*i (i∈{1,2,3,...,n}) and the course vector R*j (j∈{1,2,3,...,n}), which are used as the input vectors of the two back-propagation neural networks respectively. The SGD (Stochastic Gradient Descent) algorithm is used to pass through the middle hidden layer l i After training, the student vector pi and course vector qj are output respectively, and the cosine similarity method is used to calculate the score prediction value. The output of the ML_BPNN neural network is shown in Formula 3.

[0021]

[0022] Among them, u i =R*i(i∈{1,2,3,...,n}) represents the score of the i-th student on all courses, v j =R*j(j∈{1,2,3,...,n}) represents the ratings of all students for the jth course. i is the output of student vector training, q j is the output result of the course vector training, is the final student i's score prediction training result for course j, by calculating p i and q j The cosine similarity is obtained.

[0023] Step 14. Let the implicit feedback vector C of student i for course j be ij =(C ij1 ,C ij2 ), explicit feedback vector E ij =(E ij1 ,E ij2 ). Define explicit feedback and implicit feedback as intermediary objects, and use the intermediary principle to update the loss function calculation formula of the ML_BPNN neural network in step 13. The loss function calculation formula of the fusion intermediary fuzzy logic is shown in Equation 4.

[0024]

[0025] r α =min(||C ij -C im ||),m∈[1,n],||C ij -C im || is the Euclidean distance between implicit feedback data vectors;

[0026] r β =min(||E ij -E jm ||),m∈[1,n],||E ij -E jm || is the Euclidean distance between explicit feedback data vectors;

[0027] Where R is the set of courses, is the predicted value, R ij is the initial value, p∈[0,1], which is used to adjust the influence of implicit feedback and explicit feedback on the final result.

[0028] According to the above method, we can finally get a mixed score matrix containing course and student characteristics.

[0029] Preferably, step 2 of the present invention is specifically:

[0030] Step 21. Based on the Poi-FHSM model, learning similarity is introduced to represent the global similarity relationship between students. Let U and U' be the student vector matrix, V and W be the course vector matrix, and predefine the similarity Reflect local similarity relationships by learning similarity Reflecting the global similarity relationship, it is extended to the PoiU-FHSM model. The student-student mixed similarity can obtain more relationships between students' courses and can capture the relationships in the dataset with close connections between students.

[0031]

[0032] Among them, U u· is the potential vector corresponding to student u in the student vector matrix, U′ u′· is the latent vector corresponding to student u' in the student vector matrix.

[0033] The Pearson similarity calculation method is used. The student-student predefined similarity calculation formula is shown in Formula 6, and the course-course predefined similarity calculation formula is shown in Formula 6.

[0034]

[0035] Among them, N(u,u') is the set of courses selected by students u and u'. u,v is the rating of student u for course v, R u′,v is the rating of student u' for course v, is the average rating of student u for all courses. is the average rating of student u' across all rated courses.

[0036] Step 22. Based on Formula 5, use As a normalization factor, the mixed similarity is normalized and the prediction rule is introduced from the course selection set C of student u. u Capturing student bias in u and course deviation b i To represent the student course preference, the updated hybrid similarity calculation formula is shown in Formula 7.

[0037]

[0038] Step 23. Use the stochastic descent gradient (SGD) algorithm to update the model parameters and obtain the student course preference matrix through the PoiU-FHSM model as shown in Formula 8.

[0039]

[0040] Among them, p ijrepresents the preference of student i for course j. The target student i’s preference for the course is P i* ={p i1 ,p i2 ,...,p m1}.

[0041] Preferably, step 3 of the present invention is as follows:

[0042] Step 31. Extract course labels from the teaching plans of colleges and universities, and use a bidirectional chain structure to establish the precedence and postorder relationship between course labels. The model layer saves course labels and course label relationships. The order of course labels represents the prev sequence and next sequence of corresponding course learning; the data layer saves all MOOC course information. A course label in the model layer corresponds to a MOOC course set in the data layer. The course label to which the MOOC course in the data layer belongs is determined by the label attribute.

[0043] Step 32. Based on step 23, merge sort and insertion sort are combined to sort the online students' preferences and obtain the top K MOOC courses as the TopK recommendation results.

[0044] Step 33. Based on step 31 and step 32, for the Top1 MOOC course, obtain the course label label at the data layer; obtain 0-m pre-order labels prelabel and 0-n post-order labels postlabel of label according to the prev and next attributes corresponding to label label at the model layer; return to the data layer, find the corresponding 0-m pre-order course sets and 0-n post-order course sets according to the pre-order label and post-order label; sort the course sets obtained by the pre-order label and the post-order label according to step 32, and obtain 1-k MOOC courses as the pre-requisite MOOC courses and post-requisite MOOC courses of the Top1 course. Combine the PoiU-FHSM hybrid similarity calculation results obtained through the above steps with the course mapping relationship established according to the course arrangement plan of the college syllabus, and finally realize the recommendation of the Top recommended MOOC courses and the pre-requisite and post-requisite MOOC courses associated with them.

[0045] Beneficial effects of the present invention:

[0046] (1) The intermediate fuzzy logic is introduced into the ML_BPNN neural network and the loss function is optimized to obtain the mixed score matrix.

[0047] (2) The global similarity between students is introduced so that the mixed similarity can capture the relationship between students in the dataset with close connections. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] Figure 1 , a flow chart of the method recommended by the present invention;

[0049] Figure 2 , is the ML_BPNN neural network model in the invention;

[0050] Figure 3 , recommend a process for prerequisite and subsequent courses in invention;

[0051] Figure 4 , a mixed recommendation model for the invention;

[0052] Figure 5 , is a recommended case. DETAILED DESCRIPTION

[0053] The present invention is further explained below in conjunction with the accompanying drawings and specific embodiments. It should be understood that the following specific embodiments are only used to illustrate the present invention and are not used to limit the scope of the present invention. It should be noted that the words "front", "rear", "left", "right", "upper" and "lower" used in the following description refer to directions in the accompanying drawings, and the words "inner" and "outer" refer to directions toward or away from the geometric center of a specific component, respectively.

[0054] like Figure 1-4 As shown, a hybrid recommendation method for MOOC courses based on fuzzy logic and teaching plan in this embodiment includes the following steps performed in sequence:

[0055] Step 1: Feature matrix construction: Based on the MOOC open course data, the intermediate fuzzy logic is introduced into the ML_BPNN neural network, the loss function is optimized, and finally a mixed score matrix containing MOOC course and online student characteristics is obtained.

[0056] First, the FCM (Fuzzy C-Means) clustering algorithm is used to cluster the student information and obtain the target student cluster. Let S = {s1, s2, ..., s m} is the set of student clusters after clustering, C = {c1, c2, ..., c n} is a set of courses. In the student cluster, an initial rating matrix is ​​constructed based on the students’ ratings of the courses. Then, the sentiment keywords in the course review information are counted, and the initial rating matrix is ​​weighted using the min-max normalization method (as shown in Formula 1) to obtain the student-course score matrix.

[0057]

[0058] Among them, X is the weighted value of student i’s review information on course j, and f ijis the keyword frequency of student i for course j, min is the minimum frequency, and max is the maximum frequency.

[0059]

[0060] Among them, r′ ij =r ij +(x ij +a) / (1+a), x ij is the weighted variable of student i’s review information on course j, a is the smoothing coefficient, r ij is the initial rating matrix of student i’s rating of course j, r′ ij is the weighted score of student i on course j.

[0061] For the student-course score matrix R, let the row vector set constitute the student matrix R*i (i∈{1,2,3,...,n}), and the column vector set constitute the course matrix R*j (j∈{1,2,3,...,m}). In the ML_BPNN neural network model, the student-course score matrix R as input is decomposed into student vectors R*i (i∈{1,2,3,...,n}) and course vectors R*j (j∈{1,2,3,...,n}), which are used as input vectors of two back propagation neural networks respectively. The SGD (Stochastic Gradient Descent) algorithm is used, and after passing through the middle hidden layer l i After training, the student vector pi and course vector qj are output respectively, and the cosine similarity method is used to calculate the score prediction value. The output of the ML_BPNN neural network is shown in Formula 3.

[0062]

[0063] Among them, u i =R*i(i∈{1,2,3,...,n}) represents the score of the i-th student on all courses, v j =R*j(j∈{1,2,3,...,n}) represents the ratings of all students on the jth course.

[0064] Let the implicit feedback vector C of student i for course j be ij =(C ij1 ,C ij2 ), explicit feedback vector E ij =(E ij1 ,E ij2 ). Explicit feedback and implicit feedback are defined as intermediary objects, and the loss function calculation formula of the ML_BPNN neural network is updated using the intermediary principle (as shown in Formula 4), and a mixed score matrix containing course and student characteristics is obtained.

[0065]

[0066] r α =min(||C ij -C im ||),m∈[1,n],||C ij -C im || is the Euclidean distance between implicit feedback data vectors;

[0067] r β =min(||E ij -E jm ||),m∈[1,n],||E ij -E jm || is the Euclidean distance between explicit feedback data vectors;

[0068] Where R is the set of courses, is the predicted value, R ij is the initial value, p∈[0,1], which is used to adjust the influence of implicit feedback and explicit feedback on the final result.

[0069] Step 2: Hybrid similarity calculation: The global similarity between online students is introduced into the point wise user pair factored heterogeneous similarity model (PoiU-FHSM). The hybrid similarity of the interaction relationship between online students and MOOC courses is composed of the predefined similarity calculated by Pearson similarity and the learning similarity calculated by the inner product of the latent vector.

[0070] Specifically:

[0071] Based on the Poi-FHSM model, learning similarity is introduced to represent the global similarity relationship between students. Let U and U' be the student vector matrix, V and W be the course vector matrix, and predefine the similarity. Reflect local similarity relationships by learning similarity Reflecting the global similarity relationship, it is extended to the PoiU-FHSM model. The student-student mixed similarity can obtain more relationships between students' courses and can capture the relationships in the dataset with close connections between students.

[0072]

[0073] Among them, U u· is the potential vector corresponding to student u in the student vector matrix, U′ u′· is the latent vector corresponding to student u' in the student vector matrix.

[0074] The Pearson similarity calculation method is used. The student-student predefined similarity calculation formula is shown in Formula 6, and the course-course predefined similarity calculation formula is shown in Formula 6.

[0075]

[0076] Among them, N(u,u') is the set of courses selected by students u and u'. u,v is the rating of student u for course v, R u′,v is the rating of student u' for course v, is the average rating of student u for all courses. is the average rating of student u' across all rated courses.

[0077] use As a normalization factor, the mixed similarity is normalized and the prediction rule is introduced from the course selection set C of student u. u Capturing student bias in u and course deviation b i To represent the student course preference, the updated hybrid similarity calculation formula is shown in Formula 7.

[0078]

[0079] The stochastic descent gradient (SGD) algorithm is used to update the model parameters, and the student course preference matrix is ​​obtained through the PoiU-FHSM model as shown in Formula 8.

[0080]

[0081] Among them, p ij represents the preference of student i for course j. The target student i’s preference for the course is P i* ={p i1 ,p i2 ,...,p m1}.

[0082] Step 3: Hybrid Course Recommendations:

[0083] Course labels are extracted from the teaching plan, and a bidirectional linked list is used to establish the precedence and postorder relationship between course labels. The pattern layer saves the course labels and the relationship between course labels. The order of course labels represents the prev preorder and next postorder of the corresponding course learning; the data layer saves all MOOC course information. A course label in the pattern layer corresponds to a MOOC course set in the data layer. The course label to which the MOOC course in the data layer belongs is determined by the label attribute.

[0084] Combine merge sort and insertion sort to sort students' preferences and obtain the top K courses as the TopK recommendation results.

[0085] For the Top1 MOOC course, obtain the course label label at the data layer; obtain 0-m pre-order labels prelabel and 0-n post-order labels postlabel of label at the model layer according to the prev and next attributes corresponding to the label label; return to the data layer, and find the corresponding 0-m pre-order course sets and 0-n post-order course sets according to the pre-order label and post-order label. After sorting the course sets obtained by the pre-order label and the course sets obtained by the post-order label according to step 32, obtain 1-k MOOC courses as the prerequisite MOOC courses and post-requisite MOOC courses of the Top1 course. Obtain recommendations for the Top recommended MOOC courses and their associated prerequisite and post-requisite MOOC courses.

[0086] experiment:

[0087] In order to verify the effectiveness of the course hybrid recommendation method, the present invention is experimented on the public ML100K dataset and the MOOC experimental dataset (as shown in Table 1).

[0088] Table 1 Experimental dataset

[0089]

[0090] The MOOC experimental dataset contains 1,000 student attribute information, 1,351 course attribute information, and 17,613 evaluation information.

[0091] Table 2 shows the recommendation algorithms with different loss functions. It can be observed from the table that as the k neighbor value gradually increases, the MAE value gradually stabilizes; under the same k neighbor value, the MAE value of the algorithm that introduces the loss function calculation of the intermediate fuzzy logic is smaller, and the recommendation effect is better than the algorithm that does not introduce the loss function calculation of the intermediate fuzzy logic.

[0092] Table 2 MAE values ​​with and without the introduction of intermediary fuzzy logic

[0093]

[0094] Table 3 shows the experimental results of the three hybrid similarity algorithms. It can be observed from the table that, when the nearest neighbor K is determined, for the public ML100K dataset, the PoiU-FHSM algorithm proposed in this paper does not improve the accuracy and recall rate; for the MOOC experimental dataset, PoiU-FHSM improves the recommendation accuracy by 7.43% and 5.17% compared with Poi-FHSM and Poil-FHSM, and the recall rate by 4.12% and 2.45%. The reason is that the connection between items in the public ML100K dataset is relatively close. When the user-to-user similarity is added, the overall algorithm performance is not improved. However, for the MOOC experimental dataset, the connection between students is closer than that between courses. Considering the global similarity, introducing the similarity between students can improve the recommendation effect.

[0095] Table 3 Comparison of experimental results of PoiU-FHSM hybrid similarity algorithm

[0096]

[0097]

[0098] Poi-FHSM is the original model algorithm, Poil-FHSM is a model algorithm that uses user2user similarity instead of item2item similarity for calculation, and PoiU-FHSM is a model algorithm proposed by the present invention that adds a user2user algorithm to calculate similarity based on the initial similarity calculation method.

[0099] Case verification:

[0100] In order to verify the rationality and effectiveness of the course hybrid recommendation method proposed in this paper, two real recommendation cases ( Figure 5 The proposed course mixing recommendation method is tested in (as shown).

[0101] The top 1 recommended MOOC course for learner ID 38 is “Signals and Systems”, the recommended prerequisite courses for this course are “Advanced Mathematics” and “Analog Electronic Technology”, and the recommended post-requisite courses for this course are “Digital Signal Processing” and “Principles of Automatic Control”.

[0102] The top 1 recommended MOOC course for learner ID 38 is “Principles of Communications”, the recommended prerequisite courses for this course are “Digital Signal Processing” and “Integrated Circuits”, and the recommended post-requisite courses for this course are “Data Communications” and “Mobile Communications”.

[0103] The technical means disclosed in the scheme of the present invention are not limited to the technical means disclosed in the above-mentioned implementation mode, but also include technical schemes composed of any combination of the above technical features.

Claims

1. A hybrid recommendation method for MOOC courses based on fuzzy logic and teaching plan, characterized in that: The steps include the following steps which are performed in the following order: Step 1. Based on the MOOC open course data, the intermediate fuzzy logic is introduced into the ML_BPNN neural network to optimize the loss function, and finally a mixed score matrix containing the characteristics of MOOC courses and online students is obtained; Step 2. The global similarity between online students is introduced into the point-wise user decomposition hybrid similarity model. The hybrid similarity of the interaction relationship between online students and MOOC courses is composed of the predefined similarity calculated by Pearson similarity and the learning similarity calculated by the inner product of the latent vector. Step 2 is as follows: Step 21. Based on the Poi-FHSM model, learning similarity is introduced to represent the global similarity relationship between students. Let U and U' be the student vector matrix, V and W be the course vector matrix, and predefine the similarity Reflect local similarity relationships by learning similarity Reflecting the global similarity relationship, it is extended to the PoiU-FHSM model; the student-student mixed similarity can obtain more relationships between students' courses and can capture the relationships in the data set with close connections between students; the formula is shown in Formula 5: Among them, U u· is the potential vector corresponding to student u in the student vector matrix, U′ u′. is the potential vector corresponding to student u' in the student vector matrix; The Pearson similarity calculation method is used. The student-student predefined similarity calculation formula is shown in Formula 6, and the course-course predefined similarity calculation formula is shown in Formula 6; Where N(u,u') is the set of courses selected by students u and u'; R u,v is the rating of student u for course v, R u′,v is the rating of student u' for course v, is the average rating of all courses given by student u; is the average rating of student u' for all rated courses; Step 22. Based on Formula 5, use As a normalization factor, the mixed similarity is normalized and the prediction rule is introduced from the course selection set C of student u. u Capturing student bias in u and course deviation b i To represent the student course preference, the updated hybrid similarity calculation formula is shown in Formula 7; Step 23. Use the stochastic descent gradient algorithm to update the model parameters, and obtain the student course preference matrix through the PoiU-FHSM model as shown in Formula 8; Among them, p ij represents the preference of student i for course j; the preference of target student i for course is P i* ={p i1 ,p i2 ,...,p m1 }; Step 3. Combine the PoiU-FHSM hybrid similarity calculation results with the course mapping relationship established according to the course schedule plan of the college syllabus, and finally output the top recommended MOOC courses and their associated prerequisite and post-recommended MOOC courses.

2. The MOOC course hybrid recommendation method based on fuzzy logic and teaching plan according to claim 1, characterized in that: Step 1 is as follows: Step 11. Use the FCM clustering algorithm to cluster the online student information and obtain the target student cluster. Let S = {s1, s2, ..., s m } is the set of student clusters after clustering, C = {c1, c2, ..., c n } is a set of courses; within the student cluster, an initial rating matrix is ​​constructed based on the students’ ratings of the courses, where r ij is the initial score of student i on course j, m is the number of students in the student cluster where the target student belongs, and n is the number of courses; Step 12. By counting the sentiment keywords in the MOOC course review information, the min-max normalization method is used; as shown in Formula 1; the initial rating matrix of step 11 is weighted. Among them, x ij is the weighted value of student i’s review information on course j, f ij is the keyword frequency of student i for course j, min is the minimum frequency, and max is the maximum frequency; Finally, we can get the student-course score matrix; as shown in Formula 2: Among them, r′ ij =r ij +(x ij +a) / (1+a), x ij is the weighted variable of student i’s review information on course j, a is the smoothing coefficient, r ij is the initial rating matrix of student i’s rating of course j, r′ ij is the weighted score of student i on course j; Step 13. For the student-course score matrix R, let the row vector set constitute the student matrix R*i, i∈{1,2,3,...,n}, and the column vector set constitute the course matrix R*j, j∈{1,2,3,...,m}; In the ML_BPNN neural network model, the student-course score matrix R as the input is decomposed into student vectors R*i, i∈{1,2,3,...,n} and course vectors R*jj∈{1,2,3,...,n}, which are used as the input vectors of two back-propagation neural networks respectively. The SGD algorithm is used to pass through the middle hidden layer l i After training, the student vector pi and course vector qj are output respectively, and the cosine similarity method is used to calculate the score prediction value. The output of the ML_BPNN neural network is shown in Formula 3; Among them, u i =R*i, i∈{1,2,3,...,n} represents the score of the i-th student on all courses, v j =R*j, j∈{1,2,3,...,n} represents the ratings of all students for the jth course; p i is the output of student vector training, q j is the output result of the course vector training, is the final student i's score prediction training result for course j, by calculating p i and q j The cosine similarity is obtained; Step 14. Let the implicit feedback vector C of student i for course j be ij =(C ij1 ,C ij2 ), explicit feedback vector E ij =(E ij1 ,E ij2 ); define explicit feedback and implicit feedback as intermediary objects, and use the intermediary principle to update the loss function calculation formula of the ML_BPNN neural network in step 13; the loss function calculation formula of the fusion intermediary fuzzy logic is shown in Formula 4: r α =min(||C ij -C im ||),m∈[1,n],||C ij -C im || is the Euclidean distance between implicit feedback data vectors; r β =min(||E ij -E jm ||),m∈[1,n],||E ij -E jm || is the Euclidean distance between explicit feedback data vectors; Where R is the set of courses, is the predicted value, R ij is the initial value, p∈[0,1], which is used to adjust the influence of implicit feedback and explicit feedback on the final result; According to the above method, we can finally get a mixed score matrix containing MOOC course and online student characteristics.

3. According to the MOOC course hybrid recommendation method based on fuzzy logic and teaching plan according to claim 1, step 3 is specifically: Step 31. Extract course labels from the teaching plans of colleges and universities, and use a bidirectional chain structure to establish the order and sequence relationship between course labels. The model layer saves the course labels and the course label relationship. The course label sequence represents the prev order and next order of the corresponding course learning. The data layer saves all MOOC course information. A course label in the model layer corresponds to a MOOC course set in the data layer. The course label to which the MOOC course in the data layer belongs is determined by the label attribute. Step 32. Based on step 23, merge sort and insertion sort are combined to sort the students' preferences, and the top K courses are obtained as the MOOC course TopK recommendation results; Step 33. Based on step 31 and step 32, for the Top1 MOOC course, obtain the course label label at the data layer; obtain 0-m pre-order labels prelabel and 0-n post-order labels postlabel of label1 at the model layer according to the prev and next attributes corresponding to the label label; return to the data layer, and search for the corresponding 0-m pre-order course sets and 0-n post-order course sets according to the pre-order label and post-order label; sort the course sets obtained by the pre-order label and the course sets obtained by the post-order label according to step 32, and obtain 1-k MOOC courses as the pre-requisite MOOC courses and post-requisite MOOC courses of the Top1 course; The PoiU-FHSM hybrid similarity calculation results obtained through the above steps are combined with the course mapping relationship established according to the course arrangement plan of the college syllabus, and finally the recommendation of the top recommended MOOC courses and their associated prerequisite and subsequent MOOC courses can be achieved.

Images