A personalized homework intelligent recommendation method and device based on student portrait

Abstract

The present invention is applicable to the computer field, and provides a personalized homework intelligent recommendation method and device based on student portraits. In the recommendation method, the Word2Vec model is used to convert the student portrait of each student into a feature word vector, and the similarity between each feature word vector is calculated based on the cosine similarity model; for a given target student, two student portraits similar to the target student are determined based on the calculation result of the similarity, and there is at least one common interest knowledge point in the interest knowledge point set of the two student portraits; according to the common interest knowledge point, one of the knowledge point matching models is used to match multiple target interest knowledge points in each interest knowledge point set, and the common interest knowledge points and the target interest knowledge points are used as a knowledge point set to be recommended; the homework corresponding to each knowledge point in the knowledge point set to be recommended is pushed to the target student as a homework to be recommended. The present invention can provide students with required homework exercises based on student portraits.

Description

Technical Field

[0001] The present invention belongs to the field of computer technology, and in particular provides a personalized homework intelligent recommendation method and device based on student portraits. Background Art

[0002] In the context of the rapid development of Internet education, online exercises have become popular due to their convenience and customizability. Faced with the huge exercise database on the Internet and in the examination system, how to help students discover and fill in the gaps in their knowledge, and how to accurately recommend exercises based on students' needs, has become a focal issue in the field of personalized learning recommendations. In the field of student homework recommendations, related technologies analyze the similarities between students' user attributes (such as student ID, name, gender, department, major, grade, and identity type). Then, based on these similarities, other students who are most similar to the target student are found. Finally, exercises are recommended for the target student based on the exercise choices of these similar students. However, this method does not take into account the differences in the efficiency of each student in learning and mastering knowledge points, and the recommended exercises are single, which cannot ensure whether the exercises recommended to students really meet their needs. Summary of the invention

[0003] The purpose of the embodiments of the present invention is to provide a personalized homework intelligent recommendation method and device based on student portraits, aiming to solve the technical problem that the traditional recommendation method does not take into account the differences in the efficiency of each student in learning and mastering knowledge points, the recommended exercises are single, and it is impossible to ensure whether the exercises recommended to students really meet their needs.

[0004] To achieve the above object, the present invention provides the following technical solutions.

[0005] An embodiment of the present invention provides a personalized homework intelligent recommendation method based on student portraits, and the recommendation method includes the following steps:

[0006] Analyze and process the historical learning data of n students to obtain student portraits of n students, each of which corresponds to a set of interest knowledge points;

[0007] The Word2Vec model is used to convert the student portrait of each student into a feature word vector, and the similarity between each feature word vector is calculated based on the cosine similarity model;

[0008] For a given target student, determining two student portraits similar to the target student based on the calculation result of the similarity, wherein there is at least one common interest knowledge point in the interest knowledge point sets of the two student portraits;

[0009] Determine a selected interest knowledge point selected from the common interest knowledge points; use the knowledge point matching model to match multiple target interest knowledge points related to the selected interest knowledge point in two interest knowledge point sets respectively, and obtain two correlation sets;

[0010] In two relevance sets sorted by relevance, based on preset knowledge point allocation weights, multiple target interest knowledge points that match the allocation weights are selected from the two relevance sets to obtain two target interest knowledge point sets; the target interest knowledge points and common interest knowledge points of the two target interest knowledge point sets are used as knowledge points to be recommended;

[0011] The homework corresponding to the knowledge points to be recommended is pushed to the target students as the homework to be recommended.

[0012] Furthermore, the step of converting each student's student portrait into a feature word vector using the Word2Vec model includes: obtaining data of each student portrait, including the student's personal information, grades, learning behavior habits and interests and hobbies; preprocessing the collected text data, and constructing the preprocessed text data into training corpus; selecting the Skip-Gram model based on the Hierarchical Softmax structure to complete corpus training; and using the gensim library corpus training to obtain the feature word vector of the student user portrait.

[0013] Furthermore, in the step of calculating the similarity between each feature word vector based on the cosine similarity model, the following formula is used to calculate the similarity:

[0014] ;

[0015] In the formula, Indicates the similarity between A and B; A and B represent the feature word vectors of two student portraits respectively; and Respectively represent the coordinate values ​​of the two students in the i-th attribute vector space, , ; I indicates the type of attribute.

[0016] Furthermore, the step of determining two student portraits similar to the target student based on the calculation result of the similarity includes:

[0017] Get all student portrait feature word vectors containing the target student;

[0018] Calculate the similarity between the target student's feature word vector and each student's portrait feature word vector in turn, and the multiple similarities obtained constitute a similarity set ;

[0019] Sort the similarity set from large to small according to the similarity, select the student portrait corresponding to the first similarity based on the sorting result, and retrieve the set of interest knowledge points corresponding to the student portrait ;

[0020] Starting from the second place in the sorting results, select student portraits in turn, and select the first appearance and interest knowledge point set There are student portraits with common knowledge points of interest, and the set of knowledge points of interest corresponding to the student portraits is retrieved .

[0021] Furthermore, the step of using the knowledge point matching model to respectively match multiple target interest knowledge points related to the selected interest knowledge point in the two interest knowledge point sets to obtain two correlation sets includes:

[0022] In the interest knowledge point set In the process, the knowledge point matching model is used to calculate the selected interest knowledge points and the interest knowledge point set respectively. The correlation of other knowledge points of interest in the ;

[0023] In the interest knowledge point set In the process, the knowledge point matching model is used to calculate the selected interest knowledge points and the interest knowledge point set respectively. The correlation of other knowledge points of interest in the .

[0024] Furthermore, in the step of calculating the relevance of the knowledge points of interest using the knowledge point matching model, the following formula is used to calculate the relevance:

[0025] ;

[0026] In the formula, represents the correlation between the interest knowledge point a and the interest knowledge point b; M(a) represents the set of associated knowledge points of knowledge point a; s represents the weight, , R represents a real number set, d represents the dimension; q represents the deviation, ; The vector representation of the relationship between the interest knowledge point a and the interest knowledge point b, ; Represents the representation of the knowledge point of interest b in the vector space; The vector representation of the relationship between the interest knowledge point a and the associated knowledge point z, ; Represents the representation of the associated knowledge point z in the vector space; Represents a vector concatenation operation.

[0027] Furthermore, in the two relevance sets sorted according to the relevance size, based on the preset knowledge point allocation weights, a plurality of target interest knowledge points that match the allocation weights are respectively selected from the two relevance sets to obtain two target interest knowledge point sets, including:

[0028] Based on the preset weights of knowledge points, a set of relevance points is sorted by relevance. Select one or more knowledge points of interest as a first target knowledge point set of interest;

[0029] And the set of correlations sorted by correlation size One or more knowledge points of interest are selected as a second target knowledge point set of interest.

[0030] Furthermore, before the step of pushing the homework corresponding to each knowledge point in the set of knowledge points to be recommended as the homework to be recommended to the target student, the step of verifying the set of knowledge points to be recommended is also included, including:

[0031] Obtain and process the historical learning data of the target students to obtain the knowledge points that the target students have mastered;

[0032] Based on the knowledge points that the target student has mastered, the set of knowledge points to be recommended is updated; wherein, updating the set of knowledge points to be recommended includes: deleting the same knowledge points as the knowledge points that the target student has mastered and supplementing the reserve knowledge points, the reserve knowledge points are selected from the relevance set and / or dependency sets Retrieved from .

[0033] Furthermore, the step of processing the historical learning data of the target student to obtain the knowledge points that the target student has mastered includes: extracting the historical test data of the target student w for the current knowledge point u, and calculating the loss ratio of the target student for the current knowledge point based on the historical test data, expressed as: ; The target students have done v exercises on the current knowledge point, and the occurrence ratio of the current knowledge point is counted, expressed as: ; Calculate the target student's score loss rate for the current knowledge point as: ;The score loss rate is used to represent the target students' mastery of the current knowledge points, and the score loss rate is inversely proportional to the mastery level; where, represents the total number of exercises for knowledge point u; e represents the exercises completed; v represents the number of exercises completed; represents the loss rate of student w who has done the exercises; A vector representation of the target student's mastery of knowledge point u; It indicates the correct rate of the target students in the zth exercise they have done.

[0034] Another embodiment of the present invention provides a personalized homework intelligent recommendation device based on student portraits, and the recommendation device includes the following modules:

[0035] The data analysis module is used to analyze and process the historical learning data of n students to obtain student portraits of n students, each of which corresponds to a set of interest knowledge points;

[0036] The similarity calculation module is used to convert the student portrait of each student into a feature word vector using the Word2Vec model, and calculate the similarity between each feature word vector based on the cosine similarity model;

[0037] A student portrait matching module is used to determine, for a given target student, two student portraits similar to the target student based on the similarity calculation result, wherein there is at least one common interest knowledge point in the interest knowledge point sets of the two student portraits;

[0038] The knowledge point recommendation module is used to determine a selected interest knowledge point selected from the common interest knowledge points; use the knowledge point matching model to match multiple target interest knowledge points related to the selected interest knowledge point in two interest knowledge point sets respectively, to obtain two correlation sets; in the two correlation sets sorted according to the size of the correlation, based on the preset knowledge point allocation weights, respectively select multiple target interest knowledge points that match the allocation weights from the two correlation sets to obtain two target interest knowledge point sets; use the target interest knowledge points and the common interest knowledge points of the two target interest knowledge point sets as the knowledge points to be recommended;

[0039] The exercise recommendation module is used to push the homework corresponding to each knowledge point in the set of knowledge points to be recommended to the target students as the homework to be recommended.

[0040] Compared with the prior art, the method and device for intelligently recommending personalized homework based on student portraits of the present invention have the following beneficial effects:

[0041] First, the present invention uses the Word2Vec model to convert the student portrait of each student into a feature word vector, and calculates the similarity between each feature word vector based on the cosine similarity model; for a given target student, two student portraits similar to the target student are determined based on the calculation result of the similarity. Through the above scheme, the present invention analyzes the historical learning data of each student and constructs a student portrait. The method can capture the set of knowledge points of interest of each student, thereby matching multiple similar student portraits for each student, and providing students with personalized and diversified learning resources and homework recommendations;

[0042] Second, the present invention selects and uses a knowledge point matching model to match multiple target interest knowledge points in each interest knowledge point set based on the common interest knowledge points, and uses the common interest knowledge points and the target interest knowledge points as the knowledge point set to be recommended; the homework corresponding to each knowledge point in the knowledge point set to be recommended is pushed to the target student as the homework to be recommended; therefore, the present invention can match and recommend related homework based on the common interest knowledge points and the target interest knowledge points. This method not only takes into account the common characteristics of students, but also takes into account the individuality of students, making the recommendation more accurate. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings in the following description are only some embodiments of the present invention.

[0044] Figure 1 This is a flowchart of an implementation method of a personalized homework intelligent recommendation method based on student portraits of the present invention;

[0045] Figure 2 A sub-flow chart of a personalized homework intelligent recommendation method based on student portraits of the present invention;

[0046] Figure 3 This is a structural block diagram of a personalized homework intelligent recommendation device based on student portraits of the present invention. DETAILED DESCRIPTION

[0047] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0048] At present, in the related technologies in the field of student homework recommendation, the similarities between the user attributes of students are analyzed. Then, based on these similarities, other students who are most similar to the target student are found. Finally, exercises are recommended for the target student based on the exercise selections of these similar students. However, this method does not take into account the differences in the learning efficiency of each student in terms of knowledge points, and the recommended exercises are single, which cannot ensure whether the exercises recommended to students really meet their needs.

[0049] To solve the above problems, on the one hand, the present invention adopts the Word2Vec model to convert the student portrait of each student into a feature word vector, and calculates the similarity between each feature word vector based on the cosine similarity model; for a given target student, two student portraits similar to the target student are determined based on the calculation result of the similarity. Through the above scheme, the present invention analyzes the historical learning data of each student and constructs a student portrait. The method can capture the set of knowledge points of interest of each student, thereby matching multiple similar student portraits for each student, and providing students with personalized and diversified learning resources and homework recommendations; on the other hand, the present invention selects one of the common interest knowledge points and uses the knowledge point matching model to match multiple target interest knowledge points in each interest knowledge point set according to the common interest knowledge points, and uses the common interest knowledge points and the target interest knowledge points as the knowledge point set to be recommended; the homework corresponding to each knowledge point in the knowledge point set to be recommended is pushed to the target student as the homework to be recommended; therefore, the present invention can match and recommend related homework based on the common interest knowledge points and the target interest knowledge points. This method not only takes into account the commonality of students, but also takes into account the individuality of students, making the recommendation more accurate.

[0050] The specific implementation of the present invention is described in detail below in conjunction with specific embodiments.

[0051] Please refer to Figure 1 In one embodiment of the present invention, a personalized homework intelligent recommendation method based on student portraits is provided, and the recommendation method includes the following steps:

[0052] S101: Analyze and process historical learning data of n students to obtain student portraits of the n students, each student portrait corresponding to a set of interest knowledge points;

[0053] In an embodiment of the present invention, the collected data needs to be cleaned and preprocessed to ensure the quality and availability of the data; a multi-dimensional student portrait is further constructed, which includes the student's mastery of knowledge points, and can also analyze the student's learning habits, learning style and other multi-dimensional data to construct a more comprehensive student portrait.

[0054] S102: The Word2Vec model is used to convert the student portrait of each student into a feature word vector, and the similarity between the feature word vectors is calculated based on the cosine similarity model;

[0055] Among them, see Figure 2 In a preferred embodiment of the present invention, the step of converting the student portrait of each student into a feature word vector using the Word2Vec model includes:

[0056] S201: Obtain data of each student's portrait, including the student's personal information, grades, learning behavior habits, and interests and hobbies;

[0057] S202: preprocessing the collected text data, and constructing the preprocessed text data into training corpus;

[0058] S203: Select the Skip-Gram model based on the Hierarchical Softmax structure to complete the corpus training; use the gensim library corpus training to obtain the feature word vector of the student user portrait.

[0059] Please continue to refer to Figure 1 The personalized homework intelligent recommendation method based on student portraits of the embodiment of the present invention further includes the following steps:

[0060] S103: for a given target student, determining two student portraits similar to the target student based on the similarity calculation result, wherein there is at least one common interest knowledge point in the interest knowledge point sets of the two student portraits;

[0061] Exemplarily, an embodiment of the present invention adopts matching two student portraits similar to the target student. In actual operation, the specific number is not limited. Through the above scheme, the present invention analyzes the historical learning data of each student and constructs a student portrait. The method can capture the set of knowledge points of interest of each student, thereby matching multiple similar student portraits for each student, and providing students with personalized and diversified learning resources and homework recommendations.

[0062] In the step where there is at least one common interest knowledge point in the interest knowledge point sets of the two student portraits, when there is no common interest knowledge point in the interest knowledge point sets of the two student portraits, the student portrait with the largest similarity calculation result (i.e. the most similar) is retained, and the other student portrait is discarded, and then new student portraits are traversed according to the similarity calculation results until there is at least one common interest knowledge point in the interest knowledge point sets of the two student portraits.

[0063] S104: determining a selected interest knowledge point selected from the common interest knowledge points; using the knowledge point matching model to respectively match multiple target interest knowledge points related to the selected interest knowledge point in two interest knowledge point sets, to obtain two correlation sets;

[0064] S105: In the two correlation sets sorted according to the relevance, based on the preset knowledge point allocation weights, multiple target interest knowledge points that match the allocation weights are selected from the two correlation sets to obtain two target interest knowledge point sets; the target interest knowledge points and common interest knowledge points of the two target interest knowledge point sets are used as knowledge points to be recommended;

[0065] According to the common interest knowledge points, the present invention selects one of the knowledge point matching models to match multiple target interest knowledge points in each interest knowledge point set, and uses the common interest knowledge points and the target interest knowledge points as the knowledge point set to be recommended;

[0066] In the embodiment of the present invention, based on the common interest knowledge points and the target interest knowledge points, relevant homework can be matched and recommended, which not only considers the common characteristics of students, but also takes into account the individuality of students, making the recommendation more accurate.

[0067] S106: Pushing the homework corresponding to the knowledge point to be recommended to the target student as the homework to be recommended.

[0068] Furthermore, in the step of calculating the similarity between each feature word vector based on the cosine similarity model, the following formula is used to calculate the similarity:

[0069] ;

[0070] In the formula, Indicates the similarity between A and B; A and B represent the feature word vectors of two student portraits respectively; and Respectively represent the coordinate values ​​of the two students in the i-th attribute vector space, , ; I indicates the type of attribute.

[0071] Furthermore, the step of determining two student portraits similar to the target student based on the calculation result of the similarity includes: obtaining all student portrait feature word vectors containing the target student; calculating the similarity between the feature word vector of the target student and each student portrait feature word vector in turn, and the obtained multiple similarities constitute a similarity set ; Sort the similarity set from large to small according to the similarity, select the student portrait corresponding to the first similarity based on the sorting result, and retrieve the interest knowledge point set corresponding to the student portrait ; Starting from the second place in the sorting results, select student portraits in turn, select the first appearance and interest knowledge point set There are student portraits with common knowledge points of interest, and the set of knowledge points of interest corresponding to the student portraits is retrieved .

[0072] Furthermore, the step of selecting a knowledge point matching model to match a plurality of target interest knowledge points in each interest knowledge point set according to the common interest knowledge points comprises: determining a selected interest knowledge point selected from the common interest knowledge points; In the process, the knowledge point matching model is used to calculate the selected interest knowledge points and the interest knowledge point set respectively. The correlation of other knowledge points of interest in the ; In the set of interest knowledge points In the process, the knowledge point matching model is used to calculate the selected interest knowledge points and the interest knowledge point set respectively. The correlation of other knowledge points of interest in the ;

[0073] Based on the preset weights of knowledge points, a set of relevance points is sorted by relevance. Select one or more knowledge points of interest as the first target knowledge point set of interest, and sort the relevance set according to the relevance size. One or more knowledge points of interest are selected as a second target knowledge point set of interest.

[0074] In the embodiment of the present invention, the weight is used to represent the correlation set. Or a set of dependencies The number of knowledge points selected in the selection, the greater the weight, the more knowledge points selected; as a preference, the relevance set The corresponding first allocation weight and correlation set The sum of the corresponding second allocation weights is 1. When in the correlation set When the number of selected knowledge points is large, correspondingly, select the correlation set The number of knowledge points should be reduced accordingly.

[0075] Furthermore, in the step of calculating the relevance of the knowledge points of interest using the knowledge point matching model, the following formula is used to calculate the relevance:

[0076]

[0077] In the formula, represents the correlation between the interest knowledge point a and the interest knowledge point b; M(a) represents the set of associated knowledge points of knowledge point a; s represents the weight, , R represents a real number set, d represents the dimension; q represents the deviation, ; The vector representation of the relationship between the interest knowledge point a and the interest knowledge point b, ; Represents the representation of the knowledge point of interest b in the vector space; The vector representation of the relationship between the interest knowledge point a and the associated knowledge point z, ; Represents the representation of the associated knowledge point z in the vector space; Represents a vector concatenation operation.

[0078] Furthermore, in the step of taking the common interest knowledge points and the target interest knowledge points as the knowledge point set to be recommended, the target interest knowledge points and the common interest knowledge points of the first target interest knowledge point set and the second target interest knowledge point set are taken as the knowledge point set to be recommended.

[0079] Furthermore, before the step of pushing the homework corresponding to each knowledge point in the set of knowledge points to be recommended as the homework to be recommended to the target student, the step of verifying the set of knowledge points to be recommended is also included, including:

[0080] Obtain and process the historical learning data of the target students to obtain the knowledge points that the target students have mastered;

[0081] Based on the knowledge points that the target students have mastered, updating the set of knowledge points to be recommended;

[0082] In the embodiment of the present invention, updating the set of knowledge points to be recommended includes: deleting knowledge points that are the same as the knowledge points that the target students have mastered and supplementing reserve knowledge points, wherein the reserve knowledge points are selected from the relevance set. and / or dependency sets Retrieved from .

[0083] Furthermore, the step of processing the historical learning data of the target student to obtain the knowledge points that the target student has mastered includes: extracting the historical test data of the target student w for the current knowledge point u, and calculating the loss ratio of the target student for the current knowledge point based on the historical test data, expressed as: ; The target students have done v exercises on the current knowledge point, and the occurrence ratio of the current knowledge point is counted, expressed as: ; Calculate the target student's score loss rate for the current knowledge point as: ;The score loss rate is used to represent the target students' mastery of the current knowledge points, and the score loss rate is inversely proportional to the mastery level; where, represents the total number of exercises for knowledge point u; e represents the exercises completed; v represents the number of exercises completed; represents the loss rate of student w who has done the exercises; A vector representation of the target student's mastery of knowledge point u; Indicates the correct rate of the target student in the zth exercise he has done; in the embodiment of the present invention, the knowledge points recommended to the students are updated accordingly according to the students' mastery of the knowledge points, so that the knowledge points and corresponding exercises recommended to the target students by the recommendation method of the present invention are what the students need.

[0084] Please refer to Figure 3 In another embodiment of the present invention, a personalized homework intelligent recommendation device based on student portraits is provided, and the recommendation device includes the following modules:

[0085] The data analysis module 301 is used to analyze and process the historical learning data of n students to obtain student portraits of the n students, each student portrait corresponding to a set of interest knowledge points;

[0086] A similarity calculation module 302 is used to convert the student portrait of each student into a feature word vector using a Word2Vec model, and calculate the similarity between each feature word vector based on a cosine similarity model;

[0087] The student portrait matching module 303 is used to determine, for a given target student, two student portraits similar to the target student based on the similarity calculation result, wherein there is at least one common interest knowledge point in the interest knowledge point sets of the two student portraits;

[0088] The knowledge point recommendation module 304 is used to determine a selected interest knowledge point selected from the common interest knowledge points; use the knowledge point matching model to match multiple target interest knowledge points related to the selected interest knowledge point in two interest knowledge point sets respectively, to obtain two correlation sets; in the two correlation sets sorted according to the size of the correlation, based on the preset knowledge point allocation weights, respectively select multiple target interest knowledge points that match the allocation weights from the two correlation sets to obtain two target interest knowledge point sets; use the target interest knowledge points and the common interest knowledge points of the two target interest knowledge point sets as the knowledge points to be recommended;

[0089] The exercise recommendation module 305 is used to push the homework corresponding to each knowledge point in the set of knowledge points to be recommended to the target students as the homework to be recommended.

[0090] To sum up, the present invention analyzes each student's historical learning data and constructs a student portrait. This method can capture the set of knowledge points of interest of each student, thereby matching multiple similar student portraits for each student, and providing students with personalized and diversified learning resources and homework recommendations; in addition, the present invention can match and recommend related homework based on common interest knowledge points and target interest knowledge points. This method not only takes into account the common characteristics of students, but also takes into account the individuality of students, making the recommendation more accurate.

[0091] The above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that the technical solutions described in the above embodiments may still be modified, or some or all of the technical features may be replaced by equivalents. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the scope of the technical solutions of the embodiments of the present invention.

[0092] It should be understood that the present disclosure is not limited to the exact structures that have been described above and shown in the drawings, and that various modifications and changes may be made without departing from the scope thereof. The scope of the present disclosure is limited only by the appended claims.

Claims

1. A personalized homework intelligent recommendation method based on student portraits, characterized in that: The recommended approach includes the following steps: Analyze and process the historical learning data of n students to obtain student portraits of n students, each of which corresponds to a set of interest knowledge points; The Word2Vec model is used to convert the student portrait of each student into a feature word vector, and the similarity between each feature word vector is calculated based on the cosine similarity model; For a given target student, determining two student portraits similar to the target student based on the calculation result of the similarity, wherein there is at least one common interest knowledge point in the interest knowledge point sets of the two student portraits; Determine a selected interest knowledge point selected from the common interest knowledge points; use the knowledge point matching model to match multiple target interest knowledge points related to the selected interest knowledge point in two interest knowledge point sets respectively, and obtain two correlation sets; In two relevance sets sorted by relevance, based on preset knowledge point allocation weights, multiple target interest knowledge points that match the allocation weights are selected from the two relevance sets to obtain two target interest knowledge point sets; the target interest knowledge points and common interest knowledge points of the two target interest knowledge point sets are used as knowledge points to be recommended; Pushing the homework corresponding to each knowledge point in the knowledge point set to be recommended to the target students as the homework to be recommended; The step of determining two student portraits similar to the target student based on the similarity calculation result includes: Get all student portrait feature word vectors containing the target student; Calculate the similarity between the target student’s feature word vector and each student’s portrait feature word vector in turn, and the obtained multiple similarities constitute the similarity set S N ={S1,S2,...,S n }; Sort the similarity set from large to small according to the similarity, select the student portrait corresponding to the first similarity based on the sorting result, and retrieve the interest knowledge point set C1 corresponding to the student portrait; Starting from the second position of the sorting result, student portraits are selected in sequence, and the student portrait that first appears with a common interest knowledge point with the knowledge point set C1 is selected, and the interest knowledge point set C2 corresponding to the student portrait is retrieved; The step of using the knowledge point matching model to respectively match multiple target interest knowledge points related to the selected interest knowledge point in two interest knowledge point sets to obtain two correlation sets includes: In the interest knowledge point set C1, the knowledge point matching model is used to calculate the correlation between the selected interest knowledge point and the remaining interest knowledge points in the interest knowledge point set C1 to obtain the correlation set G1; In the interest knowledge point set C2, the correlation between the selected interest knowledge point and the remaining interest knowledge points in the interest knowledge point set C2 is calculated using the knowledge point matching model to obtain a correlation set G2; In the step of calculating the relevance of the knowledge points of interest using the knowledge point matching model, the following formula is used to calculate the relevance: ； Where p(a, b) represents the correlation between interest knowledge point a and interest knowledge point b; M(a) represents the set of associated knowledge points of knowledge point a; s represents the weight, s∈R d , R represents a real number set, d represents the dimension; q represents the deviation, q∈R; r ab The vector representation of the relationship between the interest knowledge point a and the interest knowledge point b, r ab ∈R d ;e b represents the representation of the knowledge point of interest b in the vector space; r az The vector representation of the relationship between the interest knowledge point a and the associated knowledge point z, r az ∈R d ;e z Represents the representation of the associated knowledge point z in the vector space; Represents the concatenation operation of vectors; In two relevance sets sorted according to relevance, based on preset knowledge point allocation weights, a plurality of target interest knowledge points that match the allocation weights are respectively selected from the two relevance sets to obtain two target interest knowledge point sets, including: Based on the preset allocation weights of the knowledge points, one or more knowledge points of interest are selected from the relevance set G1 sorted by relevance as a first target knowledge point set of interest; And one or more knowledge points of interest are selected from the relevance set G2 sorted according to the relevance size as the second target knowledge point set of interest.

2. The personalized homework intelligent recommendation method based on student portrait according to claim 1 is characterized in that: The steps of converting the student portrait of each student into a feature word vector using the Word2Vec model include: obtaining data of each student portrait, including the student's personal information, grades, learning behavior habits and interests; preprocessing the collected text data, and constructing the preprocessed text data into training corpus; selecting a Skip-Gram model based on a Hierarchical Softmax structure to complete corpus training; and using the gensim library corpus training to obtain the feature word vector of the student user portrait.

3. The personalized homework intelligent recommendation method based on student portrait according to claim 2 is characterized in that: In the step of calculating the similarity between each feature word vector based on the cosine similarity model, the similarity is calculated using the following formula: ； In the formula, S （A,B） represents the similarity between A and B; A and B represent the feature word vectors of two student portraits respectively; A i and B i Respectively represent the coordinate values ​​of the two students in the i-th attribute vector space, A i ∈A,B i ∈B; I represents the type of attribute.

4. The personalized homework intelligent recommendation method based on student portrait according to claim 3 is characterized in that: Before the step of pushing the homework corresponding to each knowledge point in the set of knowledge points to be recommended as the homework to be recommended to the target student, the step of verifying the set of knowledge points to be recommended is also included, including: Obtain and process the historical learning data of the target students to obtain the knowledge points that the target students have mastered; Based on the knowledge points that the target students have mastered, the set of knowledge points to be recommended is updated; wherein, updating the set of knowledge points to be recommended includes: deleting the knowledge points that are the same as the knowledge points that the target students have mastered and supplementing the reserve knowledge points, and the reserve knowledge points are retrieved from the relevance set G1 and / or the relevance set G2.

5. The personalized homework intelligent recommendation method based on student portrait according to claim 4 is characterized in that: The step of processing the historical learning data of the target student to obtain the knowledge points that the target student has mastered includes: extracting the historical test data of the target student w for the current knowledge point u, and calculating the loss ratio of the target student for the current knowledge point based on the historical test data, expressed as: ; The target students have done v exercises on the current knowledge point, and the occurrence ratio of the current knowledge point is counted, expressed as: ; Calculate the target student's score loss rate for the current knowledge point as: ;The score loss rate is used to represent the target students’ mastery of the current knowledge points, and the score loss rate is inversely proportional to the mastery level; where V u represents the total number of exercises for knowledge point u; e represents the exercises completed; v represents the number of exercises completed; represents the percentage of students w losing points after doing the exercises; r u The vector representation of the target student’s mastery of knowledge point u; e z It indicates the correct rate of the target students in the zth exercise they have done.

6. A recommendation device for implementing the personalized intelligent recommendation method for homework based on student portraits as claimed in any one of claims 1 to 5, characterized in that: The recommended device includes the following modules: The data analysis module is used to analyze and process the historical learning data of n students to obtain student portraits of n students, each of which corresponds to a set of interest knowledge points; The similarity calculation module is used to convert the student portrait of each student into a feature word vector using the Word2Vec model, and calculate the similarity between each feature word vector based on the cosine similarity model; A student portrait matching module is used to determine, for a given target student, two student portraits similar to the target student based on the similarity calculation result, wherein there is at least one common interest knowledge point in the interest knowledge point sets of the two student portraits; The knowledge point recommendation module is used to determine a selected interest knowledge point selected from the common interest knowledge points; use the knowledge point matching model to match multiple target interest knowledge points related to the selected interest knowledge point in two interest knowledge point sets respectively, to obtain two correlation sets; in the two correlation sets sorted according to the size of the correlation, based on the preset knowledge point allocation weights, respectively select multiple target interest knowledge points that match the allocation weights from the two correlation sets to obtain two target interest knowledge point sets; use the target interest knowledge points and the common interest knowledge points of the two target interest knowledge point sets as the knowledge points to be recommended; The exercise recommendation module is used to push the homework corresponding to each knowledge point in the recommended knowledge point set to the target students as the recommended homework.

Images