A course teaching resource recommendation method
By constructing a deep network based on attention mechanism and Transformer model, the text and image features of course content and resources are learned. By combining low-rank matrix and sparse matrix, the problem of low efficiency in university course resource retrieval is solved, and efficient resource recommendation is achieved.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- CHANGSHU INSTITUTE OF TECHNOLOGY
- Filing Date
- 2023-03-21
- Publication Date
- 2026-06-05
AI Technical Summary
The lack of intelligent recommendation systems for teaching resources in existing university courses leads to low efficiency for teachers when browsing or searching, and makes it impossible to effectively utilize massive amounts of multimodal data.
A deep network employing an attention mechanism learns the text and image features of course content and teaching resources, constructs text, image, and image-text similarity matrices, and combines low-rank and sparse matrices to optimize the recommendation model parameters through gradient descent, thereby achieving intelligent recommendation of course resources.
It improves the utilization rate of course teaching resources and achieves efficient resource recommendation through deep learning feature analysis and matrix factorization, meeting the diverse, practical and collaborative needs of the education field.
Smart Images
Figure CN116561410B_ABST
Abstract
Description
Technical Field
[0001] This invention relates to a resource recommendation method, and more particularly to a method for recommending course teaching resources. Background Technology
[0002] Many universities across the country have built a large number of high-quality course teaching resource databases, creating a vast amount of course teaching resources for disciplines, majors, and courses. These resources include large-scale, non-linear, multimodal data such as text, video, images, audio, and documents. However, they lack catalogs and indexes, making manual sequential browsing or retrieval time-consuming, laborious, and inefficient. There is an urgent need for intelligent recommendation systems to provide course teaching resource recommendations to teachers based on course content, thereby enhancing the value of these resources.
[0003] Unlike e-commerce product recommendations, news pushes, music recommendations, movie recommendations, and TV or video recommendations in the internet and other fields, teaching resources in the education field are characterized by diversity, practicality, collaboration, timeliness, and developability, resulting in a limited number of existing teaching resource recommendation systems. Summary of the Invention
[0004] Purpose of the invention: The technical problem to be solved by the present invention is to provide a method for recommending course teaching resources, which addresses the shortcomings of the existing technology.
[0005] To address the aforementioned technical problems, this invention discloses a method for recommending course teaching resources, comprising the following steps:
[0006] Step 1: Establish a course teaching resource recommendation dataset, and learn the text and image features of course content and course teaching resources in the training set; the course teaching resource recommendation dataset includes a training set and a test set;
[0007] Furthermore, step 1 specifically includes the following steps:
[0008] Step 1-1: Establish a recommended dataset for course teaching resources, and divide the dataset into a training set and a test set;
[0009] Steps 1-2: Design a deep network with an attention mechanism. Learn the text features of the course content and teaching resources in the training set from four levels: words, sentences, paragraphs, and documents. That is, divide the course content documents and teaching resource documents in the training set into more than one paragraph, each paragraph into more than one sentence, and each sentence into more than one word. Learn text features based on documents, paragraphs, sentences, and words.
[0010] The deep network with the aforementioned attention mechanism includes:
[0011] The first layer uses a CLIP pre-trained model with an attention mechanism to convert words into word vectors and then normalizes them. The second layer uses a Transformer model with an attention mechanism to convert word vectors in each sentence into sentence vectors with positional information and then normalizes them. The third layer uses a Transformer model to convert each sentence vector into paragraph vectors with sentence positional information and then normalizes them. The fourth layer uses a fully connected operation to convert all paragraph vectors into document feature vectors and then normalizes them.
[0012] Steps 1-3 involve using a deep network with an attention mechanism to learn the image features of the course content and teaching resources in the training set, specifically including:
[0013] The image pixel size in the course content and course teaching resource documents in the training set is uniformly adjusted to 224. 224, according to 16 Divide the image into 16-pixel blocks, and stretch each block into 196 columns. An array of 1s; using the CLIP deep network pre-trained model with attention mechanism, the image is converted into a feature vector and normalized to obtain the image features of course content and course teaching resources.
[0014] Step 2: Based on the text features and image features, construct a similarity matrix for the course content and course teaching resources. The similarity matrix includes: a text similarity matrix, an image similarity matrix, and an image-text pair similarity matrix.
[0015] The construction of the similarity matrix between course content and course teaching resources includes the following steps:
[0016] Step 2-1: Construct a text similarity matrix for course content and teaching resources, including:
[0017] Design course content document and The text feature vectors are respectively and Calculate the cosine similarity metric matrix between the feature vectors of the course text content. , No. row and number Column elements for:
[0018]
[0019] Where n is the number of documents in the training set, and cos() is the cosine similarity measure function of the feature vectors;
[0020] Course teaching resource documents and The text feature vectors are respectively and Calculate the cosine similarity metric matrix between feature vectors of course teaching text resources. , No. row and number Column elements for:
[0021]
[0022] Where m is the number of documents in the course teaching resources of the training set;
[0023] Step 2-2: Construct an image similarity matrix for course content and teaching resources; including:
[0024] Calculate the cosine similarity metric matrix between feature vectors of image content in the course document. , No. 3rd line and the Column elements for:
[0025]
[0026] Calculate the similarity metric matrix between feature vectors of course teaching image resources. , No. row and number Column elements for:
[0027]
[0028] Steps 2-3 involve constructing a text-image similarity matrix for course content and teaching resources, including:
[0029] Calculate the cosine similarity metric matrix between the feature vectors of text and image content in the course document. , No. 5th line and the Column elements for:
[0030]
[0031] Calculate the cosine similarity matrix between the feature vectors of course teaching texts and image resources. , No. row and number Column elements for:
[0032]
[0033] Step 3: Construct a course teaching resource recommendation model based on similarity matrix, low-rank matrix, and sparse matrix; design an objective function based on similarity matrix, low-rank matrix, and sparse matrix; and use gradient descent method to solve for the parameters of the course teaching resource recommendation model.
[0034] Furthermore, step 3 specifically includes the following steps:
[0035] Step 3-1: Construct a course teaching resource recommendation model based on similarity matrices, low-rank matrices, and sparse matrices. Specific methods include:
[0036] The scoring matrix is constructed from the use of course teaching resources in the training set's course content. The scoring matrix row and number Column matrix element values Indicates course teaching resources ( Suitable course content documents ( The degree of ) element value =0 indicates that there are no recommended scores yet; element value When it is non-zero, it indicates the first... The course content uses the first The number of times course teaching resources are listed. Indicates the maximum number of times it can be used;
[0037] Scoring Matrix Decomposed into Low-rank matrix of dimension and sparse matrix of dimension , It is related to the course content. One potential factor, It is related to course teaching resources If there are 10 potential factors, then the scoring matrix will be 100. Approximate estimate The product of the similarity matrix of the course content, the low-rank matrix of the course content, the similarity matrix of the course teaching resources, and the low-rank matrix of the course teaching resources is calculated as follows:
[0038]
[0039] Where T is the matrix transpose, and the similarity matrix of the course content. It is the mean of the text, image, and image-text similarity matrices of the course content, calculated as follows:
[0040]
[0041] Similarity matrix of course teaching resources It is a text, image, and image-text similarity matrix of course teaching resources. , and The mean is calculated as follows:
[0042]
[0043] Step 3-2: Construct the objective function based on the similarity matrix, low-rank matrix, and sparse matrix. Specific methods include:
[0044] Define the scoring matrix and approximate estimation The sum of squared errors is used as the loss term. ,Right now:
[0045]
[0046] in, The Frobenius norm of the matrix;
[0047] In addition, the magnitudes of the low-rank matrix U of the course content and the sparse matrix V of the course teaching resources are added as penalty terms. ,Right now:
[0048]
[0049] Low-rank matrix based on course content and similarity matrix The sparse matrix V and similarity matrix of the course teaching resources Construct a similarity-assisted loss function for:
[0050]
[0051] In summary, the objective function Designed as follows:
[0052]
[0053] in, , and The weight coefficients of the error term, penalty term, and auxiliary term in the loss function are... .
[0054] Step 3-3: Iteratively update and optimize the model parameters using gradient descent. Specific methods include:
[0055] The gradient descent method is used to iteratively optimize and solve for the parameters in the course teaching resource recommendation model, namely the low-rank matrix related to the course content. sparse matrix related to course teaching resources The steps are as follows: Generate random numbers and initialize. and , along the gradient direction with step size Update parameters and Iterate repeatedly until the objective function is achieved. Convergence; objective function For low-rank matrices Find the partial derivative:
[0056]
[0057] objective function For sparse matrices Find the partial derivative:
[0058]
[0059] low-rank matrix The update formula during the iteration process is:
[0060]
[0061] Sparse matrix The update formula during the iteration process is:
[0062]
[0063] Step 4: Extract text and image features from the course content in the test set. Find the corresponding course teaching resources from the approximate estimate of the scoring matrix formed by the frequency of course teaching resources used in the training set. Then, sort the resources according to their scoring items to complete the course teaching resource recommendation. Specific methods include:
[0064] Based on the updated low-rank matrix U and sparse matrix V, the similarity matrix of the course content... Similarity matrix of course teaching resources Calculate the scoring matrix Approximate estimation matrix The scoring matrix is composed of the number of times the course content in the training set uses the course teaching resources.
[0065] Calculate the text and image features of the course content documents in the test set, retrieve the course content in the training set that is most similar to the above features, find the row corresponding to the course content in the scoring matrix, retrieve all columns in the row with a score greater than 0, sort them from high to low score, and recommend the corresponding course teaching resources.
[0066] Beneficial effects:
[0067] This invention addresses the need for course teaching resource recommendation in the education field. Based on deep learning feature analysis of course content and course teaching resources (text and images), it analyzes the similarity relationships between them and combines the low-rank and sparsity of the course content matrix to propose a recommendation method for course teaching resources, thereby improving the utilization rate of course teaching resources. Attached Figure Description
[0068] The present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments, and the advantages of the present invention in the above and / or other aspects will become clearer.
[0069] Figure 1 This is a flowchart of the present invention. Detailed Implementation
[0070] The present invention will be further described below with reference to the accompanying drawings and embodiments.
[0071] This invention relates to a method for recommending course teaching resources. It utilizes techniques such as the CLIP and Transformer models based on attention mechanisms, and recommendation models based on text, image and image-text similarity matrices, low-rank matrices, and sparse matrix factorization. The innovative aspects of this invention include:
[0072] (1) A four-layer network learning text feature of course content and course teaching resources, consisting of documents, paragraphs, sentences and words, was proposed.
[0073] (2) A text, image, and graphic-text similarity matrix was constructed for course content and course teaching resources.
[0074] (3) A course teaching resource recommendation model based on similarity matrix, low-rank matrix and sparse matrix was constructed.
[0075] (4) The error term, penalty term and similarity auxiliary loss term of the objective function were constructed based on the similarity matrix, low-rank matrix and sparse matrix.
[0076] The specific workflow is as follows: Figure 1 As shown.
[0077] Step 1: Extract relevant course content and teaching resources from CNKI (China National Knowledge Infrastructure), university course teaching databases, course teaching competitions, and related internet web pages using manual methods or web scraping software to create a dataset. Randomly select 80% of the course content and teaching resources as the training set, and the remainder as the test set. Construct a scoring matrix based on the course teaching resources used in the course content, initializing it to 0 to indicate no recommended teaching resources. For each instance of a resource being recommended, increment the score of the corresponding element in the scoring matrix by 1.
[0078] Learning and training focus on deep learning features of text and images in course content and teaching resources;
[0079] Step (11): Design a deep network with an attention mechanism to learn course content and text features of course teaching resources at four levels: words, sentences, paragraphs, and documents;
[0080] The deep network for the attention mechanism is as follows:
[0081] Layer 1 uses a CLIP pre-trained model with an attention mechanism (Reference: Learning transferable visual models from natural language supervision[J / OL]. https: / / arxiv.org / abs / 2103.00020v1, 2021-2-26) to convert text words into word vectors and perform normalization. Layer 2 uses a Transformer model with an attention mechanism (Reference: An image is worth 16x16 words: Transformers for image recognition at scale[C]. In proceedings of the 9th International Conference on Learning Representations, Virtual Event, Austria, 2021. DOI:10.48550 / arXiv.2010.11929.) to convert the word vectors of each sentence into sentence vectors with positional information and perform normalization. Layer 3 uses a Transformer model to convert the sentence vectors of each paragraph into paragraph vectors with positional information and perform normalization. Layer 4 uses a fully connected operation to convert all paragraph vectors into document feature vectors and perform normalization.
[0082] Step (12): Use a deep network with an attention mechanism to learn the image features of the course content and course teaching resources;
[0083] The image pixel size in the course documents and course teaching resource documents in the training set is uniformly adjusted to 224. 224, according to 16 Divide the image into 16-pixel blocks, and stretch each block into 196 columns. An array of 1s; using the CLIP pre-trained model to convert the images into feature vectors and perform normalization processing to obtain the image features of course content and course teaching resources.
[0084] Step 2: Construct text similarity matrices, image similarity matrices, and image-text pair similarity matrices for course content and teaching resources;
[0085] Step (21): Construct a text similarity matrix of course content and course teaching resources;
[0086] Assuming course content document and The text feature vectors are respectively and Calculate the cosine similarity metric matrix between the feature vectors of the course content text. , No. row and number Column elements for:
[0087]
[0088] Where n is the number of documents containing course content in the training set, and cos() is the cosine similarity measure function of the feature vectors.
[0089] Course teaching resource documents and The text feature vectors are respectively and Calculate the cosine similarity metric matrix between the feature vectors of course teaching resource texts. , No. row and number Column elements for:
[0090]
[0091] Where m is the number of documents in the course teaching resources of the training set.
[0092] Step (22): Construct an image similarity matrix for course content and course teaching resources;
[0093] Calculate the cosine similarity metric matrix between feature vectors of image content in the course document. , No. row and number Column elements are
[0094]
[0095] Calculate the similarity metric matrix between feature vectors of course teaching image resources. , No. row and number Column elements for:
[0096]
[0097] Step (23): Construct a text-image similarity matrix for course content and teaching resources;
[0098] Calculate the cosine similarity metric matrix between the feature vectors of text and image content in the course document. , No. row and number Column elements are
[0099]
[0100] Calculate the cosine similarity matrix between the feature vectors of the text and image content of the course teaching resources. , No. row and number Column elements are
[0101]
[0102] Step 3: Construct a course teaching resource recommendation model based on similarity matrix, low-rank matrix, and sparse matrix; design an objective function based on similarity matrix, low-rank matrix, and sparse matrix; and solve for the model parameters using gradient descent.
[0103] Step (31): Construct a course teaching resource recommendation model based on similarity matrix, low-rank matrix and sparse matrix;
[0104] The scoring matrix is composed of the course content in the training set and the usage of course teaching resources. The scoring matrix row and number Column matrix element values Indicates course teaching resources ( Suitable course content documents ( The degree of ) element value =0 indicates that there are no recommended scores yet; element value Non-zero indicates the first The course content uses the first The number of times course teaching resources are listed. Represents the maximum number of uses; scoring matrix It can be approximately decomposed into Low-rank matrix of dimension and sparse matrix of dimension , It is related to the course content. One potential factor, It is related to course teaching resources If there are 10 potential factors, then the scoring matrix will be 100. Approximate estimate Similarity matrix of course content Low-rank matrices related to course content Similarity matrix of course teaching resources The low-rank matrix of course teaching resources The product is calculated using the following formula:
[0105]
[0106] Where T is the matrix transpose, and the similarity matrix of the course content. It is the mean of the text, image, and image-text similarity matrices of the course content, calculated as follows:
[0107]
[0108] Similarity matrix of course teaching resources It is a text, image, and image-text similarity matrix of course teaching resources. , and The mean is calculated as follows:
[0109]
[0110] Step (32): Construct an objective function based on the similarity matrix, low-rank matrix, and sparse matrix;
[0111] Define the scoring matrix and approximate estimation The sum of squared errors is used as the loss term. ,Right now:
[0112]
[0113] in, The Frobenius norm of the matrix;
[0114] In addition, the magnitudes of the low-rank matrix U of the course content and the sparse matrix V of the course teaching resources are added as penalty terms. ,Right now:
[0115]
[0116] Low-rank matrix based on course content and similarity matrix The sparse matrix V and similarity matrix related to course teaching resources Construct a similarity-assisted loss function for:
[0117]
[0118] In summary, the objective function Designed as follows:
[0119]
[0120] in, , and The weight coefficients of the error term, penalty term, and auxiliary term in the loss function are... .
[0121] Step (33): Use gradient descent to iteratively update and optimize the model parameters.
[0122] The gradient descent method is used to iteratively optimize and solve for the parameters in the course teaching resource recommendation model, namely the low-rank matrix related to the course content. Sparse matrix related to course teaching resources The steps are as follows: Generate random numbers and initialize. and , along the gradient direction with step size Update parameters and Iterate repeatedly until the objective function is achieved. Convergence; objective function For low-rank matrices Find the partial derivative:
[0123]
[0124] objective function For sparse matrices Find the partial derivative:
[0125]
[0126] low-rank matrix The update formula during the iteration process is:
[0127]
[0128] Sparse matrix The update formula during the iteration process is:
[0129]
[0130] Step 4: Based on the updated low-rank matrix U and sparse matrix V, the similarity matrix of the course content... Similarity matrix of course teaching resources Calculate the scoring matrix Approximate estimation matrix ;
[0131] Calculate the text and image features of the course content documents in the test set, retrieve the course content in the training set that is most similar to the above features, find the row corresponding to the course content in the scoring matrix, retrieve all columns in the row with a score greater than 0, and recommend the corresponding course teaching resources in descending order of score.
[0132] In its specific implementation, this application provides a computer storage medium and a corresponding data processing unit. The computer storage medium is capable of storing a computer program, which, when executed by the data processing unit, can run the content of the course teaching resource recommendation method provided by this invention and some or all of the steps in various embodiments. The storage medium can be a magnetic disk, optical disk, read-only memory (ROM), or random access memory (RAM), etc.
[0133] Those skilled in the art will clearly understand that the technical solutions in the embodiments of the present invention can be implemented using computer programs and their corresponding general-purpose hardware platforms. Based on this understanding, the technical solutions in the embodiments of the present invention, or the parts that contribute to the prior art, can be embodied in the form of computer programs, i.e., software products. These computer program software products can be stored in a storage medium and include several instructions to cause a device containing a data processing unit (which may be a personal computer, server, microcontroller, MUU, or network device, etc.) to execute the methods described in various embodiments or certain parts of the embodiments of the present invention.
[0134] This invention provides an idea and method for recommending course teaching resources. Many methods and approaches exist for implementing this technical solution; the above description is merely a preferred embodiment of the invention. It should be noted that those skilled in the art can make various improvements and modifications without departing from the principles of this invention, and these improvements and modifications should also be considered within the scope of protection of this invention. All components not explicitly stated in this embodiment can be implemented using existing technologies.
Claims
1. A method for recommending course teaching resources, characterized in that, Includes the following steps: Step 1: Establish a course teaching resource recommendation dataset, and learn the text and image features of course content and course teaching resources in the training set; the course teaching resource recommendation dataset includes a training set and a test set; Step 2: Based on the text features and image features, construct a similarity matrix for the course content and course teaching resources. The similarity matrix includes: a text similarity matrix, an image similarity matrix, and an image-text pair similarity matrix. Step 3: Construct a course teaching resource recommendation model based on similarity matrix, low-rank matrix, and sparse matrix; design an objective function based on similarity matrix, low-rank matrix, and sparse matrix; and use gradient descent method to solve for the parameters of the course teaching resource recommendation model. Step 4: Extract text and image features of course content in the test set, find corresponding course teaching resources from the approximate estimate of the scoring matrix formed by the number of times course teaching resources are selected in the course content of the training set, and complete the course teaching resource recommendation according to the scoring items. Step 3 includes the following steps: Step 3-1: Construct a course teaching resource recommendation model based on similarity matrix, low-rank matrix, and sparse matrix; Step 3-2: Construct an objective function based on the similarity matrix, low-rank matrix, and sparse matrix; Step 3-3: Use gradient descent to iteratively update and optimize the model parameters. Step 3-1 describes the construction of a course and teaching resource recommendation model based on similarity matrices, low-rank matrices, and sparse matrices. The specific methods include: The scoring matrix is constructed from the use of course teaching resources in the training set's course content. The scoring matrix row and number Column matrix element values Indicates course teaching resources ( Suitable course content documents ( The degree of ) element value =0 indicates that there are no recommended scores yet; element value When it is non-zero, it indicates the first... The course content uses the first The number of times course teaching resources are listed. Indicates the maximum number of times it can be used; Scoring Matrix Decomposed into Low-rank matrix of dimension and sparse matrix of dimension , It is related to the course content. One potential factor, It is related to course teaching resources If there are 10 potential factors, then the scoring matrix will be 100. Approximate estimate The product of the similarity matrix of the course content, the low-rank matrix of the course content, the similarity matrix of the course teaching resources, and the low-rank matrix of the course teaching resources is calculated as follows: ； Where T is the matrix transpose. This is a similarity matrix of course teaching resources. This is the similarity matrix of the course content; It is the mean of the text, image, and image-text similarity matrices of the course content, calculated as follows: ； Similarity matrix of course teaching resources It is a text, image, and image-text similarity matrix of course teaching resources. , and The mean is calculated as follows: ； Step 3-2 describes the construction of the objective function based on the similarity matrix, low-rank matrix, and sparse matrix. The specific methods include: Define the scoring matrix and approximate estimation The sum of squared errors is used as the loss term. ,Right now: ； in, The Frobenius norm of the matrix; In addition, the magnitudes of the low-rank matrix U of the course content and the sparse matrix V of the course teaching resources are added as penalty terms. ,Right now: ； Low-rank matrix based on course content and similarity matrix The sparse matrix V and similarity matrix of the course teaching resources Construct a similarity-assisted loss function for: ； In summary, the objective function Designed as follows: ； in, , and The weight coefficients of the error term, penalty term, and auxiliary term in the loss function are... ; Step 3-3 describes the iterative update and optimization of the model parameters using gradient descent, specifically including: The gradient descent method is used to iteratively optimize and solve for the parameters in the course teaching resource recommendation model, namely the low-rank matrix related to the course content. sparse matrix related to course teaching resources The steps are as follows: Generate random numbers and initialize. and , along the gradient direction with step size Update parameters and Iterate repeatedly until the objective function is achieved. Convergence; objective function For low-rank matrices Find the partial derivative: ； objective function For sparse matrices Find the partial derivative: ； low-rank matrix The update formula during the iteration process is: ； Sparse matrix The update formula during the iteration process is: 。 2. The method for recommending course teaching resources according to claim 1, characterized in that, Step 1 specifically includes the following steps: Step 1-1: Establish a recommended dataset for course teaching resources, and divide the dataset into a training set and a test set; Steps 1-2: Design a deep network with an attention mechanism. Learn the text features of the course content and teaching resources in the training set from four levels: words, sentences, paragraphs, and documents. That is, divide the course content documents and teaching resource documents in the training set into more than one paragraph, each paragraph into more than one sentence, and each sentence into more than one word. Learn text features based on documents, paragraphs, sentences, and words. Steps 1-3: Using a deep network with an attention mechanism, learn the image features of the course content and course teaching resources in the training set.
3. The method for recommending course teaching resources according to claim 2, characterized in that, The deep network with the attention mechanism described in steps 1-2 includes: The first layer uses a CLIP pre-trained model with an attention mechanism to convert words into word vectors and then normalizes them. The second layer uses a Transformer model with an attention mechanism to convert word vectors in each sentence into sentence vectors with positional information and then normalizes them. The third layer uses a Transformer model to convert each sentence vector into paragraph vectors with sentence positional information and then normalizes them. The fourth layer uses a fully connected operation to convert all paragraph vectors into document feature vectors and then normalizes them.
4. The method for recommending course teaching resources according to claim 3, characterized in that, The image features of the course content and teaching resources in the learning and training set mentioned in steps 1-3 specifically include: The image pixel size in the course content and course teaching resource documents in the training set is uniformly adjusted to 224. 224, according to 16 Divide the image into 16-pixel blocks, and stretch each block into 196 columns. An array of 1s; using the CLIP deep network pre-trained model with attention mechanism, the image is converted into a feature vector and normalized to obtain the image features of course content and course teaching resources.
5. The method for recommending course teaching resources according to claim 4, characterized in that, Step 2, which involves constructing a similarity matrix between course content and teaching resources, includes the following steps: Step 2-1: Construct a text similarity matrix for course content and teaching resources, including: Design course content document and The text feature vectors are respectively and Calculate the cosine similarity metric matrix between the feature vectors of the course text content. , No. row and number Column elements for: ； Where n is the number of documents in the training set, and cos() is the cosine similarity measure function of the feature vectors; Course teaching resource documents and The text feature vectors are respectively and Calculate the cosine similarity metric matrix between feature vectors of course teaching text resources. , No. row and number Column elements for: ； Where m is the number of documents in the course teaching resources of the training set; Step 2-2: Construct an image similarity matrix for course content and teaching resources; including: Calculate the cosine similarity metric matrix between feature vectors of image content in the course document. , No. 3rd line and the Column elements for: ； Calculate the similarity metric matrix between feature vectors of course teaching image resources. , No. row and number Column elements for: ； Steps 2-3 involve constructing a text-image similarity matrix for course content and teaching resources, including: Calculate the cosine similarity metric matrix between the feature vectors of text and image content in the course document. , No. 5th line and the Column elements for: ； Calculate the cosine similarity matrix between the feature vectors of course teaching texts and image resources. , No. row and number Column elements for: 。 6. The method for recommending course teaching resources according to claim 5, characterized in that, Step 4, which involves recommending course teaching resources, includes the following specific methods: Based on the updated low-rank matrix U and sparse matrix V, the similarity matrix of the course content... Similarity matrix of course teaching resources Calculate the scoring matrix Approximate estimation matrix The scoring matrix is composed of the number of times the course content in the training set uses the course teaching resources. Calculate the text and image features of the course content documents in the test set, retrieve the course content in the training set that is most similar to the above features, find the row corresponding to the course content in the scoring matrix, retrieve all columns in the row with a score greater than 0, sort them from high to low score, and recommend the corresponding course teaching resources.