Job generation method and system based on multi-source data fusion
Through multi-source data fusion, students' knowledge weaknesses and learning interests are identified, and personalized homework is generated, which solves the problems of low efficiency and lagging feedback in the traditional education model, significantly improving learning effect and education quality.
Patent Information
- Application Number
- CN202411000195.X
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2024-07-24
- Publication Date
- 2024-11-19
- Estimated Expiration
- 2044-07-24
AI Technical Summary
The lack of personalization of traditional education models makes it difficult to target students' learning levels and needs, which are inefficient and lagging in feedback, which affects learning results.
Using a multi-source data fusion method, students' classroom performance, test scores, homework completion status and online learning data are obtained, semantic information is extracted through semantic knowledge base, word vector matrix is generated, knowledge weaknesses and learning interests are identified, and homework guidance is trained. Model, generate personalized assignments.
The generation of personalized homework is achieved, learning effect and efficiency is improved, the homework content is highly correlated with student needs, and the quality of education is improved.
Smart Images

Figure CN118570030B_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of educational technology, and in particular to a homework generation method and system based on multi-source data fusion. Background Art
[0002] With the continuous development of information technology and the increasing abundance of educational resources, the field of education is undergoing a profound change. The traditional education model mainly relies on the experience of teachers and the content of teaching materials. The homework content is usually designed in a unified way, lacking in personalization and pertinence, which to a certain extent restricts the improvement of students' learning effect and teaching quality. In this context, the use of big data, artificial intelligence and data fusion technology to generate personalized homework content has gradually become a research hotspot in the field of educational technology.
[0003] Among them, the traditional way of generating homework has the limitations of lack of personalization, low efficiency and delayed feedback. The reason is that traditional homework is mainly based on textbooks and teachers' experience, and it is difficult to adjust according to each student's learning level and needs, resulting in some students' homework difficulty not being suitable for their learning status, affecting the learning effect. Secondly, teachers need to spend a lot of time and energy designing homework, checking and feedback on homework results, which is particularly difficult in large-class teaching. Moreover, after completing the homework, students need to wait for teachers to correct and feedback, and this process is likely to result in missing the best time for learning and error correction.
[0004] With the development of educational informatization, online learning platforms and learning management systems are also becoming more and more popular. With the emergence of a large number of online learning platforms, these learning platforms can record students' detailed learning behavior data, such as the time they watch videos, the situation of answering questions, the degree of course completion, etc., providing data support for personalized education. At the same time, learning management systems can also conveniently manage course content, student records and learning activity data. These systems provide infrastructure for collecting and analyzing students' learning behaviors.
[0005] However, although the current online learning platforms and learning management systems can record students’ learning status and course management online, the data sources covering students’ learning progress and learning status are wide-ranging, including but not limited to online learning platforms, classroom performance, test scores, teacher evaluations, etc. How to effectively integrate these heterogeneous data is the key to achieving personalized education. In particular, how to provide homework content suitable for each student’s learning path based on their characteristics and needs is the key to effectively improving students’ learning efficiency and effectiveness. Summary of the invention
[0006] In view of this, in order to solve the above problems, the present invention proposes a homework generation method and system based on multi-source data fusion, which aims to accurately identify students' knowledge weaknesses and learning interests by fusing multi-source heterogeneous data, and generate personalized learning homework to improve learning effects.
[0007] The present invention is implemented by the following technical solutions:
[0008] In a first aspect, the present invention provides a method for generating a job based on multi-source data fusion, the method comprising the following steps:
[0009] Obtain multi-source heterogeneous data for data fusion, and after standardizing the multi-source heterogeneous data, obtain multi-source data that fully covers the students' learning situation;
[0010] The obtained multi-source data is converted into text data, and the semantic information in the multi-source data is extracted according to the semantic knowledge base, and projected into the shared semantic space to generate a word vector matrix, and the word vector matrix is decomposed to obtain the semantic representation vector, and the semantic representation vector is fused to obtain the fused multi-source learning data;
[0011] Build student portraits based on the integrated multi-source learning data to identify each student's knowledge weaknesses and learning interests;
[0012] The homework guidance model is trained using semantic representation vectors and student portraits including knowledge weaknesses and learning interests as inputs and personalized homework recommendations as labels;
[0013] Traverse the online selected content uploaded by the student end chapter by chapter, identify the keyword data of the selected content, generate the knowledge point architecture tree of the corresponding data database, and filter out the resource sub-dataset of the knowledge point architecture tree;
[0014] Based on the knowledge weaknesses and learning interests corresponding to the homework guidance model and the student portrait, resource data filtered according to the keyword data is pushed from the resource sub-dataset to generate personalized homework that suits each student.
[0015] Through the above steps, the present invention can effectively utilize multi-source data fusion to generate homework that meets the personalized needs of students, helping students to better master knowledge points and improve their learning interest and effect.
[0016] As a further solution of the present invention, the acquired multi-source heterogeneous data includes but is not limited to:
[0017] Students’ classroom performance data (such as attendance, class participation, etc.).
[0018] Students' test score data (such as midterm exams, final exams, etc.).
[0019] Students' homework completion data (such as homework completion time, accuracy rate, etc.).
[0020] Students’ online learning data (such as online course viewing records, online test results, etc.).
[0021] Students’ learning behavior data (such as learning time distribution, learning path, etc.).
[0022] As a further solution of the present invention, when standardizing multi-source heterogeneous data, it also includes interpolation and deviation correction processing of the multi-source heterogeneous data, wherein the multi-source heterogeneous data is interpolated to a grid of a specified resolution using inverse distance weighted interpolation, and the deviation of the current data is corrected using historical data. During interpolation, the estimated value calculation formula of the interpolation point is: In the formula, Interpolation point The estimated value at is the total number of known data points involved in the interpolation calculation, From 1 to The index is used to mark each known data point. For the The value of the known point, is the interpolation point and The distance between known points, is the weight index.
[0023] Among them, the interpolation point Estimated value of is a known point value The weighted average of the known points and the interpolation points is the inverse of the distance between them. Power, where In the calculation formula, the numerator For all known point values According to its weight Perform weighted summation, the denominator To calculate the sum of the weights of all known points, the estimated value Divide the numerator by the denominator to get the interpolation point The estimated value at .
[0024] As a further solution of the present invention, when extracting semantic information from multi-source data based on a semantic knowledge base, a semantic knowledge base covering the field of multi-source heterogeneous data is selected and constructed based on the existing WordNet, DBpedia and ConceptNet knowledge bases; Stanford NER is used to identify named entities on the standardized multi-source data, and the identified entities are matched with entries in the semantic knowledge base, word sense disambiguation is performed on the identified entities, and the semantic relationships between entities and concepts are extracted from the semantic knowledge base, and a semantic graph containing entities and relationships is constructed and generated. Through the above detailed steps, semantic information can be systematically extracted from multi-source data to generate an accurate and clearly structured semantic graph. Each step has a clear execution method, and the innovativeness and legality of the method are guaranteed by the claims.
[0025] As a further solution of the present invention, projecting to a shared semantic space to generate a word vector matrix, and decomposing the word vector matrix to obtain a semantic representation vector, including:
[0026] Input the text data after extracting semantic information into the pre-trained word vector model to generate the corresponding word vector;
[0027] Map word vectors to a shared semantic space and use adversarial training to align word vectors from different sources;
[0028] The aligned word vectors are combined to form a complete word vector matrix;
[0029] The word vector matrix is centered, the mean of each column is subtracted, and the singular value decomposition is applied to decompose the word vector matrix into three matrices U, Σ, ,in: In the formula, is an m×n word vector matrix; U is an m×n orthogonal matrix, and the column vector of U is The left singular vector of ; Σ is an m×n diagonal matrix, and the elements on the diagonal of Σ are The singular values of is an m×n orthogonal matrix, The row vector is The transpose of the right singular vectors of ;
[0030] Select the first k singular values and their corresponding singular vectors to obtain a low-dimensional semantic representation vector; after selecting the first k singular values, the formula is expressed as: In the formula, is the first k columns of U, of size m×k; are the first k singular values of Σ, corresponding to a k×k diagonal matrix; yes The first k rows of U are of size k×n; the first k columns of the decomposed U matrix are used as the low-dimensional semantic representation vector of the word.
[0031] Among them, the word vector model is a pre-trained Word2Vec model, which is obtained by training on a corpus using Word2Vec. During adversarial training, it also includes a designed adversarial training framework, which includes a generator and a discriminator. The generator is responsible for converting the word vector to a shared semantic space, and the discriminator evaluates whether the word vector comes from the same data source. By repeatedly training the generator and the discriminator, the word vectors from different sources are consistent in the shared semantic space.
[0032] Through the above steps, the word vectors in multi-source data can be aligned to a shared semantic space, and a low-dimensional semantic representation vector can be generated through the singular value decomposition method. This process ensures that data from different sources can be compared and analyzed in a unified semantic space, providing efficient and accurate semantic representation.
[0033] As a further solution of the present invention, when constructing a student portrait, a student portrait containing multi-dimensional features of knowledge level, learning habits, and hobbies is constructed, and the fused multi-source learning data is Decompose into multi-dimensional feature subsets and establish a knowledge point mapping matrix: In the formula, Indicates The learning activities involve The degree of mastery of each knowledge point is calculated; the degree of mastery of each knowledge point is calculated: in, is the matrix of students’ mastery of each knowledge point, where is the feature subset decomposed into;
[0034] Identify weaknesses and set a mastery threshold , identify the knowledge points whose mastery level is below the threshold: In the formula, W is the set of students' weak points in knowledge; K-means is used to cluster the learning interest feature subsets, identify different points of interest, annotate the clustering results, and identify the meaning of each point of interest;
[0035] Integrate the above calculation results to generate a detailed portrait report for each student.
[0036] As a further solution of the present invention, when training the homework guidance model, the generated semantic representation vector is paired with the corresponding semantic label to form a training data set model for training; wherein the homework semantic representation vector is used as input, and the semantic information corresponding to the knowledge weaknesses and learning interest points is used as the label; the Transformer model is selected and the generated training data set is used for model training to obtain the homework guidance model, and the trained model is used to generate personalized homework guidance.
[0037] As a further solution of the present invention, when generating a knowledge point architecture tree, a hierarchical structure of the knowledge point architecture tree is initially established based on the chapter content and extracted keywords, and the relationship between each knowledge point is determined through semantic analysis and association rule mining, redundant nodes are removed, and the architecture tree structure is optimized. The knowledge points are matched with the resources in the data database, and the corresponding resource sub-datasets are filtered out based on the obtained knowledge point architecture tree.
[0038] In a second aspect, the present invention also includes a job generation system based on multi-source data fusion, the system comprising:
[0039] Multi-source data acquisition module: used to acquire multi-source heterogeneous data and perform standardization on the acquired multi-source heterogeneous data to generate multi-source data that fully covers students' learning situations;
[0040] Semantic information extraction module: used to convert the standardized multi-source data into text data, extract the semantic information in the multi-source data according to the semantic knowledge base, and project it into the shared semantic space to generate a word vector matrix;
[0041] Word vector decomposition and fusion module: used to decompose the word vector matrix to obtain semantic representation vectors, and fuse these vectors to generate fused multi-source learning data;
[0042] Student portrait construction module: Based on the integrated multi-source learning data, student portraits are constructed to identify each student's knowledge weaknesses and learning interests;
[0043] Homework guidance model training module: used to train the homework guidance model by taking the semantic representation vector as input and the corresponding semantic information as label, combined with the student profile and their knowledge weaknesses and learning interests;
[0044] Online content recognition module: used to traverse the online selected content uploaded by the student end chapter by chapter and identify the keyword data of the selected content;
[0045] Architecture tree generation module: used to generate a knowledge point architecture tree of the corresponding data database according to the identified keyword data, and filter out resource sub-datasets of the knowledge point architecture tree;
[0046] Homework generation module: used to filter resource data from the resource sub-dataset based on the homework guidance model and the knowledge weaknesses and learning interests corresponding to the student portrait, and generate personalized homework for each student.
[0047] As a further solution of the present invention, the homework generation system also includes a data push module for pushing the generated personalized homework to the student end, ensuring that each student can receive homework that meets his or her learning needs and interests.
[0048] Through the collaborative work of the above modules, the homework generation system of the present invention can efficiently and accurately generate personalized learning homework, helping students to better master knowledge and improve learning effects.
[0049] The present invention also includes a computer device, comprising: at least one processor, and a memory communicatively connected to the at least one processor, wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor executes the job generation method based on multi-source data fusion.
[0050] The present invention also includes a computer-readable storage medium, wherein the computer-readable storage medium stores computer instructions, and the computer instructions are used to enable the computer to execute the job generation method based on multi-source data fusion.
[0051] Compared with the prior art, the job generation method and system based on multi-source data fusion provided by the present invention has the following beneficial effects:
[0052] 1. By acquiring and integrating multi-source heterogeneous data, such as academic performance, classroom performance, homework grades, online learning records, etc., it comprehensively covers the students' learning situation, helps to accurately evaluate the students' learning status, and after eliminating the differences between different data sources, obtains multi-source data that comprehensively covers the students' learning situation.
[0053] 2. Semantic information extraction and multi-source data fusion are realized. Semantic information in multi-source data is extracted using the semantic knowledge base and projected into a shared semantic space to generate a word vector matrix. Semantic representation vectors are generated through decomposition and fusion to ensure that important information in the data can be fully utilized.
[0054] 3. Ability to build personalized student portraits and homework guidance models. Based on the integrated multi-source learning data, a detailed student portrait is built to identify each student's knowledge weaknesses and learning interests, providing a basis for personalized learning; using semantic representation vectors as input, combined with student portraits and their knowledge weaknesses and learning interests, an accurate homework guidance model is trained to ensure that the generated homework can target students' specific needs and interests.
[0055] 4. It realizes intelligent identification of online content and resource data screening, and generates homework in a targeted manner. By traversing the online selected content uploaded by the student end chapter by chapter, identifying the keyword data of the selected content, and generating the corresponding knowledge point architecture tree, it ensures that the homework content is highly relevant to the students' actual learning content. Based on the generated knowledge point architecture tree and student portrait, the most suitable resource data is screened from the resource sub-dataset to ensure that the homework content obtained by the students is the most suitable for their learning needs. Through the homework guidance model and student portrait, personalized homework is generated for each student, helping students to study more targetedly and improve learning effects.
[0056] In summary, the homework generation method and system based on multi-source data fusion of the present invention significantly improves the accuracy and personalization of homework generation through comprehensive data acquisition, intelligent semantic analysis, precise homework guidance and efficient personalized homework generation, and promotes the improvement of students' learning effects and education quality.
[0057] These and other aspects of the present invention will be more clearly understood in the following description of the embodiments. It should be understood that the above general description and the following detailed description are only exemplary and explanatory and cannot limit the present invention. BRIEF DESCRIPTION OF THE DRAWINGS
[0058] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or related technologies, the following briefly introduces the drawings required for use in the exemplary embodiments or related technical descriptions. The drawings are used to provide a further understanding of the present invention and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention and do not constitute a limitation to the present invention. In the drawings:
[0059] Figure 1 The present invention is a flowchart of a method for generating a job based on multi-source data fusion according to an embodiment of the present invention.
[0060] Figure 2 The present invention is a flowchart of decomposing and obtaining a semantic representation vector in a job generation method based on multi-source data fusion according to an embodiment of the present invention.
[0061] Figure 3 The present invention provides a flowchart for constructing a student portrait in a homework generation method based on multi-source data fusion according to an embodiment of the present invention. DETAILED DESCRIPTION
[0062] 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 used to limit the present invention.
[0063] In some of the processes described in the specification and claims of the present invention and the above-mentioned figures, multiple operations that appear in a specific order are included, but it should be clearly understood that these operations may not be executed in the order in which they appear in this article or executed in parallel. The serial numbers of the operations, such as 101, 102, etc., are only used to distinguish different operations, and the serial numbers themselves do not represent any execution order. In addition, these processes may include more or fewer operations, and these operations may be executed in sequence or in parallel. It should be noted that the descriptions of "first", "second", etc. in this article are used to distinguish different messages, devices, modules, etc., do not represent the order of precedence, and do not limit the "first" and "second" to be different types.
[0064] The following will be combined with the drawings in the exemplary embodiments of the present invention to clearly and completely describe the technical solutions in the exemplary embodiments of the present invention. Obviously, the exemplary embodiments described are only part of the embodiments of the present invention, not all of them. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative work are within the scope of protection of the present invention.
[0065] In order to provide homework content suitable for the learning path of each student according to his / her characteristics and needs, the present invention provides a homework generation method and system based on multi-source data fusion, which accurately identifies students' knowledge weaknesses and learning interests by fusing multi-source heterogeneous data, and generates personalized learning homework to improve learning effects.
[0066] The technical solution of the present invention is further described below in conjunction with specific embodiments:
[0067] See also Figure 1 As shown, Figure 1 A flowchart of a method for generating a job based on multi-source data fusion provided by the present invention. A method for generating a job based on multi-source data fusion provided in one embodiment of the present invention comprises the following steps:
[0068] Step S10: Acquire multi-source heterogeneous data for data fusion, and standardize the multi-source heterogeneous data to obtain multi-source data that fully covers the student's learning situation.
[0069] In this step, the multi-source heterogeneous data obtained include but are not limited to:
[0070] Students’ classroom performance data (such as attendance, class participation, etc.).
[0071] Students' test score data (such as midterm exams, final exams, etc.).
[0072] Students' homework completion data (such as homework completion time, accuracy rate, etc.).
[0073] Students’ online learning data (such as online course viewing records, online test results, etc.).
[0074] Students’ learning behavior data (such as learning time distribution, learning path, etc.).
[0075] For example, when generating personalized homework through the method based on multi-source data fusion in this embodiment, the following multi-source heterogeneous data of a student (Xiao Ming) is obtained in a selected semester:
[0076] 1. Classroom performance data:
[0077] Attendance rate: 95%;
[0078] Class participation: High;
[0079] 2. Test score data:
[0080] Midterm exam score: 85 points;
[0081] Final exam score: 88 points;
[0082] 3. Job completion data:
[0083] Average homework completion time: 40 minutes;
[0084] Average accuracy: 90%;
[0085] 4. Online learning data:
[0086] Online course viewing record: completion rate 80%;
[0087] Online test results: 85%;
[0088] 5. Learning behavior data:
[0089] Study time distribution: 7 to 9 pm every day;
[0090] Learning path: Review class notes first, then complete assignments, and finally take an online test.
[0091] In this embodiment, when standardizing multi-source heterogeneous data, it also includes interpolation and deviation correction processing of the multi-source heterogeneous data, wherein the multi-source heterogeneous data is interpolated to a grid of a specified resolution using inverse distance weighted interpolation, and the deviation of the current data is corrected using historical data. During interpolation, the estimated value calculation formula of the interpolation point is: In the formula, Interpolation point The estimated value at is the total number of known data points involved in the interpolation calculation, From 1 to The index is used to mark each known data point. For the The value of the known point, is the interpolation point and The distance between known points, is the weight index. Among them, the interpolation point Estimated value of is a known point value The weighted average of the known points and the interpolation points is the inverse of the distance between them. Power, where In the calculation formula, the numerator For all known point values According to its weight Perform weighted summation, the denominator To calculate the sum of the weights of all known points, the estimated value Divide the numerator by the denominator to get the interpolation point The estimated value at .
[0092] Step S20, convert the obtained multi-source data into text data, extract semantic information from the multi-source data according to the semantic knowledge base, project it into a shared semantic space to generate a word vector matrix, decompose the word vector matrix to obtain a semantic representation vector, and fuse the semantic representation vectors to obtain fused multi-source learning data.
[0093] In this step, when extracting semantic information from multi-source data based on the semantic knowledge base, a semantic knowledge base covering the field of multi-source heterogeneous data is selected and constructed based on the existing WordNet, DBpedia and ConceptNet knowledge bases; Stanford NER is used to identify named entities on the standardized multi-source data, and the identified entities are matched with the entries in the semantic knowledge base, word sense disambiguation is performed on the identified entities, the semantic relations between entities and concepts are extracted from the semantic knowledge base, and a semantic graph containing entities and relations is constructed and generated.
[0094] Through the above detailed steps, semantic information can be systematically extracted from multi-source data to generate accurate and clearly structured semantic graphs. Each step has a clear execution method, and the innovation and legality of the method are guaranteed by the claims.
[0095] Exemplarily, when converting multi-source data into text data, the multi-source data collected above can be converted into a structured text representation: Xiao Ming's attendance rate is 95%. His class participation is very high. The midterm exam score is 85 points, and the final exam score is 88 points. The average homework completion time is 40 minutes, and the accuracy rate is 90%. He completed 80% of the online courses, and the online test result is 85%. Xiao Ming studies from 7 to 9 pm every day, and the learning path is to review the class notes first, then complete the homework, and finally take the online test.
[0096] In this embodiment, see Figure 2 As shown in the figure, the word vector matrix is generated by projecting it into the shared semantic space. After decomposing the word vector matrix, the semantic representation vector is obtained, including:
[0097] Step S201: input the text data after semantic information is extracted into a pre-trained word vector model to generate a corresponding word vector;
[0098] Step S202: Map the word vectors to a shared semantic space and use adversarial training to align word vectors from different sources;
[0099] Step S203: Combining the aligned word vectors to form a complete word vector matrix;
[0100] Step S204: Center the word vector matrix, subtract the mean of each column, and apply singular value decomposition to decompose the word vector matrix into three matrices U, Σ, ,in: In the formula, is an m×n word vector matrix; U is an m×n orthogonal matrix, and the column vector of U is The left singular vector of ; Σ is an m×n diagonal matrix, and the elements on the diagonal of Σ are The singular values of is an m×n orthogonal matrix, The row vector is The transpose of the right singular vectors of ;
[0101] Step S205: Select the first k singular values and their corresponding singular vectors to obtain a low-dimensional semantic representation vector; after selecting the first k singular values, the formula is expressed as: In the formula, is the first k columns of U, of size m×k; are the first k singular values of Σ, corresponding to a k×k diagonal matrix; yes The first k rows of U are of size k×n; Step S206, using the first k columns in the decomposed U matrix as the low-dimensional semantic representation vector of the word.
[0102] Among them, the word vector model is a pre-trained Word2Vec model, which is obtained by training on a corpus using Word2Vec. During adversarial training, it also includes a designed adversarial training framework, which includes a generator and a discriminator. The generator is responsible for converting the word vector to a shared semantic space, and the discriminator evaluates whether the word vector comes from the same data source. By repeatedly training the generator and the discriminator, the word vectors from different sources are consistent in the shared semantic space.
[0103] Through the above steps, Xiao Ming's multi-source data is successfully projected into a shared semantic space, and a low-dimensional semantic representation vector is generated. These representation vectors can be further used to generate personalized homework to ensure that the homework content can accurately reflect Xiao Ming's learning situation and needs. The word vectors in the multi-source data can be aligned to the shared semantic space, and the low-dimensional semantic representation vector is generated through the singular value decomposition method. This process ensures that data from different sources can be compared and analyzed in a unified semantic space, providing efficient and accurate semantic representation.
[0104] Step S30: construct a student portrait based on the integrated multi-source learning data to identify each student's knowledge weaknesses and learning interests.
[0105] In this step, see Figure 3 As shown in the figure, when building a student portrait, the following steps are included:
[0106] Step S301: Construct a student portrait containing multi-dimensional features including knowledge level, learning habits, and hobbies, and combine the fused multi-source learning data Decompose into multi-dimensional feature subsets and establish a knowledge point mapping matrix: In the formula, Indicates The learning activities involve The degree of knowledge points;
[0107] Step S302: Calculate the mastery of knowledge points, and calculate the mastery of each knowledge point by each student:
[0108]
[0109] in, is the matrix of students’ mastery of each knowledge point, where Step S303: weak point identification, setting a mastery threshold , identify the knowledge points whose mastery level is below the threshold: In the formula, W is the set of students’ knowledge weaknesses;
[0110] Step S304: clustering the learning interest feature subset using K-means to identify different points of interest, annotating the clustering results, and identifying the meaning of each point of interest;
[0111] Step S305: Integrate the above calculation results to generate a detailed portrait report for each student.
[0112] Exemplarily, when establishing a knowledge point mapping matrix to represent the degree of involvement of different learning activities with different knowledge points, it is assumed that the knowledge points are as follows: Knowledge point A (algebra);
[0113] Knowledge point B (Geometry);
[0114] Knowledge point C (Calculus).
[0115] The knowledge point mapping matrix is expressed as follows:
[0116] Then, when calculating the mastery of each knowledge point by student Xiao Ming, the mastery is:
[0117] Mastery = classroom performance × 0.8 + test score × 0.9 + homework completion × 0.85 + online learning × 0.8.
[0118] If Xiao Ming's calculation results are: knowledge point A 0.85; knowledge point B 0.75; knowledge point C 0.65, set the mastery threshold to 0.70 and identify the knowledge points whose mastery is below the threshold. Perform K-means clustering on the learning interest features (class participation and online course viewing records), and the clustering results are as follows:
[0119] Cluster 1: high engagement and high completion rate;
[0120] Cluster 2: Moderate engagement and moderate completion rate;
[0121] Assume that Xiaoming is clustered into cluster 1, which means that Xiaoming is interested in content with high engagement and high completion rate. Then, by integrating the results, the detailed portrait report of Xiaoming is generated as follows:
[0122] Knowledge level: Good (but weak in knowledge point C);
[0123] Study habits: high attendance, study time concentrated in the evening, and organized study path;
[0124] Interests: Interested in content with high engagement and high completion rates.
[0125] Through the above process, the present invention can generate a personalized homework for Xiao Ming, which can not only consolidate his weak knowledge points, but also stimulate his interest in learning, which can not only improve the student's learning effect, but also make the learning process more interesting and efficient.
[0126] Step S40: Using the semantic representation vector and the student portrait including knowledge weaknesses and learning interests as input, and using personalized homework recommendations as labels, a homework guidance model is trained.
[0127] In this step, when training the homework instruction model, the generated semantic representation vector is paired with the corresponding semantic label to form a training data set model for training; wherein the homework semantic representation vector is used as input, and the semantic information corresponding to the knowledge weaknesses and learning interest points is used as the label; the Transformer model is selected and the generated training data set is used for model training to obtain the homework instruction model, and the trained model is used to generate personalized homework instructions.
[0128] Step S50, traverse the online selected content uploaded by the student end chapter by chapter, identify the keyword data of the selected content, generate a knowledge point architecture tree of the corresponding data database, and filter out the resource sub-dataset of the knowledge point architecture tree.
[0129] In this step, when generating the knowledge point architecture tree, the hierarchical structure of the knowledge point architecture tree is initially established based on the chapter content and extracted keywords. Through semantic analysis and association rule mining, the relationship between each knowledge point is determined, redundant nodes are removed, the architecture tree structure is optimized, and the knowledge points are matched with the resources in the data database. According to the obtained knowledge point architecture tree, the corresponding resource sub-datasets are filtered out.
[0130] Step S60: Based on the homework guidance model and the knowledge weaknesses and learning interests corresponding to the student portrait, resource data filtered according to the keyword data is pushed from the resource sub-data set to generate personalized homework that suits each student.
[0131] For example, the online selected content uploaded by the student end is traversed chapter by chapter, a knowledge point architecture tree is generated and the resource sub-data set is filtered. When identifying the keyword data of the selected content, suppose that in a certain chapter, Xiao Ming uploaded online selected content about "calculus", which includes: defining integrals, differentials, and Newton-Leibniz formula. The following keywords are extracted through natural language processing:
[0132] (1) Define the integral;
[0133] (2) Differentiation;
[0134] (3) Newton-Leibniz formula.
[0135] When generating the knowledge point architecture tree, combine the chapter content and the extracted keywords to initially establish the hierarchical structure of the knowledge point architecture tree:
[0136] Calculus
[0137] ├── Define integral
[0138] ├── Differentiation
[0139] └── Newton-Leibniz formula
[0140] Then, semantic analysis and association rule mining are carried out to determine the relationship between each knowledge point and obtain:
[0141] Defining integrals is closely related to differentials and can be linked together via the Newton-Leibniz formula.
[0142] Then, the optimized knowledge point architecture tree is as follows:
[0143] Calculus
[0144] ├── Define integral
[0145] │ └── Newton-Leibniz formula
[0146] └── Differentiation
[0147] └── Newton-Leibniz formula
[0148] Then the knowledge points are matched with the resources in the data database, and the matching results are:
[0149] (1) Define points: match them to some exercise books, explanation videos and other resources.
[0150] (2) Differentiation: Match to relevant online courses, exercises and other resources.
[0151] (3) Newton-Leibniz formula: Matched with detailed theoretical explanations, application examples and other resources.
[0152] Finally, the corresponding resource sub-datasets are selected based on the obtained knowledge point architecture tree. The resource sub-datasets are:
[0153] ├── Define integral
[0154] │ ├── Exercise Book
[0155] │ └── Explanation video
[0156] ├── Differentiation
[0157] │ ├── Online Courses
[0158] │ └── Exercises
[0159] └── Newton-Leibniz formula
[0160] ├── Theoretical explanation
[0161] └── Application Examples
[0162] Finally, according to Xiao Ming’s student portrait:
[0163] Knowledge weaknesses: Knowledge point C (Knowledge point C includes defining integrals and differentials);
[0164] Learning interest points: High engagement content.
[0165] Based on the homework guidance model and student portrait push resources, generate personalized homework, select resources suitable for Xiao Ming from the resource sub-dataset, and push the following to Xiao Ming's weak points in knowledge (defined integrals and differentials): defined integrals exercise books and explanation videos, as well as differentials online courses and exercises. At the same time, considering Xiao Ming's learning interests, more interactive resources can be selected, such as interesting explanation videos and interactive online courses. The personalized homework generated in the end is as follows:
[0166] 1. Define the integral:
[0167] ▪Exercise Book: Complete Questions 1-10 in Chapter 3.
[0168] ▪Explanatory video: Watch the video “Application of Defined Integrals” and answer the related questions.
[0169] 2. Differentiation:
[0170] ▪Online Course: Complete Section 2 “Basic Concepts of Differentiation” and take the online quiz.
[0171] ▪Exercises: Complete Exercises 1-15 in Chapter 4.
[0172] 3. Newton-Leibniz formula:
[0173] ▪Theoretical explanation: Read the lecture notes on "Newton-Leibniz formula and its applications".
[0174] ▪Application Examples: Complete relevant application example questions and submit a report.
[0175] Through the above steps, the present invention generates a detailed and personalized homework guide for Xiao Ming, ensuring that Xiao Ming can effectively study for his weaknesses while maintaining a high interest in learning. This method not only improves the pertinence and effectiveness of learning, but also stimulates students' learning motivation.
[0176] Through the above steps, the present invention can effectively utilize multi-source data fusion to generate homework that meets the personalized needs of students, helping students to better master knowledge points and improve their learning interest and effect.
[0177] Compared with the currently proposed online learning platform and learning management system, the present invention obtains and fuses multi-source heterogeneous data, such as academic performance, classroom performance, homework performance, online learning records, etc., to comprehensively cover the learning situation of students, which is helpful to accurately evaluate the learning status of students. After eliminating the differences between different data sources, multi-source data that comprehensively covers the learning situation of students is obtained. The semantic information in the multi-source data is extracted using a semantic knowledge base, and projected into a shared semantic space to generate a word vector matrix. The semantic representation vector is generated by decomposition and fusion to ensure that the important information in the data can be fully utilized. Based on the fused multi-source learning data, a detailed student portrait is constructed to identify each student's knowledge weaknesses and learning interests, providing a basis for personalized learning; with the semantic representation vector as input, combined with the student portrait and its knowledge weaknesses and learning interests, an accurate homework guidance model is trained to ensure that the generated homework can target the specific needs and interests of the students.
[0178] The homework generation method based on multi-source data fusion of the present invention can also traverse the online clicked content uploaded by the student end chapter by chapter, identify the keyword data of the clicked content, generate the corresponding knowledge point architecture tree, and ensure the high correlation between the homework content and the actual learning content of the students. Based on the generated knowledge point architecture tree and student portrait, the most suitable resource data is screened out from the resource sub-dataset to ensure that the homework content obtained by the students is the most suitable for their learning needs. Through the homework guidance model and student portrait, personalized homework that meets each student is generated to help students learn more targetedly and improve learning effects.
[0179] It should be understood that, although described in a certain order, these steps are not necessarily performed in sequence in the above order. Unless there is clear explanation in this article, the execution of these steps does not have strict order restriction, and these steps can be performed in other orders. Moreover, a part of the steps of the present embodiment may include a plurality of steps or a plurality of stages, and these steps or stages are not necessarily performed at the same time, but can be performed at different times, and the execution order of these steps or stages is not necessarily performed in sequence, but can be performed in turn or alternately with at least a part of the steps or stages in other steps or other steps.
[0180] In one embodiment, the present invention provides a job generation system based on multi-source data fusion, which is used to execute the above-mentioned job generation method based on multi-source data fusion, and the system includes:
[0181] Multi-source data acquisition module: used to acquire multi-source heterogeneous data, and perform standardization on the acquired multi-source heterogeneous data to generate multi-source data that fully covers the students' learning situation;
[0182] Semantic information extraction module: used to convert the standardized multi-source data into text data, extract the semantic information in the multi-source data according to the semantic knowledge base, and project it into the shared semantic space to generate a word vector matrix;
[0183] Word vector decomposition and fusion module: used to decompose the word vector matrix to obtain semantic representation vectors, and fuse these vectors to generate fused multi-source learning data;
[0184] Student portrait construction module: Based on the integrated multi-source learning data, student portraits are constructed to identify each student's knowledge weaknesses and learning interests;
[0185] Homework guidance model training module: used to train the homework guidance model by taking the semantic representation vector as input and the corresponding semantic information as label, combined with the student profile and their knowledge weaknesses and learning interests;
[0186] Online content recognition module: used to traverse the online selected content uploaded by the student end chapter by chapter and identify the keyword data of the selected content;
[0187] Architecture tree generation module: used to generate a knowledge point architecture tree of the corresponding data database according to the identified keyword data, and filter out resource sub-datasets of the knowledge point architecture tree;
[0188] Homework generation module: used to filter resource data from the resource sub-dataset based on the homework guidance model and the knowledge weaknesses and learning interests corresponding to the student profile, and generate personalized homework for each student;
[0189] The data push module is used to push the generated personalized homework to the student end, ensuring that each student can receive homework that meets their learning needs and interests.
[0190] The homework generation system based on multi-source data fusion of the present invention can efficiently and accurately generate personalized learning homework through the collaborative work of the above modules, helping students to better master knowledge and improve learning effects.
[0191] In this embodiment, the job generation system based on multi-source data fusion adopts the steps of a job generation method based on multi-source data fusion as mentioned above during execution. Therefore, the operation process of the job generation system based on multi-source data fusion is no longer introduced in detail in this embodiment.
[0192] In summary, the homework generation method and system based on multi-source data fusion of the present invention significantly improves the accuracy and personalization of homework generation through comprehensive data acquisition, intelligent semantic analysis, precise homework guidance and efficient personalized homework generation, and promotes the improvement of students' learning effects and education quality.
[0193] In one embodiment, a computer device is also provided in an embodiment of the present invention, comprising at least one processor and a memory communicatively connected to the at least one processor, wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor performs the steps of the job generation method based on multi-source data fusion.
[0194] In one embodiment, the present invention further provides a computer-readable storage medium storing computer instructions, wherein the computer instructions are used to enable the computer to execute the steps of the job generation method based on multi-source data fusion.
[0195] A person skilled in the art can understand that all or part of the processes in the above-mentioned embodiments can be realized by instructing the relevant hardware through a computer program represented by computer instructions, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, storage, database or other media used in the embodiments provided in this application can include at least one of non-volatile and volatile memory.
[0196] Non-volatile memory may include read-only memory, magnetic tape, floppy disk, flash memory or optical storage, etc. Volatile memory may include random access memory or external cache memory. As an illustration and not limitation, RAM may be in various forms, such as static random access memory or dynamic random access memory, etc.
[0197] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present invention should be included in the protection scope of the present invention.
Claims
1. A method for generating jobs based on multi-source data fusion, characterized in that: The method comprises the following steps: Obtain multi-source heterogeneous data for data fusion, and after standardizing the multi-source heterogeneous data, obtain multi-source data that fully covers the students' learning situation; The obtained multi-source data is converted into text data, and the semantic information in the multi-source data is extracted according to the semantic knowledge base, and projected into the shared semantic space to generate a word vector matrix, and the word vector matrix is decomposed to obtain the semantic representation vector, and the semantic representation vector is fused to obtain the fused multi-source learning data; Build student portraits based on the integrated multi-source learning data to identify each student's knowledge weaknesses and learning interests; The homework guidance model is trained using semantic representation vectors and student portraits including knowledge weaknesses and learning interests as inputs and personalized homework recommendations as labels; Traverse the online selected content uploaded by the student end chapter by chapter, identify the keyword data of the selected content, generate the knowledge point architecture tree of the corresponding data database, and filter out the resource sub-dataset of the knowledge point architecture tree; Based on the knowledge weaknesses and learning interests corresponding to the homework guidance model and the student portrait, resource data filtered according to the keyword data is pushed from the resource sub-dataset to generate personalized homework that suits each student; Among them, the word vector matrix is generated by projecting into the shared semantic space, and the semantic representation vector is obtained after decomposing the word vector matrix, including: Input the text data after extracting semantic information into the pre-trained word vector model to generate the corresponding word vector; Map word vectors to a shared semantic space and use adversarial training to align word vectors from different sources; The aligned word vectors are combined to form a complete word vector matrix; Center the word vector matrix, subtract the mean of each column, and apply singular value decomposition to decompose the word vector matrix into three matrices , , ; Select the first k singular values and their corresponding singular vectors to obtain a low-dimensional semantic representation vector; Use the first k columns in the decomposed U matrix as the low-dimensional semantic representation vector of the word.
2. The method for generating a job based on multi-source data fusion according to claim 1, characterized in that: The multi-source heterogeneous data obtained include students' classroom performance data, test score data, homework completion data, online learning data, and learning behavior data.
3. The method for generating a job based on multi-source data fusion according to claim 2, characterized in that: When standardizing multi-source heterogeneous data, it also includes interpolation and deviation correction of multi-source heterogeneous data. In this process, inverse distance weighted interpolation is used to interpolate multi-source heterogeneous data onto a grid of a specified resolution, and historical data is used to correct the deviation of the current data. During interpolation, the estimated value of the interpolation point is calculated as follows: In the formula, Interpolation point The estimated value at, interpolation point Estimated value of is a known point value The weighted average of the known points and the interpolation points is the inverse of the distance between them. Power; is the weight index; is the total number of known data points involved in the interpolation calculation, From 1 to The index is used to mark each known data point. For the The value of the known point, is the interpolation point and The distance between known points; where In the calculation formula, the numerator For all known point values According to its weight Perform weighted summation, the denominator To calculate the sum of the weights of all known points, the estimated value Divide the numerator by the denominator to get the interpolation point The estimated value at .
4. The method for generating a job based on multi-source data fusion according to claim 3, characterized in that: When extracting semantic information from multi-source data according to the semantic knowledge base, a semantic knowledge base covering the field of multi-source heterogeneous data is selected and constructed based on the existing WordNet, DBpedia and ConceptNet knowledge bases; Stanford NER is used to identify named entities on the standardized multi-source data, and the identified entities are matched with the entries in the semantic knowledge base, word sense disambiguation is performed on the identified entities, the semantic relations between entities and concepts are extracted from the semantic knowledge base, and a semantic graph containing entities and relations is constructed and generated.
5. The method for generating a job based on multi-source data fusion according to claim 4, characterized in that: The word vector model is a pre-trained Word2Vec model, which is obtained by training on a corpus using Word2Vec. During adversarial training, a designed adversarial training framework is also included. The adversarial training framework includes a generator and a discriminator. The generator is responsible for converting the word vector to a shared semantic space, and the discriminator evaluates whether the word vector comes from the same data source.
6. The method for generating a job based on multi-source data fusion according to claim 5, characterized in that: When building a student profile, include: Construct a student portrait that includes multi-dimensional features such as knowledge level, learning habits, and interests and hobbies, decompose the fused multi-source learning data into multi-dimensional feature subsets, and establish a knowledge point mapping matrix; Calculate the mastery of knowledge points and calculate the mastery of each knowledge point by each student; Set a mastery threshold and identify knowledge points whose mastery is below the threshold; Use K-means to cluster the learning interest feature subsets, identify different points of interest, annotate the clustering results, and identify the meaning of each point of interest; Integrate the above calculation results to generate a detailed portrait report for each student.
7. The method for generating a job based on multi-source data fusion according to claim 6, characterized in that: When training the homework instruction model, the generated semantic representation vector is paired with the corresponding semantic label to form a training data set model for training; the homework semantic representation vector is used as input, and the semantic information corresponding to the knowledge weaknesses and learning interest points is used as the label; the Transformer model is selected and the generated training data set is used for model training to obtain the homework instruction model, and the trained model is used to generate personalized homework instructions.
8. A job generation system based on multi-source data fusion, characterized in that: The method for generating a job based on multi-source data fusion according to any one of claims 1 to 7, wherein the system for generating a job based on multi-source data fusion comprises: Multi-source data acquisition module: used to acquire multi-source heterogeneous data and perform standardization on the acquired multi-source heterogeneous data to generate multi-source data that fully covers students' learning situations; Semantic information extraction module: used to convert standardized multi-source data into text data, extract semantic information from multi-source data according to the semantic knowledge base, and project it into a shared semantic space to generate a word vector matrix; wherein, projecting into a shared semantic space to generate a word vector matrix, decomposing the word vector matrix to obtain a semantic representation vector, including: inputting the text data after extracting semantic information into a pre-trained word vector model to generate the corresponding word vector; mapping the word vector to a shared semantic space, and using adversarial training to align word vectors from different sources; combining the aligned word vectors to form a complete word vector matrix; centering the word vector matrix, subtracting the mean of each column, and applying singular value decomposition to decompose the word vector matrix into three matrices , , ; Select the first k singular values and their corresponding singular vectors to obtain a low-dimensional semantic representation vector; use the first k columns in the decomposed U matrix as the low-dimensional semantic representation vector of the word; Word vector decomposition and fusion module: used to decompose the word vector matrix to obtain semantic representation vectors, and fuse these vectors to generate fused multi-source learning data; Student portrait construction module: Based on the integrated multi-source learning data, student portraits are constructed to identify each student's knowledge weaknesses and learning interests; Homework guidance model training module: used to train the homework guidance model by taking the semantic representation vector as input and the corresponding semantic information as label, combined with the student profile and their knowledge weaknesses and learning interests; Online content recognition module: used to traverse the online selected content uploaded by the student end chapter by chapter and identify the keyword data of the selected content; Architecture tree generation module: used to generate a knowledge point architecture tree of the corresponding data database according to the identified keyword data, and filter out resource sub-datasets of the knowledge point architecture tree; Homework generation module: used to filter resource data from the resource sub-dataset based on the homework guidance model and the knowledge weaknesses and learning interests corresponding to the student portrait, and generate personalized homework for each student.
Citation Information
Patent Citations
Question routing method based on word vectors
CN104636456A
Multi-source and multi-dimensional school teaching transverse information difference comparison and analysis method
CN111104455A
Method for establishing student growth portraits based on group sparse fusion hospital big data
CN111652291A
Teaching method and system based on knowledge graph
CN113886567A
Multi-source data fusion method and system based on word vector matrix decomposition technology
CN114169320A
