Learning path recommendation method and system based on big data analysis

Abstract

The invention relates to the field of intelligent recommendation, and particularly discloses a learning path recommendation method and system based on big data analysis, and the method comprises the steps: obtaining the multi-dimensional data of a learner, and the text introduction and course information of a to-be-recommended learning course through employing an artificial intelligence technology based on a deep neural network model; different text data are mapped into a semantic space through a semantic encoder to extract key information of a learner, weighting is carried out through a bidirectional attention mechanism module, key matching information is emphasized, and a classification result used for representing whether a course is recommended to the learner or not is obtained. Therefore, recommended courses meeting interests and requirements of the learners can be further provided for the learners, and the satisfaction degree and the learning effect of the learners are improved.

Description

Technical Field

[0001] This application relates to the field of intelligent recommendation, and more specifically, to a learning path recommendation method and system based on big data analysis. Background Art

[0002] With the continuous rise of the Internet, online education and online learning have become a trend, providing great convenience for learners. The courses in online learning systems are more abundant, more targeted, and more complete. The learning system can recommend corresponding learning courses for learners according to their usual subject preferences, knowledge levels, learning goals and other characteristics. Learners can find the courses they are interested in for learning and practice without being restricted by time and space, master key knowledge, and improve learning efficiency and effect. However, due to the inaccurate learning path recommendation of learning websites in the prior art, the system recommends irrelevant courses, wasting learners' time, reducing learners' learning efficiency, and reducing learners' efficiency and experience.

[0003] Therefore, an optimized learning path recommendation scheme based on big data analysis is expected. Summary of the Invention

[0004] To solve the above technical problems, this application is proposed. Embodiments of this application provide a learning path recommendation method and system based on big data analysis, which adopt artificial intelligence technology based on a deep neural network model, obtain multi-dimensional data of learners and text introductions and course information of learning courses to be recommended, map different text data to a semantic space through a semantic encoder to extract key information of learners, and then perform weighting through a bidirectional attention mechanism module to emphasize key matching information to obtain a classification result for indicating whether to recommend the course to the learner. In this way, recommended courses that meet the interests and needs of learners can be further provided for learners, improving learners' satisfaction and learning effect.

[0005] According to one aspect of this application, a learning path recommendation method based on big data analysis is provided, which includes: Obtain multi-dimensional data of learners, where the multi-dimensional data of learners includes learning history, hobbies, and learning goals; After cleaning the multi-dimensional data of the learners, pass it through a first semantic encoder including a word embedding layer to obtain multiple learner data feature vectors and concatenate them to obtain a learner data semantic feature vector; Obtain the text introduction and course information of the learning course to be recommended; After performing word segmentation on the text introduction and course information of the learning course to be recommended, pass it through a second semantic encoder including a word embedding layer to obtain a semantic feature vector of the course to be recommended; Perform weighted fusion on the semantic feature vector of the learner data and the semantic feature vector of the course to be recommended to obtain a course-learner association feature vector; Perform potential kernel space remapping calibration based on modal decomposition on the course-learner association feature vector to obtain a course-learner matching feature vector; Calculate the product of the course-learner matching feature vector and its own transpose to obtain the course-learner matching feature matrix; Pass the course-learner matching feature matrix through a bidirectional attention mechanism module to obtain a classification feature matrix; Pass the classification feature matrix through a classifier to obtain a classification result, and the classification result is used to indicate whether to recommend the course to the learner.

[0006] In the above learning path recommendation method based on big data analysis, after cleaning the learner multi-dimensional data, pass it through a first semantic encoder including a word embedding layer to obtain multiple learner data feature vectors and concatenate them to obtain a learner data semantic feature vector, including: passing the learner multi-dimensional data through the embedding layer of the first semantic encoder including the word embedding layer to convert each dimension data in the multi-dimensional data into an embedding vector to obtain a sequence of embedding vectors, where the embedding layer uses a learnable embedding matrix to perform embedding encoding on each dimension data; and inputting the sequence of embedding vectors into the first semantic encoder including the word embedding layer to obtain the multiple learner data feature vectors.

[0007] In the above learning path recommendation method based on big data analysis, inputting the sequence of embedding vectors into the first semantic encoder including the word embedding layer to obtain the multiple learner data feature vectors includes: arranging the sequence of embedding vectors as an input vector; converting the input vector into a query vector and a key vector respectively through a learnable embedding matrix; calculating the product between the query vector and the transpose vector of the key vector to obtain a self-attention association matrix; performing normalization processing on the self-attention association matrix to obtain a normalized self-attention association matrix; inputting the normalized self-attention association matrix into a Softmax activation function for activation to obtain a self-attention feature matrix; and multiplying the self-attention feature matrix with each embedding vector in the sequence of embedding vectors as a value vector respectively to obtain the multiple learner data feature vectors.

[0008] In the above learning path recommendation method based on big data analysis, after performing word segmentation on the text introduction and course information of the learning course to be recommended, a second semantic encoder including a word embedding layer is used to obtain a semantic feature vector of the course to be recommended, including: performing word segmentation on the text introduction and course information of the learning course to be recommended to convert the text introduction and course information of the learning course to be recommended into a word sequence composed of multiple words; using the embedding layer of the second semantic encoder including the embedding layer to map each word in the word sequence into a word embedding vector respectively to obtain a sequence of learning course word embedding vectors; using the transformer of the second semantic encoder including the embedding layer to perform global context semantic encoding based on the transformer idea on the sequence of learning course word embedding vectors to obtain multiple context semantic feature vectors; and concatenating the multiple context semantic feature vectors to obtain the semantic feature vector of the course to be recommended.

[0009] In the above learning path recommendation method based on big data analysis, using the transformer of the second semantic encoder including the embedding layer to perform global context semantic encoding based on the transformer idea on the sequence of learning course word embedding vectors to obtain multiple context semantic feature vectors, including: arranging the sequence of learning course word embedding vectors in one dimension to obtain a global feature vector; calculating the product between the global feature vector and the transposed vectors of each vector in the sequence of learning course word embedding vectors to obtain multiple self-attention correlation matrices; respectively performing normalization processing on each self-attention correlation matrix in the multiple self-attention correlation matrices to obtain multiple normalized self-attention correlation matrices; passing each normalized self-attention correlation matrix in the multiple normalized self-attention correlation matrices through the Softmax classification function to obtain multiple probability values; and respectively using each probability value in the multiple probability values as a weight to weight each vector in the sequence of learning course word embedding vectors to obtain the multiple context semantic feature vectors.

[0010] In the above learning path recommendation method based on big data analysis, passing the course-learner matching feature matrix through a bidirectional attention mechanism module to obtain a classification feature matrix, including: performing pooling on the course-learner matching feature matrix along the horizontal direction and the vertical direction respectively to obtain a first-direction pooling vector and a second-direction pooling vector; performing correlation encoding on the first-direction pooling vector and the second-direction pooling vector to obtain a bidirectional correlation matrix; inputting the bidirectional correlation matrix into the Sigmoid activation function to obtain a bidirectional correlation weight matrix; and calculating the element-wise product between the bidirectional correlation weight matrix and the course-learner matching feature matrix to obtain the classification feature matrix.

[0011] According to another aspect of the present application, a learning path recommendation system based on big data analysis is provided, which includes: A learner data acquisition module for acquiring multi-dimensional data of learners, where the multi-dimensional data of learners includes learning history, hobbies, and learning goals; A learner data semantic extraction module for performing data cleaning on the multi-dimensional data of learners and then obtaining multiple learner data feature vectors through a first semantic encoder including a word embedding layer and concatenating them to obtain a learner data semantic feature vector; A learning course information acquisition module for acquiring the text introduction and course information of the learning courses to be recommended; A learning course semantic extraction module for performing word segmentation on the text introduction and course information of the learning courses to be recommended and then obtaining a semantic feature vector of the course to be recommended through a second semantic encoder including a word embedding layer; An interpolation and ordering fusion module for performing weighted fusion on the learner data semantic feature vector and the semantic feature vector of the course to be recommended to obtain a course-learner association feature vector; A feature remapping and calibration module for performing potential kernel space remapping and calibration based on modal decomposition on the course-learner association feature vector to obtain a course-learner matching feature vector; A matching module for obtaining the course-learner matching feature matrix by calculating the product of the course-learner matching feature vector and its own transpose; A bidirectional attention module for obtaining a classification feature matrix by passing the course-learner matching feature matrix through a bidirectional attention mechanism module; A recommendation result judgment module for obtaining a classification result by passing the classification feature matrix through a classifier, and the classification result is used to indicate whether to recommend the course to the learner.

[0012] Compared with the prior art, a learning path recommendation method and system based on big data analysis provided by the present application adopt artificial intelligence technology based on a deep neural network model, acquire multi-dimensional data of learners and the text introduction and course information of the learning courses to be recommended, map different text data into the semantic space through a semantic encoder to extract the key information of learners, and then perform weighting through a bidirectional attention mechanism module to emphasize the key matching information, so as to obtain a classification result for indicating whether to recommend the course to the learner. In this way, recommended courses that meet the interests and needs of learners can be further provided for learners, improving the satisfaction and learning effect of learners. BRIEF DESCRIPTION OF THE DRAWINGS

[0013] The above and other objects, features, and advantages of the present application will become more apparent by describing the embodiments of the present application in more detail with reference to the accompanying drawings. The drawings are used to provide a further understanding of the embodiments of the present application and constitute a part of the specification. Together with the embodiments of the present application, they are used to explain the present application and do not constitute a limitation to the present application. In the drawings, the same reference numerals generally represent the same components or steps.

[0014] Figure 1 It is a flowchart of a learning path recommendation method based on big data analysis according to an embodiment of the present application.

[0015] Figure 2 It is a schematic structural diagram of a learning path recommendation method based on big data analysis according to an embodiment of the present application.

[0016] Figure 3 It is a flowchart of inputting a sequence of the embedding vectors into a first semantic encoder including a word embedding layer to obtain a plurality of learner data feature vectors in a learning path recommendation method based on big data analysis according to an embodiment of the present application.

[0017] Figure 4 It is a block diagram of a learning path recommendation system based on big data analysis according to an embodiment of the present application.

[0018] Figure 5 It is a block diagram of an electronic device according to an embodiment of the present application. Detailed Embodiments

[0019] Next, exemplary embodiments according to the present application will be described in detail with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all embodiments of the present application. It should be understood that the present application is not limited by the exemplary embodiments described herein.

[0020] Figure 1 It is a flowchart of a learning path recommendation method based on big data analysis according to an embodiment of the present application. Figure 2 It is a schematic structural diagram of a learning path recommendation method based on big data analysis according to an embodiment of the present application. As Figure 1 and Figure 2As shown, according to an embodiment of the present application, it includes: S110, obtaining multi-dimensional data of the learner, where the multi-dimensional data of the learner includes learning history, hobbies, and learning goals; S120, performing data cleaning on the multi-dimensional data of the learner and then passing it through a first semantic encoder including a word embedding layer to obtain multiple learner data feature vectors and concatenating them to obtain a learner data semantic feature vector; S130, obtaining the text introduction and course information of the learning course to be recommended; S140, performing word segmentation on the text introduction and course information of the learning course to be recommended and then passing it through a second semantic encoder including a word embedding layer to obtain a semantic feature vector of the course to be recommended; S150, performing weighted fusion on the learner data semantic feature vector and the semantic feature vector of the course to be recommended to obtain a course-learner association feature vector; S160, performing potential kernel space remapping calibration based on modal decomposition on the course-learner association feature vector to obtain a course-learner matching feature vector; S170, calculating the product of the course-learner matching feature vector and its own transpose to obtain the course-learner matching feature matrix; S180, passing the course-learner matching feature matrix through a bidirectional attention mechanism module to obtain a classification feature matrix; and S190, passing the classification feature matrix through a classifier to obtain a classification result, where the classification result is used to indicate whether to recommend the course to the learner.

[0021] In step S110, multi-dimensional data of the learner is obtained, where the multi-dimensional data of the learner includes learning history, hobbies, and learning goals. Considering that the learning history of the learner can provide valuable information, including the courses already studied, the fields and knowledge points learned, etc. By analyzing the learning history, the knowledge level and interest preferences of the learner can be understood, providing a basis for recommending relevant courses. In addition, the hobbies of the learner are important factors determining their learning motivation and preferences. By understanding the hobbies of the learner, courses related to their interests can be recommended, improving the initiative and participation of the learner. Furthermore, the learning goals of the learner can reflect their learning needs and expectations. Understanding the learning goals of the learner can recommend courses related to the goals targeted, helping the learner achieve their learning goals. Therefore, by obtaining multi-dimensional data, the system can establish a user model of the learner, matching and comparing the characteristics and needs of the learner with the learning courses to be recommended. This can provide personalized course recommendations for the learner, improving the accuracy and satisfaction of the recommendations. At the same time, the acquisition of multi-dimensional data can also provide more comprehensive and integrated learning support for the learner, promoting the improvement of their learning effect.

[0022] In step S120, after cleaning the multi-dimensional data of the learner, a first semantic encoder including a word embedding layer is used to obtain multiple learner data feature vectors and concatenate them to obtain a learner data semantic feature vector. Considering that the multi-dimensional data of the learner is usually unstructured text data, which contains rich information. Through the semantic encoder, the text data can be transformed into low-dimensional semantic feature vectors, extracting the key information therein and reducing the dimension of the data. This can reduce the redundancy of the data and better represent the characteristics and needs of the learner. Specifically, the semantic encoder can map different text data to vector representations in the semantic space. By calculating the similarity or distance between vectors, the semantic similarity between learners can be measured. This helps the recommendation system to find learners similar to a given learner according to their characteristics and interests, and recommend the courses they like. In addition, through the data cleaning process of the semantic encoder, the influence of these interference factors can be reduced, improving the quality and consistency of the data. This can more accurately represent the characteristics and needs of the learner, improving the accuracy and reliability of the recommendation system.

[0023] Specifically, in the embodiment of the present application, after cleaning the multi-dimensional data of the learner, a first semantic encoder including a word embedding layer is used to obtain multiple learner data feature vectors and concatenate them to obtain a learner data semantic feature vector, including: passing the multi-dimensional data of the learner through the embedding layer of the first semantic encoder including a word embedding layer to transform each dimension data in the multi-dimensional data into an embedding vector to obtain a sequence of embedding vectors, where the embedding layer uses a learnable embedding matrix to perform embedding coding on each dimension data; and inputting the sequence of embedding vectors into the first semantic encoder including a word embedding layer to obtain the multiple learner data feature vectors.

[0024] Figure 3 The flowchart for inputting the sequence of embedding vectors into the first semantic encoder including a word embedding layer to obtain the multiple learner data feature vectors in the learning path recommendation method based on big data analysis according to the embodiment of the present application. More specifically, in the embodiment of the present application, as Figure 3As shown, inputting the sequence of the embedding vectors into the first semantic encoder including a word embedding layer to obtain the multiple learner data feature vectors includes: S210, arranging the sequence of the embedding vectors into an input vector; S220, respectively transforming the input vector into a query vector and a key vector through a learnable embedding matrix; S230, calculating the product between the query vector and the transposed vector of the key vector to obtain a self-attention correlation matrix; S240, performing a normalization process on the self-attention correlation matrix to obtain a normalized self-attention correlation matrix; S250, inputting the normalized self-attention correlation matrix into a Softmax activation function for activation to obtain a self-attention feature matrix; and S260, multiplying the self-attention feature matrix with each embedding vector in the sequence of the embedding vectors as value vectors respectively to obtain the multiple learner data feature vectors.

[0025] In step S130, obtain the text introduction and course information of the learning course to be recommended. Considering that the text introduction and course information of the learning course can provide detailed descriptions about the course content, objectives, and learning resources. These information can help the recommendation system understand the characteristics of the course in aspects such as the theme, difficulty, learning objectives, etc., so as to match the needs of the learners. Specifically, the interests of the learners are one of the important factors determining their course selection. By analyzing the text introduction and course information of the learning course, the theme, field, and characteristics of the course can be understood, so as to match the interests of the learners. This helps to recommend courses related to the learners' interests and improve the learners' participation and satisfaction. In addition, the text introduction and course information of the learning course can also provide the attributes and characteristics of the course, such as the course type, learning method, learning time, etc. These information can help the recommendation system recommend courses that meet the requirements according to the needs and constraints of the learners.

[0026] In step S140, after performing word segmentation on the text introduction and course information of the learning course to be recommended, pass them through the second semantic encoder including a word embedding layer to obtain the semantic feature vectors of the course to be recommended. Considering that the text introduction and course information of the learning course to be recommended are usually unstructured text data, which contains rich information. Through the semantic encoder, the text data can be transformed into low-dimensional semantic feature vectors, extracting the key information therein and reducing the dimension of the data. This can reduce the redundancy of the data and better represent the characteristics and content of the course. In addition, there may be noise, errors, or inconsistent information in the text introduction and course information of the learning course to be recommended. Through the data processing process of the semantic encoder, the influence of these interference factors can be reduced, improving the quality and consistency of the data.

[0027] Specifically, in the embodiments of the present application, after performing word segmentation on the text introduction and course information of the learning course to be recommended, a second semantic encoder including a word embedding layer is used to obtain a semantic feature vector of the course to be recommended, including: performing word segmentation on the text introduction and course information of the learning course to be recommended to convert the text introduction and course information of the learning course to be recommended into a word sequence composed of multiple words; using the embedding layer of the second semantic encoder including the embedding layer to map each word in the word sequence into a word embedding vector respectively to obtain a sequence of learning course word embedding vectors; using the transformer of the second semantic encoder including the embedding layer to perform global context semantic encoding based on the idea of the transformer on the sequence of learning course word embedding vectors to obtain multiple context semantic feature vectors; and concatenating the multiple context semantic feature vectors to obtain the semantic feature vector of the course to be recommended.

[0028] More specifically, in the embodiments of the present application, using the transformer of the second semantic encoder including the embedding layer to perform global context semantic encoding based on the idea of the transformer on the sequence of learning course word embedding vectors to obtain multiple context semantic feature vectors, including: arranging the sequence of learning course word embedding vectors in one dimension to obtain a global feature vector; calculating the product between the global feature vector and the transposed vectors of each vector in the sequence of learning course word embedding vectors to obtain multiple self-attention correlation matrices; respectively performing normalization processing on each self-attention correlation matrix in the multiple self-attention correlation matrices to obtain multiple normalized self-attention correlation matrices; passing each normalized self-attention correlation matrix in the multiple normalized self-attention correlation matrices through the Softmax classification function to obtain multiple probability values; and respectively using each probability value in the multiple probability values as a weight to weight each vector in the sequence of learning course word embedding vectors to obtain the multiple context semantic feature vectors.

[0029] In step S150, the semantic feature vector of the learner data and the semantic feature vector of the course to be recommended are weighted and fused to obtain a course-learner association feature vector. It should be understood that since the semantic feature vector of the learner data contains multi-dimensional key information such as the learning history, hobbies, and learning goals of the learner, and the semantic feature vector of the course to be recommended contains content features such as the text introduction and course information of the learning course to be recommended, and the two perform semantic representations from the perspectives of learner needs and course content respectively. Therefore, when analyzing the association relationship between the learner and the course, it is necessary to integrate these two types of features belonging to different subjects. Therefore, they are weighted and fused, and by assigning appropriate weights to different feature dimensions, the key matching information is highlighted, so as to obtain a course-learner association feature vector that can comprehensively reflect the association degree between the learner and the course, providing a basis for further analyzing the matching relationship between the two.

[0030] In step S160, the potential kernel space remapping calibration based on modal decomposition is performed on the course-learner association feature vector to obtain the course-learner matching feature vector. It should be understood that in the process of weighted fusion of the learner data semantic feature vector and the course semantic feature vector to be recommended to obtain the course-learner association feature vector, the structural information inside the original features may be ignored due to direct linear combination. In this case, simply relying on weighted fusion may not be able to fully capture the deep association between learner preferences and course characteristics, especially when these features have complex non-linear relationships. Since data from different sources (such as learning history, hobbies, etc.) may be in different feature spaces, the information density and importance they carry are also different, and direct fusion may cause some key feature information to be submerged or weakened, thus affecting the accuracy of the final matching result.

[0031] The potential kernel space remapping calibration method based on modal decomposition can effectively solve the above problems. By identifying and separating various modal components in the course-learner association feature vector, this method can deeply explore the complex structural information hidden inside. While maintaining the original feature expression ability, this method can map data of different modalities to a unified potential kernel space, where more accurate feature calibration and recombination are carried out. Therefore, by performing modal decomposition-based processing on the course-learner association feature vector, not only can the understanding of the learner's true preferences be enhanced, but also the accuracy and personalization level of course recommendation can be improved, making the finally generated course-learner matching feature vector more accurately reflect the matching degree between the two.

[0032] Specifically, performing the potential kernel space remapping calibration based on modal decomposition on the course-learner association feature vector to obtain the course-learner matching feature vector includes: First, constructing the cross-pixel fine-grained association matrix of the course-learner association feature vector, which is expressed by the formula: ; where, represents the course-learner association feature vector, and respectively represent the feature values at the th and th positions of the course-learner association feature vector, represents calculating the Euclidean distance, represents the feature value at the position of the cross-pixel fine-grained association matrix.

[0033] That is, by establishing a cross-pixel granularity correlation mapping, the course learner association feature vector and the course feature vector are transformed into an analytic matrix form. This fine-grained correlation not only focuses on the corresponding relationship of a single feature dimension, but also emphasizes the cross-domain synergy effect between different feature units. For example, a certain knowledge weakness of a learner may trigger changes in the correlation weights of multiple course knowledge modules, or a specific interest tag may activate the implicit progressive learning needs in the course system through the correlation path. The generated cross-pixel fine-grained correlation matrix breaks through the dependence on the orthogonal hypothesis of traditional dot-product similarity calculation, captures the potential correlations in non-orthogonal directions in the feature space, provides a structured prior for the subsequent attention mechanism, and makes the recommendation decision conform to both the global matching rule and the local feature synergy effect.

[0034] Then, based on the convolutional layer, deep kernel feature mining is performed on the cross-pixel fine-grained correlation matrix to obtain the dynamic response matrix of the potential kernel space of the course learner association feature, which is expressed by the formula: ; Where, represents the cross-pixel fine-grained correlation matrix, represents the convolutional layer, represents the dynamic response matrix of the potential kernel space of the course learner association feature.

[0035] That is, by designing a depthwise separable convolution architecture, a multi-order correlation transfer path is established on the cross-pixel fine-grained correlation matrix. Specifically, the convolutional kernel is no longer limited to extracting a single correlation topology at this stage, but gradually expands the receptive field through stacked dilated convolution layers, thereby identifying the composite structure of "local correlation clusters - global correlation chains". A dynamic response matrix of the potential kernel space of the course learner association feature that can dynamically reflect the complex interaction relationship between features is formed, providing a richer and task-specific feature basis for the subsequent potential kernel space remapping and calibration.

[0036] Next, orthogonal basis decomposition is performed on the cross-pixel fine-grained correlation matrix to obtain a set of course learner association feature component representation vectors, which is expressed by the formula: ; Where, represents the transpose of the vector, represents the diagonal matrix, and respectively represent the first and the th eigenvalues on the diagonal of the diagonal matrix, represents the set of course learner association feature component representation vectors, , , respectively represent the first, second, and representation vectors of the course learner association feature components.

[0037] That is, by means of orthogonal basis decomposition technology, the high-dimensional cross-pixel fine-grained association matrix is mapped into a linear space spanned by orthogonal basis vectors, so that these orthogonal bases, as "basic atomic patterns", can capture the main change directions of data correlation, and thus decompose the complex association structure into a set of independent representation vectors of course learner association feature components, providing a concise and representative feature basis for subsequent feature processing. In this way, the core feature components that can best reflect the essence of the association between learners and courses can be extracted from the high-dimensional association data, redundant information can be removed, and the key structure can be retained, providing a low-dimensional, orthogonal, and physically meaningful feature representation for subsequent potential kernel space remapping or attention mechanism processing.

[0038] Then, each representation vector of the course learner association feature components in the set of the representation vectors of the course learner association feature components is input into the dynamic focusing of feature importance based on the self-attention mechanism to obtain a set of representation vectors of the course learner association feature mode components, which is expressed by the formula: ; where represents a sequence model based on the self-attention mechanism, represents the set of representation vectors of the course learner association feature mode components, , , respectively represent the first, second, and representation vectors of the course learner association feature mode components.

[0039] That is, through the self-attention mechanism, the relative importance of each feature component in the global structure is dynamically evaluated, breaking the limitations of traditional fixed weights, enabling the model to adaptively enhance the core structure patterns highly relevant to the learner's needs according to the feature correlation context of the current data instance, and suppressing secondary or irrelevant feature components. In this way, the model can dynamically adjust the attention to the feature components according to the personalized data features of different learners, form a feature representation more suitable for the specific recommendation scenario, and provide a more discriminative input for subsequent matching feature calibration or classification decision-making.

[0040] Next, each representation vector of the course learner association feature mode components in the set of the representation vectors of the course learner association feature mode components is embedded into the dynamic response matrix of the course learner association feature potential kernel space to obtain a set of course learner association feature kernel space masked feature vectors, which is expressed by the formula: ; Among them, represents matrix multiplication, represents the feature scale of the dynamic response matrix of the potential kernel space of the course learner associated features, represents the th course learner associated feature modal component representation vector, represents the length of the course learner associated feature modal component representation vector, represents the th course learner associated feature kernel space masking feature vector.

[0041] That is, non-linearly couple the dynamically focused course learner associated feature modal component representation vector with the dynamically responsive matrix of the potential kernel space of the course learner associated features mined through the convolutional layer, so that the former can perceive and incorporate the high-order data dependencies encoded by the latter, thereby generating a course learner associated feature kernel space masking feature vector with both global structural basis and deep context dependencies, enabling the model to capture the matching points between learner needs and course content from the dual perspectives of global structure and local details, and enhancing the adaptability to complex feature association scenarios.

[0042] Finally, fuse the set of the course learner associated feature kernel space masking feature vectors to obtain the course-learner matching feature vector, which is expressed by the formula: ; Among them, represents the concatenation function, , , respectively represent the first, second, and th course learner associated feature kernel space masking feature vectors, represents the course-learner matching feature vector.

[0043] That is, through specific operations, converge into a course-learner matching feature vector with both global structural integrity and local detail richness, enabling it to comprehensively represent the complex matching relationship between the learner and the course. The fused course-learner matching feature vector can effectively integrate the key information extracted in different processing stages, including both the core association patterns refined by orthogonal basis decomposition (such as basic structures like global average association and contrast association), and the high-order local interaction relationships learned by the convolutional kernel (such as complex dependency topologies between feature positions), while retaining the course learner associated feature kernel space masking feature vector highly relevant to the current recommendation context through the self-attention mechanism, providing a structurally complete and semantically rich feature basis for subsequent calculation of the matching matrix through matrix transpose product and weighted by the bidirectional attention mechanism.

[0044] In step S170, the course-learner matching feature matrix is obtained by calculating the product of the course-learner matching feature vector and its own transpose. By calculating the product of the course-learner matching feature vector and its own transpose to obtain the course-learner matching feature matrix, an association measure between features can be obtained. This association measure can be used to measure the correlation or similarity between features, thereby helping to understand the relationship between features and find potential association patterns. This is beneficial for further analyzing and interpreting the relationship between features and subsequent processing.

[0045] In step S180, the course-learner matching feature matrix is passed through a bidirectional attention mechanism module to obtain a classification feature matrix. Considering the bidirectional attention mechanism, the course-learner matching feature matrix can be weighted, so that key matching information gets higher weights. This helps the system to pay more attention to features that contribute significantly to the matching degree and ignore features that have less impact on the matching degree. By emphasizing key matching information, the accuracy and sensitivity of the classification feature matrix to the matching degree can be improved. Specifically, the bidirectional attention mechanism can take into account the bidirectional relationship between the course and the learner, that is, the degree of attention of the learner to the course and the degree of adaptation of the course to the learner. By performing bidirectional attention calculations on the matching feature matrix, this context relationship can be captured to better understand and represent the matching degree between the course and the learner.

[0046] Specifically, in the embodiment of the present application, passing the course-learner matching feature matrix through a bidirectional attention mechanism module to obtain a classification feature matrix includes: performing pooling on the course-learner matching feature matrix along the horizontal direction and the vertical direction respectively to obtain a first-direction pooling vector and a second-direction pooling vector; performing association encoding on the first-direction pooling vector and the second-direction pooling vector to obtain a bidirectional association matrix; inputting the bidirectional association matrix into a Sigmoid activation function to obtain a bidirectional association weight matrix; and calculating the element-wise product between the bidirectional association weight matrix and the course-learner matching feature matrix to obtain the classification feature matrix.

[0047] In step S190, the classification feature matrix is passed through a classifier to obtain a classification result, which is used to indicate whether to recommend the course to the learner. Considering that the classifier can make decisions on recommendation and filtering based on the characteristics of the learner and the course, as well as the degree of matching between them. Specifically, the classifier maps the classification feature matrix to a binary classification result, usually represented by "recommend" or "not recommend". Such a classification result can intuitively tell the system whether to recommend the course to the learner. For example, for a learner, if the classification result is "recommend", the system can present the course as the main recommendation result to the learner; if the classification result is "not recommend", the system can filter out the course and not display it in the recommendation results. This can better meet the personalized needs of learners and help improve the accuracy, personalization level, and user satisfaction of the recommendation system.

[0048] In summary, the learning path recommendation method based on big data analysis according to the embodiments of the present application is elucidated. It uses artificial intelligence technology based on a deep neural network model to obtain multi-dimensional data of the learner and the text introduction and course information of the learning courses to be recommended. The semantic encoder maps different text data into the semantic space to extract the key information of the learner, and then the bidirectional attention mechanism module is used for weighting to emphasize the key matching information, so as to obtain a classification result indicating whether to recommend the course to the learner. This can further provide recommended courses that meet the interests and needs of the learners and improve the satisfaction and learning effect of the learners.

[0049] Exemplary system Figure 4 is a block diagram of a learning path recommendation system based on big data analysis according to an embodiment of the present application. As Figure 4As shown, the learning path recommendation system 100 based on big data analysis according to an embodiment of the present application includes: a learner data acquisition module 110 for acquiring multi-dimensional learner data, where the multi-dimensional learner data includes learning history, hobbies, and learning goals; a learner data semantic extraction module 120 for performing data cleaning on the multi-dimensional learner data and then obtaining multiple learner data feature vectors through a first semantic encoder including a word embedding layer and concatenating them to obtain a learner data semantic feature vector; a learning course information acquisition module 130 for acquiring text introductions and course information of learning courses to be recommended; a learning course semantic extraction module 140 for performing word segmentation on the text introductions and course information of the learning courses to be recommended and then obtaining a semantic feature vector of the course to be recommended through a second semantic encoder including a word embedding layer; an interpolation and ordering fusion module 150 for performing weighted fusion on the learner data semantic feature vector and the semantic feature vector of the course to be recommended to obtain a course-learner association feature vector; a feature remapping and calibration module 160 for performing potential kernel space remapping and calibration based on modal decomposition on the course-learner association feature vector to obtain a course-learner matching feature vector; a matching module 170 for obtaining the course-learner matching feature matrix by calculating the product of the course-learner matching feature vector and its own transpose; a bidirectional attention module 180 for obtaining a classification feature matrix by passing the course-learner matching feature matrix through a bidirectional attention mechanism module; and a recommendation result judgment module 190 for obtaining a classification result by passing the classification feature matrix through a classifier, where the classification result is used to indicate whether to recommend the course to the learner.

[0050] Here, those skilled in the art can understand that the specific operations of each step in the above learning path recommendation system based on big data analysis have been introduced in detail in the description of the Figures 1 to 3 learning path recommendation method based on big data analysis above, and therefore, the repeated description thereof will be omitted.

[0051] Exemplary electronic device Next, reference will be made to Figure 5 to describe the electronic device according to an embodiment of the present application. Figure 5 is a block diagram of an electronic device according to an embodiment of the present application. As Figure 5 shown, the electronic device 10 includes one or more processors 11 and a memory 12.

[0052] The processor 11 may be a central processing unit (CPU) or other forms of processing units having data processing capabilities and / or instruction execution capabilities, and may control other components in the electronic device 10 to perform desired functions.

[0053] The memory 12 may include one or more computer program products, and the computer program products may include various forms of computer-readable storage media, such as volatile memory and / or non-volatile memory. The volatile memory may include, for example, random access memory (RAM) and / or cache memory, etc. The non-volatile memory may include, for example, read-only memory (ROM), hard disk, flash memory, etc. One or more computer program instructions may be stored on the computer-readable storage media, and the processor 11 may run the program instructions to implement the functions in the learning path recommendation method based on big data analysis in various embodiments of the present application described above and / or other desired functions. Various contents such as learner multi-dimensional data and text introductions and course information of the learning courses to be recommended may also be stored in the computer-readable storage media.

[0054] In one example, the electronic device 10 may further include: an input device 13 and an output device 14, and these components are interconnected through a bus system and / or other forms of connection mechanisms (not shown).

[0055] The input device 13 may include, for example, a keyboard, a mouse, and the like.

[0056] The output device 14 may output various information to the outside, including classification results, etc. The output device 14 may include, for example, a display, a speaker, a printer, and a communication network and its connected remote output devices, etc.

[0057] Of course, for simplicity, Figure 5 only some of the components related to the present application in the electronic device 10 are shown, and components such as a bus, an input / output interface, etc. are omitted. In addition, according to specific application scenarios, the electronic device 10 may further include any other appropriate components.

[0058] Exemplary computer program products and computer-readable storage media In addition to the above methods and devices, the embodiments of the present application may also be computer program products, which include computer program instructions that, when run by a processor, cause the processor to execute the steps in the functions in the learning path recommendation method based on big data analysis according to various embodiments of the present application described in the "Exemplary Method" section above.

[0059] The computer program product may be written in any combination of one or more programming languages for performing the operations of the embodiments of the present application. The programming languages include object-oriented programming languages such as Java, C++, etc., and also include conventional procedural programming languages such as the "C" language or similar programming languages. The program code may be executed entirely on the user's computing device, partially on the user's device, executed as a stand-alone software package, partially on the user's computing device and partially on a remote computing device, or entirely on a remote computing device or server.

[0060] In addition, an embodiment of the present application may also be a computer-readable storage medium having computer program instructions stored thereon, and when the computer program instructions are run by a processor, the processor is caused to execute the steps in the functions of the learning path recommendation method based on big data analysis according to various embodiments of the present application described in the "Exemplary Method" section above of this specification.

[0061] The computer-readable storage medium may employ any combination of one or more readable media. The readable media may be a readable signal medium or a readable storage medium. The readable storage medium may, for example, include but is not limited to an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination of the above. More specific examples (a non-exhaustive list) of the readable storage medium include: an electrical connection having one or more wires, a portable disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disc read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above.

Claims

1. A learning path recommendation method based on big data analysis, characterized in that: Including: Obtain multi-dimensional data of the learner, where the multi-dimensional data of the learner includes learning history, hobbies, and learning goals; After cleaning the multi-dimensional data of the learner, use a first semantic encoder including a word embedding layer to obtain multiple learner data feature vectors and concatenate them to obtain a learner data semantic feature vector; Obtain the text introduction and course information of the learning course to be recommended; After performing word segmentation on the text introduction and course information of the learning course to be recommended, use a second semantic encoder including a word embedding layer to obtain a semantic feature vector of the course to be recommended; Perform weighted fusion on the learner data semantic feature vector and the semantic feature vector of the course to be recommended to obtain a course-learner association feature vector; Perform potential kernel space remapping calibration based on modal decomposition on the course-learner association feature vector to obtain a course-learner matching feature vector; Calculate the product of the course-learner matching feature vector and its own transpose to obtain the course-learner matching feature matrix; Pass the course-learner matching feature matrix through a bidirectional attention mechanism module to obtain a classification feature matrix; Pass the classification feature matrix through a classifier to obtain a classification result, and the classification result is used to indicate whether to recommend the course to the learner.

2. The learning path recommendation method based on big data analysis according to claim 1, characterized in that After cleaning the multi-dimensional data of the learner, use a first semantic encoder including a word embedding layer to obtain multiple learner data feature vectors and concatenate them to obtain a learner data semantic feature vector, including: Pass the multi-dimensional data of the learner through the embedding layer of the first semantic encoder including a word embedding layer to convert each dimension data in the multi-dimensional data into an embedding vector to obtain a sequence of embedding vectors, where the embedding layer uses a learnable embedding matrix to perform embedding encoding on each dimension data; Input the sequence of embedding vectors into the first semantic encoder including a word embedding layer to obtain the multiple learner data feature vectors.

3. The learning path recommendation method based on big data analysis according to claim 2, wherein, Input the sequence of embedding vectors into the first semantic encoder including a word embedding layer to obtain the multiple learner data feature vectors, including: Arrange the sequence of embedding vectors into an input vector; Convert the input vector into a query vector and a key vector respectively through a learnable embedding matrix; Calculate the product between the query vector and the transposed vector of the key vector to obtain a self-attention association matrix; Perform normalization processing on the self-attention association matrix to obtain a normalized self-attention association matrix; Input the normalized self-attention association matrix into a Softmax activation function for activation to obtain a self-attention feature matrix; Multiply the self-attention feature matrix with each embedding vector in the sequence of embedding vectors as a value vector respectively to obtain the multiple learner data feature vectors.

4. The learning path recommendation method based on big data analysis according to claim 3, wherein, After performing word segmentation on the text introduction and course information of the learning course to be recommended, use a second semantic encoder including a word embedding layer to obtain a semantic feature vector of the course to be recommended, including: Perform word segmentation on the text introduction and course information of the to-be-recommended learning course to convert the text introduction and course information of the to-be-recommended learning course into a word sequence composed of multiple words; Use the embedding layer of the second semantic encoder including the embedding layer to map each word in the word sequence into a word embedding vector respectively to obtain a sequence of learning course word embedding vectors; Use the transformer of the second semantic encoder including the embedding layer to perform global context semantic encoding based on the transformer idea on the sequence of learning course word embedding vectors to obtain multiple context semantic feature vectors; Concatenate the multiple context semantic feature vectors to obtain the to-be-recommended course semantic feature vector.

5. The learning path recommendation method based on big data analysis according to claim 4, wherein Use the transformer of the second semantic encoder including the embedding layer to perform global context semantic encoding based on the transformer idea on the sequence of learning course word embedding vectors to obtain multiple context semantic feature vectors, including: Arrange the sequence of learning course word embedding vectors in one dimension to obtain a global feature vector; Calculate the product between the global feature vector and the transposed vectors of each vector in the sequence of learning course word embedding vectors to obtain multiple self-attention correlation matrices; Perform normalization processing on each self-attention correlation matrix in the multiple self-attention correlation matrices respectively to obtain multiple normalized self-attention correlation matrices; Pass each normalized self-attention correlation matrix in the multiple normalized self-attention correlation matrices through the Softmax classification function to obtain multiple probability values; Use each probability value in the multiple probability values as a weight to weight each vector in the sequence of learning course word embedding vectors respectively to obtain the multiple context semantic feature vectors.

6. The learning path recommendation method based on big data analysis according to claim 5, wherein Perform potential kernel space remapping calibration based on modal decomposition on the course learner association feature vector to obtain a course-learner matching feature vector, including: Construct a cross-pixel fine-grained association matrix of the course learner association feature vector; Perform deep kernel feature mining on the cross-pixel fine-grained association matrix based on the convolutional layer to obtain a course learner association feature potential kernel space dynamic response matrix; Perform orthogonal basis decomposition on the cross-pixel fine-grained association matrix to obtain a set of course learner association feature component representation vectors; Input each course learner association feature component representation vector in the set of course learner association feature component representation vectors into the feature importance dynamic focusing based on the self-attention mechanism to obtain a set of course learner association feature modal component representation vectors; Embed each course learner association feature modal component representation vector in the set of course learner association feature modal component representation vectors into the course learner association feature potential kernel space dynamic response matrix to obtain a set of course learner association feature kernel space masked feature vectors; Fuse the set of course learner association feature kernel space masked feature vectors to obtain the course-learner matching feature vector.

7. The learning path recommendation method based on big data analysis according to claim 6, wherein Pass the course-learner matching feature matrix through the bidirectional attention mechanism module to obtain a classification feature matrix, including: Pool the course-learner matching feature matrix along the horizontal and vertical directions respectively to obtain a first-direction pooling vector and a second-direction pooling vector; Perform correlation encoding on the first-direction pooling vector and the second-direction pooling vector to obtain a bidirectional correlation matrix; Input the bidirectional correlation matrix into the Sigmoid activation function to obtain a bidirectional correlation weight matrix; Calculate the element-wise multiplication between the bidirectional correlation weight matrix and the course-learner matching feature matrix to obtain the classification feature matrix.

8. A learning path recommendation system based on big data analysis, characterized in that, Including: A learner data acquisition module for acquiring multi-dimensional learner data, where the multi-dimensional learner data includes learning history, hobbies, and learning goals; A learner data semantic extraction module for performing data cleaning on the multi-dimensional learner data and then passing it through a first semantic encoder including a word embedding layer to obtain multiple learner data feature vectors and concatenating them to obtain a learner data semantic feature vector; A learning course information acquisition module for acquiring the text introduction and course information of the learning course to be recommended; A learning course semantic extraction module for performing word segmentation on the text introduction and course information of the learning course to be recommended and then passing it through a second semantic encoder including a word embedding layer to obtain a semantic feature vector of the course to be recommended; An interpolation and ordering fusion module for performing weighted fusion on the learner data semantic feature vector and the semantic feature vector of the course to be recommended to obtain a course-learner association feature vector; A feature remapping and calibration module for performing latent kernel space remapping and calibration based on modal decomposition on the course-learner association feature vector to obtain a course-learner matching feature vector; A matching module for obtaining the course-learner matching feature matrix by calculating the product of the course-learner matching feature vector and its own transpose; A bidirectional attention module for passing the course-learner matching feature matrix through a bidirectional attention mechanism module to obtain a classification feature matrix; A recommendation result judgment module for passing the classification feature matrix through a classifier to obtain a classification result, and the classification result is used to indicate whether to recommend the course to the learner.

9. The learning path recommendation system based on big data analysis according to claim 8, characterized in that The learner data semantic extraction module includes: An embedding unit for passing the multi-dimensional learner data through the embedding layer of the first semantic encoder including a word embedding layer to convert each dimension data in the multi-dimensional data into an embedding vector to obtain a sequence of embedding vectors, where the embedding layer uses a learnable embedding matrix to perform embedding encoding on each dimension data; A semantic encoding unit for inputting the sequence of embedding vectors into the first semantic encoder including a word embedding layer to obtain the multiple learner data feature vectors.

10. The learning path recommendation system based on big data analysis according to claim 9, characterized in that The semantic encoding unit is used for: Arranging the sequence of embedding vectors into an input vector; Converting the input vector into a query vector and a key vector respectively through a learnable embedding matrix; Calculating the product between the query vector and the transposed vector of the key vector to obtain a self-attention correlation matrix; Performing normalization processing on the self-attention correlation matrix to obtain a normalized self-attention correlation matrix; Input the standardized self-attention correlation matrix into the Softmax activation function for activation to obtain a self-attention feature matrix; Multiply the self-attention feature matrix with each embedding vector in the sequence of the embedding vectors as value vectors respectively to obtain the multiple learner data feature vectors.

Images