CN112749342A - Personalized recommendation method for network education and teaching resources - Google Patents
Personalized recommendation method for network education and teaching resources Download PDFInfo
- Publication number
- CN112749342A CN112749342A CN202110088311.8A CN202110088311A CN112749342A CN 112749342 A CN112749342 A CN 112749342A CN 202110088311 A CN202110088311 A CN 202110088311A CN 112749342 A CN112749342 A CN 112749342A
- Authority
- CN
- China
- Prior art keywords
- educational
- user
- representing
- resources
- matrix
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 238000000034 method Methods 0.000 title claims abstract description 29
- 238000001914 filtration Methods 0.000 claims abstract description 30
- 238000004364 calculation method Methods 0.000 claims abstract description 8
- 239000011159 matrix material Substances 0.000 claims description 67
- 239000013598 vector Substances 0.000 claims description 43
- 230000017105 transposition Effects 0.000 claims description 5
- 238000000354 decomposition reaction Methods 0.000 claims description 3
- NAWXUBYGYWOOIX-SFHVURJKSA-N (2s)-2-[[4-[2-(2,4-diaminoquinazolin-6-yl)ethyl]benzoyl]amino]-4-methylidenepentanedioic acid Chemical compound C1=CC2=NC(N)=NC(N)=C2C=C1CCC1=CC=C(C(=O)N[C@@H](CC(=C)C(O)=O)C(O)=O)C=C1 NAWXUBYGYWOOIX-SFHVURJKSA-N 0.000 claims description 2
- 230000003247 decreasing effect Effects 0.000 claims description 2
- 230000009467 reduction Effects 0.000 claims description 2
- 238000007405 data analysis Methods 0.000 abstract description 2
- 230000006870 function Effects 0.000 description 10
- 230000006399 behavior Effects 0.000 description 3
- 238000011156 evaluation Methods 0.000 description 2
- 238000010801 machine learning Methods 0.000 description 2
- 238000005457 optimization Methods 0.000 description 2
- 238000012163 sequencing technique Methods 0.000 description 2
- 230000002195 synergetic effect Effects 0.000 description 2
- 230000003542 behavioural effect Effects 0.000 description 1
- 230000000295 complement effect Effects 0.000 description 1
- 238000010276 construction Methods 0.000 description 1
- 238000002790 cross-validation Methods 0.000 description 1
- 238000011157 data evaluation Methods 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 238000002474 experimental method Methods 0.000 description 1
- 230000003993 interaction Effects 0.000 description 1
- 238000010606 normalization Methods 0.000 description 1
- 230000008569 process Effects 0.000 description 1
- 238000012549 training Methods 0.000 description 1
- 238000010200 validation analysis Methods 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/90—Details of database functions independent of the retrieved data types
- G06F16/95—Retrieval from the web
- G06F16/953—Querying, e.g. by the use of web search engines
- G06F16/9535—Search customisation based on user profiles and personalisation
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/90—Details of database functions independent of the retrieved data types
- G06F16/95—Retrieval from the web
- G06F16/953—Querying, e.g. by the use of web search engines
- G06F16/9536—Search customisation based on social or collaborative filtering
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F17/00—Digital computing or data processing equipment or methods, specially adapted for specific functions
- G06F17/10—Complex mathematical operations
- G06F17/16—Matrix or vector computation, e.g. matrix-matrix or matrix-vector multiplication, matrix factorization
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/22—Matching criteria, e.g. proximity measures
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Databases & Information Systems (AREA)
- Data Mining & Analysis (AREA)
- General Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- Mathematical Physics (AREA)
- Pure & Applied Mathematics (AREA)
- Computational Mathematics (AREA)
- Mathematical Optimization (AREA)
- Mathematical Analysis (AREA)
- Bioinformatics & Computational Biology (AREA)
- Evolutionary Computation (AREA)
- Computing Systems (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Artificial Intelligence (AREA)
- Algebra (AREA)
- Life Sciences & Earth Sciences (AREA)
- Software Systems (AREA)
- Electrically Operated Instructional Devices (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
Abstract
A personalized recommendation method for network education and teaching resources belongs to the technical field of computer internet, and provides a collaborative filtering recommendation algorithm combining project similarity and alternate least square collaborative filtering based on a Spark platform, so that the prediction calculation efficiency can be improved, and the system response time can be reduced. In order to solve the problem that the appropriate network education and teaching resources cannot be accurately recommended to different users due to inaccurate models caused by data sparsity in the conventional collaborative filtering recommendation scheme, the invention optimizes the alternative least square collaborative filtering recommendation algorithm and uses the algorithm on a Spark big data analysis platform, so that the workload completed in unit time and the recommendation precision are increased by using a parallel method, and the problem that the network education and teaching resources cannot be accurately recommended is solved.
Description
Technical Field
The invention belongs to the technical field of computer internet, and relates to a personalized recommendation method aiming at the aspect of network education and teaching resources.
Background
With the continuous progress of the human informatization level, information infrastructure, information technology, software systems and platforms have developed today to become revolutionary power which promotes the development of education and teaching level to higher levels and social progress. After the thought idea of E-education is provided, various network education and teaching platforms gradually appear, particularly, in the current year, due to epidemic situations, offline education institutions are basically in a state of being incapable of normal business, however, the learning requirements of students still exist, and the advantages of online education and teaching enter the public sight at once. Under the background that various education and teaching platforms appear in succession, network education and teaching resources are also increased at a rapid speed, but under the condition that the amount of the education and teaching resources is increased sharply, a plurality of troubles and challenges are brought to teaching organizers and learners, and the teaching organizers and the learners have to spend a great deal of time and energy to screen out the education and teaching resources meeting the demands of the teaching organizers and the learners. Therefore, the personalized recommendation system widely applied to the business field is also gradually applied to the education and teaching field, personalized calculation can be performed by using the historical behavior data of the user, interest points of different users can be found, and the users are guided to gradually find information and education and teaching resources required by the users, so that the working efficiency and the learning efficiency of the users are improved to a great extent.
The recommendation algorithm is a method capable of filtering information, can effectively recommend personalized information according to information requirements, personal interests and the like of users, and has been successfully applied to a plurality of fields such as online videos, social platforms, online music, electronic commerce and the like. Along with the continuous perfection of education and teaching resources in intelligent education and teaching construction, abundant education and teaching resources such as electronic books, teaching videos, documents and the like are utilized to carry out personalized recommendation, and the learning efficiency of students is improved.
Currently, a popular recommendation algorithm is Collaborative filtering recommendation (CF for short), and Collaborative filtering is understood literally to include two operations of collaboration and filtering. The synergy means that the group behaviors are used for recommendation, and the group gradually evolves to a better state through the synergy, which is a saying that the organisms have synergistic evolution. For a recommendation system, the recommendation finally given to the user is more and more accurate through the continuous synergistic action of the user. And filtering, namely finding (filtering) the scheme preferred by the user from the feasible decision (recommendation) scheme. Specifically, the idea of collaborative filtering is to find some similarity (similarity between users or similarity between objects) through the behavior of a group, and to make decisions and recommendations for users through the similarity. In general, collaborative filtering recommendations fall into three categories. The first is user-based collaborative filtering, the second is item-based collaborative filtering, and the third is model-based collaborative filtering. Based on the user-based collaborative filtering, the main consideration is the similarity between users, and as long as the educational teaching resources liked by similar users are found out and the scores of the target users for the corresponding educational teaching resources are predicted, a plurality of educational teaching resources with the highest scores can be found out and recommended to the users. And project-based collaborative filtering is similar to user-based collaborative filtering, except that at this time, the similarity between the educational teaching resources and the educational teaching resources is found, and only if the scores of some educational teaching resources by the target user are found, the similar educational teaching resources with high similarity can be predicted, and a plurality of similar educational teaching resources with the highest scores are recommended to the user. For example, if you buy a book related to machine learning on the internet, the website will recommend a pile of machine learning immediately, and the book related to big data is given to you, so that the project-based collaborative filtering idea is obviously used.
However, the precision of many current recommendation algorithms is low, and if a collaborative filtering method is directly used, the problem that the association degree of the recommended result and the prepared course task is low when the recommended education and teaching resources are outdated due to the lack of understanding and knowledge of course contents is caused, so that a personalized recommendation method for recommending relevant education and teaching resources as accurately as possible needs to be constructed.
Disclosure of Invention
The invention provides a personalized recommendation method for network education and teaching resources. In real teaching, it is not possible for every user to have a behavioral relationship with all teaching resources. In fact, the user-resource pairs with interactions are only a small fraction. In other words, the user-resource relationship list is very sparse. If the educational teaching resource scoring matrix is very sparse, the accuracy of the model can be directly influenced. The invention aims to provide a teaching resource recommendation method which is suitable for an ultra-large sparse feature matrix and has better calculation efficiency for a large-scale data set environment. The problem that due to the fact that a model is inaccurate due to data sparsity in an existing collaborative filtering scheme, appropriate education and teaching resources cannot be accurately recommended to different users is solved.
According to the invention, an Alternating Least Square (ALS) algorithm is used on a Spark big data analysis platform, so that the workload completed in unit time and the recommendation precision are increased by a parallel method, and the problem that education and teaching resources cannot be accurately recommended is solved.
A method for personalized recommendation of network education and teaching resources, comprising:
the method comprises the following steps: firstly, a scoring matrix of a user is obtained through user scoring original data, and similarity between the user and between educational and teaching resources and educational and teaching resources is calculated.
Step two: and designing a loss function, and adding the similarity data obtained by the calculation into the corresponding function for further use.
Step three: and (5) performing iterative solution on the minimization function in the function of the step two, and finally obtaining a U matrix and a V matrix.
Drawings
FIG. 1 is a matrix decomposition model diagram
FIG. 2 iterative determination of UV map
FIG. 3 is a flow chart of an alternate least squares recommendation algorithm
FIG. 4MSE algorithm comparison graph
FIG. 5RMSE Algorithm comparison map
Detailed Description
The invention discloses an alternating least square collaborative filtering recommendation algorithm which is realized on a Spark platform, and the basic principle is as follows: if R is a user's tutorialScoring matrix of education resources by normalization to (R ═ Rm×n))∈{0,1}m×nWherein m represents the number of users and n represents the number of scoring matrices. This algorithm requires finding a low rank matrix X to approximate the matrix R and minimizing the objective function as follows:
L(x)＝∑ij(Rij-Xij)2 (1)
in the formula (1), RijDenotes a score, X, of i on the abscissa and j on the ordinate in the matrix RijRepresents the value of i on the abscissa and j on the ordinate in the low rank matrix X, (R)ij-Xij)2The squared error terms of the R matrix and the X matrix are represented. Then consider how to efficiently solve the optimization problem argminx L(x)。
As shown in fig. 1, the matrix decomposition model X ═ UVTWhere d represents the number of features of the R matrix, R represents the rank of the matrix R, RijExpressing the scores of the users i on the education and teaching resources j, U expressing the preference matrix of the users on the implicit characteristics of the education and teaching resources, UiAn implicit feature vector representing the preferences of user i, V represents a matrix of implicit features contained in the educational resources, VjRepresenting implicit feature vectors, V, contained in educational resources jTRepresenting the transpose of the matrix V, in general d is smaller than one thousandth of r, r ≈ min (m, n), i.e. equivalent to dimensionality reduction. At this time, the formula (1) can be rewritten as
L(U,V)＝∑ij(Rij-UiVj T)2 (2)
The loss function generally needs to add a regularization term to avoid the problems of overfitting and the like, and the overfitting can be prevented by adding a second-order regularization term to the formula (2), so that the formula (2) can be rewritten into
In equation (3), λ represents the coefficient of the second-order regularization term, and is generally usedThe parameter is configured by a validation set or a cross validation set, various data are tried, the best parameter is found, and the parameter is set to be a smaller value by considering the balance between training sets, so that overfitting can be avoided. Initially setting lambda to be 0.1, using a learning rate to be 0.01, then fixing the learning rate, centering the lambda on an initial value, increasing the lambda by taking 0.01 as a step length, iterating for ten times, taking the lambda obtained by the previous iteration as a reference value and increasing the lambda by taking 0.01 as the step length for each iteration, setting the lambda to be 0.1 when the lambda reaches 0.2, iterating for ten times again, taking the lambda obtained by the previous iteration as the reference value and reducing the lambda by taking 0.01 as the step length, and when the lambda is 0, selecting the lambda with the minimum error in the 20 times as the lambda to be the lambda1Then, the learning rate is reduced to 0.001, and the λ obtained in the above step is further finely adjusted to obtain the λ obtained in the previous step1Increasing and decreasing by 0.001 step size as reference value, and selecting the λ with minimum error in the twenty times after the same 20 iterations1The resultant is λ (the physical meaning of λ in the following equation is the same as equation (3)). By this time, collaborative filtering has successfully translated into an optimization problem due to the variable UiAnd VjCoupled together, the solution is not easy, so we introduce an alternating least squares method to solve, assuming that V matrix is known, we can apply Ridge Regression to extrapolate each row of the U matrix, and vice versa. Thus, fixing the V matrix, the penalty function L (U, V) is coupled to UiThe partial derivative is calculated and made equal to 0, the following solution U is obtainediFormula (2)
R in the formula (4)iIndicating i the scoring vector that the user has scored,
representation by user uiA feature matrix composed of the scored feature vectors, wherein m represents the number of users, n represents the number of scoring matrices,
indicating the amount of educational resources that user i has scored.
Similarly, the U matrix is fixed, and the following solution V can be obtainedjFormula (2)
R in the formula (5)jThe scoring vector of the scored users of the education and teaching resource j is represented, m represents the number of the users, n represents the number of scoring matrixes,
expressed as educational resources vjA feature matrix composed of feature vectors of the scored users,
expressed as educational resources vjThe number of users evaluated in the past, I in equations (4) and (5) represents a d × d identity matrix.
Based on the minimum cross-over two-multiplication of the collaborative filtering algorithm, the calculation is completed by calling the formulas (4) and (5) to calculate the algorithm until a convergence result appears, and the iteration process is shown in fig. 2.
The flow of the alternating least squares collaborative filtering recommendation algorithm is shown in fig. 3.
Step 1, firstly, obtaining a scoring matrix of a user through original data scored by the user, and calculating the similarity between the user and between educational and teaching resources and educational and teaching resources. The invention applies the vector cosine method, also called VSS (spatial similarity method), where N is(m)Represents the set of educational resources owned by user m, i.e., the promotion of educational resources by user m, N(n)Representing the educational and teaching resources possessed by the user n, the similarity formula between the users m and n is:
according to the same principle, the similarity formula between the educational teaching resource i and the educational teaching resource j is as follows:
and 2, designing a loss function, and adding the similarity data obtained by the calculation into the function of the formula (8).
In the above formula, (i, j) represents all the user-education resource pairs, rijRepresents the user i's score for the educational teaching resource j,
representing transposes of user matrices, vjRepresenting implicit feature vectors, I, contained in educational resources, jiRepresents a set of users, ukiRepresenting the user i's preference for implicit characteristics k of educational and educational resources, upImplicit feature vectors, u, representing the preferences of the user pkpExpressing the preference of the user p to implicit characteristics K of the educational and teaching resources, PC (m, n) expressing the similarity between the user and the user, K (u)i) Representing user uiN nearest neighbors of, IjRepresents a set of educational and teaching resources, vkjRepresenting implicit characteristic vectors k, v contained in educational and teaching resources jqRepresenting implicit feature vectors, v, contained in educational and teaching resources qkqRepresenting an implicit characteristic vector K contained in an educational and teaching resource q, wherein K displays an attribute, K (v)j) Representing educational resources vjThe N nearest neighbors of (c), PC (i, j) represents the similarity between the educational resources and the educational resources,
representing user uiScored educational and teaching materialsThe number of sources is such that,
expressed as educational resources vjThe number of users scored early.
And 3, performing iterative solution on the minimization function in the step 2, and finally obtaining a U matrix and a V matrix.
And (3) exactly solving U and V, solving V according to the fixed U, solving U by the fixed V, and iteratively solving by using an alternating least square method, wherein the specific steps are as follows:
in the above formula (9), IiA set of users is represented as a set of users,
transposition of implicit feature vectors, v, representing preferences of user ijRepresenting implicit feature vectors, r, contained in educational resources, jijRepresents the score v of the user i on the educational and teaching resource jkjRepresenting implicit characteristic vectors k, u contained in educational and teaching resources jkiRepresenting the user i's preference for implicit characteristics k of educational and educational resources, upImplicit feature vectors, u, representing the preferences of the user pkpExpressing the preference of the user p for implicit characteristics K of educational and teaching resources, PC (m, n) expressing the similarity between the user m and the user n, K (u)i) Representing user uiN nearest neighbors, k represents a random attribute, E represents an identity matrix, IjRepresenting a set of educational and teaching resources, RTA transpose of a scoring matrix representing a user's scoring of educational teaching resources,
representing the transpose of the matrix formed by the educational teaching resource vectors,
the matrix formed by the vector of the user set is represented, and u can be obtained through iterative solutioni。
Same as for vjTherein, there are
In the above formula (10), IiA set of users is represented as a set of users,
representing a matrix formed by vectors of a set of users, vjRepresenting the implicit feature vectors contained in the educational and educational resources j,
expressed as educational resources vjThe number of users scored early on,
expressing a matrix formed by vectors of the educational teaching resources, PC (i, j) expressing the similarity between the educational teaching resources i and j, vqRepresenting implicit feature vectors, K (v), contained in educational and teaching resources qj) Representing educational resources vjN nearest neighbors, k represents a random attribute, E represents an identity matrix, IjRepresenting a set of educational and teaching resources, RTAnd (3) representing the transposition of the scoring matrix of the educational and teaching resources by the user.
Experiment and results
To evaluate the accuracy of the proposed algorithm, we used the following evaluation indices: mean Square Error (MSE), Root Mean Square Error (RMSE), global average accuracy (MAP). MSE means the variation of the vibration range of the data evaluation, and if the range is small, the correctness of the drawn up model represents good. RMSE serves as the standard deviation. The MAP is ranking based on the evaluation index, and the higher value indicates that the accuracy of the recommended teaching resources is higher, and the sequencing of the teaching resources can still enable the user to be satisfied. The exact definitions are each:
The data are the results obtained for 200 users tested by the teaching resource recommendation method. Because the number of users is not large, the data is relatively small, and random errors exist, but the data obtained by the method can objectively show the effect of the algorithm. Compared to the algorithms presented by the present invention are content-based recommendation algorithms and association rule-based recommendation algorithms.
The results of the MSE algorithm are shown in FIG. 4, and it can be seen from the results of FIG. 4 that the proposed algorithm used in the present invention is more accurate than the two compared algorithms. The RMSE algorithm results are shown in fig. 5. It can be seen from fig. 5 that the algorithm used in the present invention greatly surpasses the algorithm to which it is compared. The MSE algorithm represents the accuracy of the algorithm provided by the invention, the RMSE algorithm can represent the sequencing of the recommendation results, and the two algorithms are complementary to meet the requirements of the system, so that the algorithm provided by the invention can be applied to a teaching resource recommendation system.
In summary, the invention is based on the application of the alternative least square collaborative filtering recommendation algorithm in the teaching resource recommendation system. Firstly, the principle of the alternative least square recommendation algorithm is analyzed and the situation of the problem is solved, so that a recommendation method which respectively calculates the similarity between users and the similarity between teaching resources on the basis of the traditional collaborative filtering recommendation algorithm and simultaneously applies the alternative least square collaborative filtering recommendation algorithm to a Spark platform so as to increase the workload finished in unit time and the recommendation precision by a parallel method is provided.
Claims (3)
1. A personalized recommendation method aiming at network education and teaching resources is characterized in that:
r is a scoring matrix of the educational and educational resources of the user, and is normalized to (R ═ Rm×n))∈{0,1}m×nWherein m represents the number of users, n represents the number of educational education scoring matrixes, a low-rank matrix X is searched to approximate a matrix R, and the following objective function is minimized:
L(x)＝∑ij(Rij-Xij)2 (1)
in the formula (1), RijRepresents the grade of the user i on the educational and teaching resource j, (R)ij-Xij)2A square error term representing the R matrix and the X matrix;
for matrix decomposition model X ═ UVTD represents the number of the characteristics of the R matrix, R represents the rank of the matrix R, U represents the preference matrix of the user for the implicit characteristics of the educational and teaching resources, and U represents the preference matrix of the user for the implicit characteristics of the educational and teaching resourcesiAn implicit feature vector representing the preferences of user i, V represents a matrix of implicit features contained in the educational resources, VjRepresenting implicit feature vectors, V, contained in educational resources jTRepresenting the transposition of a matrix V, wherein d is less than one thousandth of r, and r is approximately equal to min (m, n), namely equivalent to dimensionality reduction;
at this time, formula (1) is rewritten as
L(U,V)＝∑ij(Rij-UiVj T)2 (2)
The loss function generally needs to add a regularization term to avoid the problems of overfitting and the like, and after a second-order regularization term is added to the formula (2), overfitting can be prevented, and then the formula (2) is rewritten into
In the formula (3), lambda represents a coefficient of a second-order regularization term, an alternating least square method is introduced to solve, and assuming that a V matrix is known, each row of the U matrix can be estimated by applying Ridge Regression (Ridge Regression), and vice versa; thus, fixing the V matrix, the penalty function L (U, V) is coupled to UiThe partial derivative is calculated and made equal to 0, the following solution U is obtainediFormula (2)
R in the formula (4)iScore vector, V, representing i user scoreduiRepresenting a feature matrix consisting of feature vectors scored by users i, m representing the number of users, n representing the number of scoring matrices, nuiRepresenting the number of the educational and teaching resources scored by the user i;
similarly, the U matrix is fixed, and the following solution V is obtainedjFormula (2)
R in the formula (5)jThe scoring vector of the scored users of the education and teaching resource j is represented, m represents the number of the users, n represents the number of scoring matrixes,
a feature matrix composed of feature vectors of users scored for educational teaching resources j,
j early expressed as educational and teaching resourceThe number of users rated, I represents a unit matrix of d × d in equations (4) and (5);
based on the minimum cross double multiplication of the collaborative filtering algorithm, the formulas (4) and (5) are called to calculate the algorithm until the convergence result is generated, and then the calculation is completed;
the flow of the alternative least square collaborative filtering recommendation algorithm is as follows;
step 1, firstly, obtaining a scoring matrix of a user through original data scored by the user, and calculating the similarity between the user and between educational and teaching resources and educational and teaching resources; using vector cosine method, in which N(m)Represents the set of educational resources owned by user m, i.e., the promotion of educational resources by user m, N(n)Representing the educational and teaching resources possessed by the user n, the similarity formula between the users m and n is:
the same principle is known that the similarity formula between the education resources i and the education resources j is as follows:
step 2, adding the similarity data obtained by the calculation into a loss function of the formula (8);
in the above formula, (i, j) represents all the user-education resource pairs, rijRepresents the user i's score for the educational teaching resource j,
representing transposes of user matrices, vjImplicit features representing the inclusion of educational and educational resources jEigenvectors, IiRepresents a set of users, ukiRepresenting user i's preference u for implicit characteristics k of educational and educational resourcespImplicit feature vectors, u, representing the preferences of the user pkpExpressing the preference of the user p to implicit characteristics K of the educational and teaching resources, PC (m, n) expressing the similarity between the user and the user, K (u)i) Representing user uiN nearest neighbors of, IjRepresents a set of educational and teaching resources, vkjRepresenting implicit characteristic vectors k, v contained in educational and teaching resources jqRepresenting implicit feature vectors, v, contained in educational and teaching resources qkqRepresenting an implicit characteristic vector K contained in an educational and teaching resource q, wherein K displays an attribute, K (v)j) Representing educational resources vjThe N nearest neighbors of (c), PC (i, j) represents the similarity between the educational resources and the educational resources,
representing user uiThe amount of educational teaching resources that have been scored,
expressed as educational resources vjThe number of users rated early;
and 3, performing iterative solution on the minimization function in the step 2, and finally obtaining a U matrix and a V matrix.
2. The method of claim 1, wherein: initially setting lambda to be 0.1, centering lambda on the initial value, increasing the step length by 0.01, iterating for ten times, setting lambda to be 0.1 when lambda reaches 0.2, iterating for ten times again, reducing by taking 0.01 as the step length, and selecting the lambda with the minimum error in the 20 times as lambda to be 01Then with the obtained lambda1Increasing and decreasing according to step length of 0.001 as reference value, and selecting λ with minimum error in twenty times after 20 iterations1As the final desired lambda.
3. The method of claim 1, wherein the first and second light sources are selected from the group consisting of,the method is characterized in that: and (3) exactly solving U and V, solving V according to the fixed U, solving U by the fixed V, and iteratively solving by using an alternating least square method, wherein the specific steps are as follows:
in the above formula (9), IiA set of users is represented as a set of users,
transposition of implicit feature vectors, v, representing preferences of user ijRepresenting implicit feature vectors, r, contained in educational resources, jijRepresents the score v of the user i on the educational and teaching resource jkjRepresenting implicit characteristic vectors k, u contained in educational and teaching resources jkiRepresenting the user i's preference for implicit characteristics k of educational and educational resources, upImplicit feature vectors, u, representing the preferences of the user pkpExpressing the preference of the user p for implicit characteristics K of educational and teaching resources, PC (m, n) expressing the similarity between the user m and the user n, K (u)i) Representing user uiN nearest neighbors, k represents a random attribute, E represents an identity matrix, IjRepresenting a set of educational and teaching resources, RTA transpose of a scoring matrix representing a user's scoring of educational teaching resources,
representing the transpose of the matrix formed by the educational teaching resource vectors,
the matrix formed by the vector of the user set is represented, and u can be obtained through iterative solutioniSame as for vjTherein, there are
In the above formula (10), IiA set of users is represented as a set of users,
representing a matrix formed by vectors of a set of users, vjRepresenting the implicit feature vectors contained in the educational and educational resources j,
expressed as educational resources vjThe number of users scored early on,
expressing a matrix formed by vectors of the educational teaching resources, PC (i, j) expressing the similarity between the educational teaching resources i and j, vqRepresenting implicit feature vectors, K (v), contained in educational and teaching resources qj) Representing educational resources vjN nearest neighbors, k represents a random attribute, E represents an identity matrix, IjRepresenting a set of educational and teaching resources, RTAnd (3) representing the transposition of the scoring matrix of the educational and teaching resources by the user.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110088311.8A CN112749342A (en) | 2021-01-20 | 2021-01-20 | Personalized recommendation method for network education and teaching resources |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110088311.8A CN112749342A (en) | 2021-01-20 | 2021-01-20 | Personalized recommendation method for network education and teaching resources |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN112749342A true CN112749342A (en) | 2021-05-04 |
Family
ID=75652922
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202110088311.8A Pending CN112749342A (en) | 2021-01-20 | 2021-01-20 | Personalized recommendation method for network education and teaching resources |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN112749342A (en) |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20100185579A1 (en) * | 2009-01-22 | 2010-07-22 | Kwang Seok Hong | User-based collaborative filtering recommendation system and method for amending similarity using information entropy |
| CN110929161A (en) * | 2019-12-02 | 2020-03-27 | 南京莱斯网信技术研究院有限公司 | Large-scale user-oriented personalized teaching resource recommendation method |
-
2021
- 2021-01-20 CN CN202110088311.8A patent/CN112749342A/en active Pending
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20100185579A1 (en) * | 2009-01-22 | 2010-07-22 | Kwang Seok Hong | User-based collaborative filtering recommendation system and method for amending similarity using information entropy |
| CN110929161A (en) * | 2019-12-02 | 2020-03-27 | 南京莱斯网信技术研究院有限公司 | Large-scale user-oriented personalized teaching resource recommendation method |
Non-Patent Citations (2)
| Title |
|---|
| MOHAMMED FADHEL ALJUNID 等: "Movie Recommender System Based on Collaborative Filtering Using Apache Spark", SPRINGER NATURE SINGAPORE PTE LTD. 2019, pages 283 - 295 * |
| 迟玉良 等: "项目相似度与 ALS 结合的推荐算法研究", 软件导刊, vol. 17, no. 6, pages 81 - 84 * |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Salehi et al. | Hybrid attribute-based recommender system for learning material using genetic algorithm and a multidimensional information model | |
| Chang et al. | Using groups of items for preference elicitation in recommender systems | |
| CN111242310B (en) | Feature validity evaluation method and device, electronic equipment and storage medium | |
| CN111079028A (en) | Collaborative filtering recommendation system and method based on multi-source auxiliary information | |
| JP2017199355A (en) | Recommendation generation | |
| Kokkodis et al. | The utility of skills in online labor markets | |
| Hu et al. | Learning peer recommendation using attention-driven CNN with interaction tripartite graph | |
| Zhong et al. | Design of a personalized recommendation system for learning resources based on collaborative filtering | |
| CN113821527A (en) | Hash code generation method and device, computer equipment and storage medium | |
| Dai et al. | Study of online learning resource recommendation based on improved BP neural network | |
| CN111125428A (en) | Time-dependent movie recommendation method based on score prediction function fitting structure | |
| Sánchez-Corcuera et al. | Persuasion-based recommender system ensambling matrix factorisation and active learning models | |
| Zhu | Network Course Recommendation System Based on Double‐Layer Attention Mechanism | |
| Tzeng et al. | Massive open online course recommendation system based on a reinforcement learning algorithm | |
| Lin | Learning information recommendation based on text vector model and support vector machine | |
| CN114358807A (en) | User portrayal method and system based on predictable user characteristic attributes | |
| CN113221003A (en) | Mixed filtering recommendation method and system based on dual theory | |
| Salehi | Latent feature based recommender system for learning materials using genetic algorithm | |
| CN111401525A (en) | Adaptive learning system and method based on deep learning | |
| CN116010696A (en) | News recommendation method, system and medium integrating knowledge graph and long-term interest of user | |
| CN112749342A (en) | Personalized recommendation method for network education and teaching resources | |
| Zeng | Application of conditional random field model based on machine learning in online and offline integrated educational resource recommendation | |
| CN112818196B (en) | Data processing method, equipment, electronic device and storage medium based on electronic learning platform | |
| CN113343113A (en) | Cold start entity recommendation method for knowledge distillation based on graph convolution network | |
| Attalariq et al. | Chatbot-Based Book Recommender System Using Singular Value Decomposition |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination |