A practical training course recommendation method and system based on big data analysis

Abstract

The application provides a practical training course recommendation method and system based on big data analysis, and relates to the technical field of data processing, which comprises the following steps: collecting historical interaction data of multiple users on different practical training courses; generating a sparse matrix; constructing a matrix singular value decomposition function about the sparse matrix with the Young inequality as a constraint to suppress random noise positively correlated with the size of the historical interaction data; combining the sparse matrix filling rate to update the suppression strength of the random noise in the matrix singular value decomposition function; solving the updated matrix singular value decomposition function to output a user hidden factor matrix, a practical training course hidden factor matrix and a combination matrix; calculating a practical training course similarity matrix describing the similarity between different categories of practical training courses according to the practical training course hidden factor matrix and the combination matrix; combining the practical training course similarity matrix to calculate the predicted interest scores of the target user on each practical training course, and recommending the practical training courses to the target user according to the predicted interest scores.

Description

Technical Field

[0001] This invention relates to the field of data processing technology, and in particular to a method and system for recommending practical training courses based on big data analysis. Background Technology

[0002] Big data refers to massive, diverse, and rapidly growing datasets. By collecting, storing, processing, and analyzing this data, we can uncover potential value and patterns, providing a basis for decision-making. Practical training courses are centered on hands-on practice and aim to improve students' or learners' practical skills and application abilities. They typically combine experiments, projects, simulations, or business cases, allowing learners to master professional knowledge and skills in real or simulated environments.

[0003] Faced with the complex challenge of matching a vast amount of course resources with students' skill needs, traditional manual screening is inefficient and susceptible to subjective limitations. Personalized recommendation systems, however, can accurately identify students' knowledge base, career goals, and learning preferences, dynamically matching highly relevant practical training content. This avoids wasting ineffective learning resources, increases students' interest in recommended practical courses, and ultimately helps students quickly build the practical skills required for their positions, shortening the skills transfer cycle.

[0004] However, the recommendation process for practical training courses requires identifying similar courses with comparable content and high overlap in user interests. But in big data scenarios, most users have only rated a very small number of courses, resulting in extremely sparse data. This often leads to random statistical artifacts—coincidental noise—caused by the limited number of valid behavioral samples. Most recommendation scenarios tend to ignore this coincidental noise. Recommendations with coincidental noise contain statistical artifacts, inevitably leading to significant biases and impacting the accuracy and reliability of personalized recommendations. Summary of the Invention

[0005] To address the problem that existing technologies for recommending practical training courses require identifying similar courses with comparable content and high overlap in user interests, but in big data scenarios, most users have only rated a very small number of courses, resulting in extremely sparse data. This leads to random statistical artifacts, or coincidental noise, caused by the limited number of valid behavioral samples. Most recommendation scenarios tend to ignore this coincidental noise, and recommendations with coincidental noise contain statistical artifacts, inevitably leading to significant biases and affecting the accuracy and reliability of personalized recommendations. This invention provides a method and system for recommending practical training courses based on big data analysis.

[0006] The technical solutions provided by the embodiments of the present invention are as follows: First aspect This invention provides a method for recommending practical training courses based on big data analysis, comprising: S1: Collect historical interaction data of multiple users on different practical training courses; S2: Generate a sparse matrix based on historical interaction data; S3: Construct a matrix singular value decomposition function for sparse matrices constrained by Young's inequality to suppress random noise that is positively correlated with the scale of historical interaction data; S4: Combine the sparse matrix filling rate to update the suppression strength of random noise in the singular value decomposition function of the matrix; S5: Solve the updated singular value decomposition function of the matrix and output the user latent factor matrix, the training course latent factor matrix and the combination matrix; S6: Based on the latent factor matrix and combination matrix of the training courses, calculate the training course similarity matrix, which describes the degree of similarity between different categories of training courses; S7: Combine the similarity matrix of practical training courses to calculate the target user's predicted interest score for each practical training course, and recommend practical training courses to the target user based on the predicted interest score.

[0007] Second aspect This invention provides a practical training course recommendation system based on big data analysis, comprising: processor; A memory storing computer-readable instructions, which, when executed by the processor, implement the training course recommendation method based on big data analysis as described in the first aspect.

[0008] Third aspect The present invention provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the training course recommendation method based on big data analysis as described in the first aspect.

[0009] The beneficial effects of the technical solutions provided in the embodiments of the present invention include at least the following: In this embodiment of the invention, by introducing Young's inequality constraints into the singular value decomposition of the sparse matrix describing user-training course interaction data, the extreme deviations in the matrix or random coincidental noise caused by sample sparsity can be effectively limited during the decomposition process, thereby preventing such noise from excessively affecting the recommendation process. Furthermore, by dynamically adjusting the noise suppression strength in conjunction with the sparse matrix filling rate, it is ensured that course similarity can still be accurately captured even with limited or sparse data. By calculating the similarity between courses and making recommendations based on predicted interest scores, personalized training course recommendations that better match the actual preferences of the target users can be provided, significantly improving the accuracy and reliability of the recommendations while reducing the bias caused by data sparsity. Attached Figure Description

[0010] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0011] Figure 1 A flowchart illustrating a training course recommendation method based on big data analysis provided in an embodiment of the present invention; Figure 2 This is a schematic diagram of the structure of a training course recommendation system based on big data analysis, provided in an embodiment of the present invention. Detailed Implementation

[0012] The technical solution of the present invention will now be described with reference to the accompanying drawings.

[0013] In embodiments of the present invention, words such as "exemplarily," "for example," etc., are used to indicate that something is an example, illustration, or description. Any embodiment or design described as "exemplary" in the present invention should not be construed as being more preferred or advantageous than other embodiments or designs. Specifically, the use of the word "exemplary" is intended to present the concept in a concrete manner. Furthermore, in embodiments of the present invention, the meaning expressed by "and / or" can be both, or either one.

[0014] To make the technical problems, technical solutions and advantages of the present invention clearer, a detailed description will be given below in conjunction with the accompanying drawings and specific embodiments.

[0015] Reference manual attached Figure 1 The diagram illustrates a flowchart of a training course recommendation method based on big data analysis provided by an embodiment of the present invention.

[0016] This invention provides a method for recommending practical training courses based on big data analysis. This method can be implemented using a device for recommending practical training courses based on big data analysis, which can be a terminal or a server. The processing flow of the method for recommending practical training courses based on big data analysis may include the following steps: S1: Collect historical interaction data of multiple users on different practical training courses.

[0017] Historical interaction data refers to all past interactions between users and the training course, reflecting users' interest, preferences, and participation in the course.

[0018] It is understandable that users and training courses are described in the form of embedding vectors throughout the process. Specifically, user features and training course features are encoded into embedding vectors using tools such as TF-IDF, Word2Vec, or FastText.

[0019] In one possible implementation, historical interaction data specifically refers to user ratings of practical training courses.

[0020] The rating score quantifies a user's historical behavior toward a particular training course into a numerical value, representing the degree of the user's interest or participation.

[0021] Alternatively, the score can also be determined according to the following rules. Specifically, the score is first obtained from the practical training course record platform. i The user on the first j The historical behaviors of each training course are defined as follows: no interaction, clicking to browse the training course, enrolling in the training course, completing a predetermined percentage of the training course content, submitting the training course assignment, and completing the training course. Then, according to preset rules, these historical behaviors are converted into representations of the first training course. i The user on the first j The rating score is based on the level of interest in each practical training course. Optionally, the preset rules can be: no interaction results in a rating of 0; clicking to browse the practical training course results in a rating of 1; enrolling in the practical training course results in a rating of 2; completing a preset percentage of the practical training course content results in a rating of 3; submitting practical training course assignments results in a rating of 4; and completing the practical training course results in a rating of 5.

[0022] It should be noted that those skilled in the art can set the preset ratio according to actual needs, and this invention does not limit this. Optionally, it can be set to 0.5. Quantifying user behavior into rating values ​​can accurately reflect the degree of interest, improving the personalization and accuracy of course recommendations.

[0023] S2: Generate a sparse matrix based on historical interaction data.

[0024] A sparse matrix is ​​a matrix in which most elements are empty or zero, with only a small number of elements having actual values. In recommender systems, user-course matrices are typically sparse because most users have only interacted with a small number of courses, leaving most positions in the matrix without any recorded activity. Generating sparse matrices can systematically represent the interaction relationships between all users and courses, while effectively focusing on users' true interests by retaining actual interaction information and ignoring missing parts, thereby improving the targeting and reliability of recommendations.

[0025] In one possible implementation, the element values ​​of the sparse matrix are specifically the user ratings for the training courses, wherein the number of rows in the sparse matrix is ​​the same as the number of users, and the number of columns in the sparse matrix is ​​the same as the number of training course categories.

[0026] The formula for a sparse matrix is ​​as follows: ; in, Represents a sparse matrix. Indicates the first i The user on the first j The scoring value of each practical training course , This indicates the total number of users. , This indicates the total number of practical training courses.

[0027] It's important to note that by transforming users' historical interaction data with training courses into a sparse matrix, the relationships between all users and courses are systematically represented. Non-empty elements in the matrix reflect users' actual interests and participation behaviors, while missing elements represent interactions that did not occur. This effectively ignores irrelevant information and focuses on genuine preferences. In this way, the recommendation system can extract users' latent interests and course features in a sparse data environment, providing a reliable foundation for subsequent matrix factorization and course similarity calculations, thereby improving the accuracy and personalization of recommendations. S3: Construct a matrix singular value decomposition function for sparse matrices constrained by Young's inequality to suppress random noise that is positively correlated with the size of historical interaction data.

[0028] Young's inequality, a fundamental inequality in mathematical analysis and a special case of the weighted arithmetic-geometric mean inequality, describes how, under specific conditions, the arithmetic mean of non-negative real numbers is not less than the geometric mean, and the equality condition is related to the power function ratio between variables. In this scheme, it is a mathematical inequality used to constrain the magnitude of singular values ​​in a matrix, thereby limiting the influence of extreme values ​​on the results during matrix decomposition. The singular value decomposition function decomposes the original matrix into a product of multiple matrices, containing singular values ​​and latent factors, used to extract the latent structure and features of the matrix. The scale of historical interaction data refers to the total amount or coverage of user interactions with courses throughout history; the larger the scale, the greater the data noise and bias. Random noise refers to irregular information caused by sparse data or accidental behavior; this noise may mislead recommendation algorithms into generating incorrect associations.

[0029] In one possible implementation, the singular value decomposition function of the matrix includes a reconstruction error term, a regularization term, and a constraint term. S3 specifically includes: S301: To minimize the reconstruction error of the sparse matrix, establish a reconstruction term for the sparse matrix.

[0030] S302: To avoid overfitting of the user latent factor matrix and the training course latent factor matrix, a regularization term is established.

[0031] S303: To suppress random noise that is positively correlated with the scale of historical interaction data, a constraint term based on Young's inequality is established.

[0032] S304: Combining the reconstruction term, regularization term, and constraint term yields the singular value decomposition function of the matrix.

[0033] Specifically, the system constructs a singular value decomposition function to decompose the sparse user-course interaction matrix into user latent factor matrices and course latent factor matrices to extract latent interests and course features. In this process, a reconstruction term is first established to fit the original interaction data, a regularization term is added to prevent overfitting, and then a noise constraint term is constructed using Young's inequality to suppress random noise amplified by the increasing scale of historical interaction data. These three parts are combined to form a complete objective function, enabling the decomposition to find the latent factor embeddings that best reflect real user preferences while satisfying the noise constraint. By applying constraints to the combined matrix, excessive penalties for effective interaction information are avoided, and the suppression strength is adaptively adjusted according to the data scale, ensuring that noise generated by accidental behavior does not mislead feature learning. This allows the system to accurately capture the true semantics of courses and user interests in a sparse interaction environment, effectively improving the accuracy, stability, and personalization of recommendation results.

[0034] The formula for the singular value decomposition function of a matrix is ​​as follows: ; in, Let these represent the user latent factor matrix and the training course latent factor matrix, respectively. Denotes the square of the F-norm. Indicates the weight of the regularization term. This indicates that the constraints are met. This indicates the preset number of singular values ​​to be retained. Indicates taking the matrix The k Large singularity, This indicates the suppression strength for random noise. Represents the logarithmic function. Representing a sparse matrix The nuclear norm number.

[0035] It should be noted that this formula, by introducing a constraint term into classical matrix singular value decomposition, limits the sum of the principal singular values ​​of the combined matrix UᵀV to a threshold derived by Young's inequality ( This approach adaptively controls random noise related to the scale of historical interaction data. Compared to traditional decomposition, this method effectively suppresses noise interference with latent factor learning while fitting the original interaction data, preventing accidental behavior from being amplified into erroneous associations. This allows for the extraction of more authentic user interests and course features in sparse data environments, improving the accuracy and stability of recommendations.

[0036] The specific process of deriving the threshold based on Young's inequality is as follows: According to Young's inequality principle, for a set of non-negative numbers, the arithmetic mean is greater than or equal to the geometric mean. In this scheme, the conjugate pair parameter in Young's inequality is always 2 for derivation, i.e. The parameters that need to be constrained are as follows: According to the property of singular value decomposition, any combination matrix can be decomposed into... ,in, and All are orthogonal matrices. Diagonal matrices. The diagonal is the singular value we want to scale. .therefore, ,in, Represents the first element in the corresponding matrix. i Line number j Column elements. The above formula is essentially the sum of the products of all corresponding elements, for each pair of elements... We can obtain the following by applying the basic Young's inequality and summing the results: Alignment is performed using an orthogonal transformation (orthogonal transformations do not change the F norm of the matrix), i.e. And thus obtain The core objective of matrix factorization is to make... In other words, the product of latent factors obtained from the decomposition should approximate the original sparse matrix R as closely as possible. According to the optimality property of matrix decomposition, when the decomposition converges, the sum of squares of the F-norms of the latent factors and the nuclear norm of R satisfy a direct proportional relationship: Therefore, it is equivalently replaced. Then, it is bound to the scale of interactive data and a constraint strength coefficient is introduced (m users, n training courses, a total of m+n free variables, random coincidence noise is independent error, according to the probability concentration property of independent random variables: the greater the total degrees of freedom, the higher the probability of extreme coincidence noise, and the upper limit of the maximum fluctuation of noise). Proportional, therefore it is compared with Multiply (adapting to noise fluctuations in datasets of different sizes) to obtain .

[0037] The specific meaning of the above function is as follows: in the constraint terms Finding the minimum value of the function under constraints and , Indicates a refactoring item. This indicates a regular term.

[0038] Optionally, those skilled in the art can set the weight of the regularization term according to actual needs, and this invention does not limit this. Optionally, it can be set to 0.5. The preset number of singular values ​​represents the total number of latent features to be retained in advance. Those skilled in the art can set the preset number of singular values ​​according to actual needs, and this invention does not limit this. Optionally, it can be set to 4 or 5, etc. Sparse matrix The nuclear norm is also known as the sparse matrix. The sum of all singular values.

[0039] It should be noted that the noise-resistant matrix singular value decomposition function designed based on Young's inequality is, in general, based on classical matrix decomposition, with the addition of a noise constraint term adapted to the data scale. This controls the interference of noise on latent feature learning at its source while fitting the user's real interaction preferences. Specifically, the process first constructs a reconstruction term to fit the original interaction data and a regularization term to suppress model overfitting. Then, based on Young's inequality, a noise constraint term adapted to the overall data scale is derived. Finally, these terms are combined to obtain the complete objective function to be optimized, i.e., the matrix singular value decomposition function. The goal is to find the user and course latent factor embeddings that best fit the original data while satisfying the noise constraint. This scheme applies constraints to a combined matrix that condenses the latent feature association information, avoiding excessive penalty to the original effective interaction information and allowing the constraint strength to adaptively adjust with the data scale. It avoids the problems of excessively strong constraints losing real features or insufficient constraints failing to suppress noise, enabling the learning of latent features that better fit the true semantics of the course in sparse interaction scenarios, ultimately improving the accuracy and stability of the recommendation results.

[0040] S4: Combine the sparse matrix filling rate to update the suppression strength of random noise in the singular value decomposition function of the matrix.

[0041] The sparse matrix filling rate refers to the proportion of valid interaction data in the sparse matrix, i.e., the proportion of non-empty elements in the matrix, reflecting the completeness of user-course interaction data. Suppression strength refers to the degree to which random noise is restricted or weakened during the singular value decomposition of the matrix; the higher the suppression strength, the less interference noise causes in the extraction of latent features. By dynamically adjusting the noise suppression strength in conjunction with the sparse matrix filling rate, the impact of noise can be flexibly controlled according to the data sparsity, enabling matrix decomposition to accurately extract user interests and course features even with limited data, thereby improving the accuracy and stability of recommendation results.

[0042] In one possible implementation, S4 specifically includes: S401: Get the total number of non-zero elements in a sparse matrix.

[0043] S402: Take the quotient of the total number of non-zero elements and the total number of elements in the sparse matrix to obtain the sparse matrix fill rate.

[0044] S403: Combine the sparse matrix filling rate and update the suppression strength based on the principle that it is positively correlated with the sparse matrix filling rate.

[0045] The specific formula for updating the suppression strength is as follows: ; in, Indicates the intensity of inhibition. Represents the natural constant. Indicates the fill rate of the sparse matrix. This represents the total number of non-zero elements.

[0046] Specifically, the process first counts the number of non-zero elements in the sparse matrix and calculates the ratio of non-zero elements to the total number of elements in the matrix to obtain the sparse matrix filling rate. Then, the noise suppression intensity is dynamically adjusted based on the filling rate. The suppression intensity is positively correlated with the filling rate; that is, the higher the filling rate, the greater the allowed suppression intensity, thus adaptively controlling the impact of noise on matrix decomposition. Through this method, singular value decomposition of the matrix can flexibly suppress random noise according to the sparsity of the data, making the latent factor matrix more accurately reflect the user's potential interests and course characteristics, thereby improving the accuracy and stability of the recommendation results, while avoiding excessive penalty or over-amplification of random noise caused by sparse data.

[0047] S5: Solve the updated singular value decomposition function of the matrix and output the user latent factor matrix, the training course latent factor matrix, and the combination matrix.

[0048] The user latent factor matrix represents the latent interest features of each user, encoding user preference information across different dimensions into vector form. The training course latent factor matrix represents the latent features of each course, encoding course content characteristics and factors attracting users into vector form. The combined matrix, generated from the user and course latent factor matrices, is used to reconstruct the original interaction matrix or to assist in calculating users' predicted interest in courses.

[0049] It's important to note that the larger the sum of the user latent factor matrix and the training course latent factor matrix, the more easily even small, accidental noises in the original data (such as a few users coincidentally selecting two unrelated courses) can be amplified into significant errors affecting the final recommendation results. However, by solving for the user latent factor matrix and the training course latent factor matrix using a matrix singular value decomposition function that combines Young's inequality and suppression strength, we can effectively suppress random noise that is positively correlated with the scale of historical interaction data. This avoids erroneous similarity associations caused by accidental noise (for example, unrelated training courses mistakenly associated by a few users will not be judged as similar). This improves the accuracy of the recommendation results. It's understandable that although the final course similarity calculation only uses the training course latent factor matrix, the user latent factor matrix is ​​a necessary prerequisite for obtaining a correct training course latent factor matrix: if the user latent factor matrix is ​​only used to solve for the training course latent factor matrix, the initial random error of the user latent factor matrix will be directly transferred to the training course latent factor matrix, causing distortion of the course features extracted from the training course latent factor matrix, and ultimately reducing the accuracy of the recommendation.

[0050] In one possible implementation, S5 specifically includes: S501: Latent factor matrix of fixed practical training courses.

[0051] S502: Calculate the gradient of the user's latent factor matrix at the current iteration number.

[0052] The specific formula for calculating the gradient is as follows: ; in, This represents the gradient of the user's latent factor matrix at the current iteration number. This indicates the weight of the regularization term.

[0053] S503: Update the user's latent factor matrix along the gradient descent direction.

[0054] The updated formula is as follows: ; in, This represents the updated user latent factor matrix. This indicates the preset iteration step size.

[0055] It should be noted that those skilled in the art can set the preset iteration step size according to actual needs, and this invention does not limit this. Optionally, the preset iteration step size can be set to 0.002.

[0056] S504: Project the updated user latent factor matrix onto the constraint terms based on Young's inequality, and truncate the combined matrix composed of the updated user latent factor matrix and the training course latent factor matrix so that the singular values ​​and values ​​of the combined matrix satisfy the constraint terms.

[0057] Specifically, the truncation process is as follows: First, take the updated user latent factor matrix and the currently fixed training course latent factors to obtain a combined matrix. Then, perform singular value decomposition on this combined matrix to obtain... d Given a set of singular values, calculate their sum. Then, sum the sum against the upper bound of the constraint terms. By comparison, if the sum is less than or equal to the upper limit, the current solution is valid and no adjustment is needed. If the sum exceeds the upper limit: singular value truncation is performed, i.e., the smallest batch of singular values ​​is pruned (small singular values ​​themselves correspond to random noise), or all singular values ​​are scaled proportionally until the sum of singular values ​​exactly meets the upper limit requirement. Afterwards, a new constrained combination matrix needs to be reconstructed based on the adjusted new singular values. Conversely, the user latent factor matrix under the adjusted new combination matrix is ​​obtained. The original U with excessive noise will be modified to U that satisfies the constraints with lower noise, providing qualified fixed parameters for the next symmetric update V.

[0058] S505: The user latent factor matrix obtained after fixed truncation is used to symmetrically update the latent factor matrix of the training course.

[0059] The process involves symmetrically updating the latent factor matrix of the training course, which is the user latent factor matrix obtained after fixed truncation. Then, the gradient with respect to the latent factor matrix of the training course is calculated, followed by gradient update, and then projection, using the same update rules.

[0060] S506: Determine if the current iteration count is greater than the preset iteration count. If yes, output the updated user latent factor matrix, training course latent factor matrix, and combination matrix. Otherwise, increment the iteration count by 1 and return to step S501.

[0061] Understandably, if the current iteration number does not meet the condition of being greater than the preset iteration number, then the user latent factor matrix and the training course latent factor matrix updated under the current iteration number are used to update the user latent factor matrix and the training course latent factor matrix in step S501. The combined matrix in step S506 is obtained by combining the user latent factor matrix and the training course latent factor matrix updated under the current iteration number.

[0062] Specifically, the system solves for the latent factor matrices of users and training courses by iteratively optimizing the singular value decomposition function of the matrix. First, the course latent factor matrix is ​​fixed, the gradient of the user latent factor matrix is ​​calculated and updated along the gradient descent direction. Then, the updated user latent factor matrix is ​​projected onto constraints based on Young's inequality, and the combined matrix is ​​truncated or scaled to ensure that the sum of singular values ​​meets noise suppression requirements, thereby reducing the noise impact of accidental interactions. Subsequently, the user latent factor matrix is ​​fixed, and the course latent factor matrix is ​​symmetrically updated and similarly subjected to projection constraints to ensure accurate extraction of course features. Through multiple iterations, this method can gradually optimize the latent features of users and courses, making the latent factor matrix both realistically reflect user interests and avoid amplification of random noise, thus improving the stability and accuracy of the combined matrix. This provides a reliable foundation for subsequent course similarity calculations and personalized recommendations, significantly improving the accuracy of recommendation results.

[0063] S6: Based on the latent factor matrix and combination matrix of the training courses, calculate the training course similarity matrix, which describes the degree of similarity between different categories of training courses.

[0064] The practical training course similarity matrix quantifies the degree of similarity between different practical training courses. Each element in the matrix represents the similarity between two courses in the latent feature space, reflecting the proximity of course content and its appeal to user interests. Calculating the course similarity matrix based on the latent factor matrix and combination matrix of practical training courses fully utilizes the combined information of user latent interests and course latent features, mapping originally sparse or scattered interaction data into the latent space, thereby accurately identifying the inherent similarities between courses. This method avoids misjudgments caused by accidental behavior, enabling the recommendation system to provide more personalized course recommendations that better match users' true preferences and learning needs, improving the accuracy and reliability of recommendations.

[0065] In one possible implementation, S6 specifically includes: S601: Obtain the singular values ​​of the combined matrix.

[0066] S602: Sort the singular values ​​in descending order and construct a diagonal matrix.

[0067] S603: Calculate the similarity matrix of the training courses using the latent factor matrix and the diagonal matrix of the training courses.

[0068] The elements of the practical training course similarity matrix represent the similarity between any two practical training courses.

[0069] The formula for the similarity matrix of the practical training courses is as follows: ; in, This represents the similarity matrix of practical training courses. This represents the latent factor matrix of the practical training course. Represents a diagonal matrix The inverse matrix, subscript T Indicates transpose. Represents a diagonal matrix. Represents the first element in the combination matrix. i Large singular values, , This represents the total number of singular values.

[0070] Specifically, this process calculates the course similarity matrix using the latent factor matrix and combination matrix of the training courses. First, the singular values ​​of the combination matrix are obtained and sorted by size to construct a diagonal matrix. Then, matrix operations are performed using the course latent factor matrix and the inverse of this diagonal matrix to obtain the similarity in the latent feature space between courses. This method maps the originally sparse and scattered interaction data to the latent space, comprehensively considering user interests and course characteristics to accurately reveal the inherent connections between courses. In this way, even if a few accidental behaviors exist, they will not mislead the course similarity judgment, thereby improving the accuracy and personalization of recommendations and ensuring that the system can provide course recommendations that better match the user's true preferences.

[0071] S7: Combine the similarity matrix of practical training courses to calculate the target user's predicted interest score for each practical training course, and recommend practical training courses to the target user based on the predicted interest score.

[0072] The predicted interest score is a numerical value calculated based on the target user's potential interest characteristics and the similarity between courses. It measures the user's potential level of interest or preference for each practical training course. By calculating the predicted interest score, the system can combine the user's potential interests with course similarity, accurately identify the courses the user is most likely to be interested in, thereby achieving personalized recommendations and effectively improving user engagement and learning fit.

[0073] In one possible implementation, S7 specifically includes: S701: A training course set for acquiring historical interaction data with target users.

[0074] S702: Combining the similarity matrix of training courses and the training course set, calculate the predicted interest score of the target user for each training course for which there is no historical interaction data.

[0075] The specific formula for calculating the predicted interest score is as follows: ; in, Indicates target user i For the j Predicted interest scores for each practical training course The first term in the similarity matrix of practical training courses represents the... j Individual training courses and training course collection The Middle k The similarity between the practical training courses Indicates target user i For the k Historical interaction data of each practical training course.

[0076] S703: Recommend a preset number of training courses to target users in descending order of predicted interest scores.

[0077] It is understood that those skilled in the art can set the preset quantity according to actual needs, and this invention does not limit it. Optionally, it can be set to 10, that is, recommending 10 categories of practical training courses.

[0078] It should be noted that the predicted interest score is calculated using a similarity matrix of training courses and the target user's historical interaction data. First, the set of courses the target user has interacted with is obtained. Then, combining the similarity between courses, the user's interest information in the interacted courses is mapped to the uninterviewed courses, resulting in a predicted interest score for each course. This method, by integrating the user's latent interests and course similarity, can accurately measure the user's preference for each course and recommend the most likely courses to the user based on the predicted interest score. In this way, the recommendation system can achieve personalized course matching, effectively improving user engagement and learning experience, while reducing recommendation bias caused by data sparsity or random behavior, thus improving the accuracy and reliability of the recommendation results. The method of calculating course similarity based on the latent factor matrix and combination matrix of training courses, compared to traditional methods that only rely on surface interactions or co-occurrence relationships, can comprehensively consider the correlation between the user's latent interests and course features in the latent space, mapping sparse and scattered interaction data to the latent feature space. This way, even if a few random interactions exist, they will not mislead the course similarity judgment, thus more accurately revealing the inherent connections between courses, achieving precise matching for personalized recommendations, and improving the accuracy and reliability of recommendations.

[0079] In practical applications, personalized recommendations for training courses are achieved through big data analysis. First, historical interaction data between users and courses is collected to reflect user interests and participation levels. Then, a sparse matrix is ​​constructed to systematically represent the relationship between users and courses. Based on this, singular value decomposition of the matrix is ​​used to extract latent features of users and courses. Young's inequality constraints are introduced, combined with the sparse matrix filling rate, to dynamically suppress random noise and prevent accidental interactions from being amplified into erroneous associations. By iteratively optimizing the latent factor matrix and truncating or scaling the combined matrix, latent interests and course features are accurately extracted. Then, combined with the course similarity matrix, the predicted interest score of users for courses they have not interacted with is calculated, and recommendations are made based on the predicted interest ranking. This method can comprehensively consider latent user interests and course features in sparse data environments, mapping data to a latent space, effectively reducing noise interference, thereby improving the accuracy, personalization level, and reliability of recommendations, making the recommendations more consistent with users' true preferences and learning needs.

[0080] The beneficial effects of the technical solutions provided in the embodiments of the present invention include at least the following: In this embodiment of the invention, by introducing Young's inequality constraints into the singular value decomposition of the sparse matrix describing user-training course interaction data, the extreme deviations in the matrix or random coincidental noise caused by sample sparsity can be effectively limited during the decomposition process, thereby preventing such noise from excessively affecting the recommendation process. Furthermore, by dynamically adjusting the noise suppression strength in conjunction with the sparse matrix filling rate, it is ensured that course similarity can still be accurately captured even with limited or sparse data. By calculating the similarity between courses and making recommendations based on predicted interest scores, personalized training course recommendations that better match the actual preferences of the target users can be provided, significantly improving the accuracy and reliability of the recommendations while reducing the bias caused by data sparsity.

[0081] Reference manual attached Figure 2 The diagram shows a structural schematic of a training course recommendation system based on big data analysis provided by the present invention.

[0082] This invention also provides a training course recommendation system 20 based on big data analysis, applied to the above-mentioned training course recommendation method based on big data analysis, including: Processor 201.

[0083] The memory 202 stores computer-readable instructions, which, when executed by the processor 201, implement the training course recommendation method based on big data analysis as described in the method embodiment.

[0084] The training course recommendation system 20 based on big data analysis provided by this invention can execute the above-mentioned training course recommendation method based on big data analysis and achieve the same or similar technical effects. To avoid duplication, this invention will not elaborate further.

[0085] It should be understood that the processor in the embodiments of the present invention can be a central processing unit (CPU), or it can be other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor can be a microprocessor or any conventional processor.

[0086] It should also be understood that the memory in the embodiments of the present invention can be volatile memory or non-volatile memory, or may include both volatile and non-volatile memory. The non-volatile memory can be read-only memory (ROM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), or flash memory. The volatile memory can be random access memory (RAM), which is used as an external cache. By way of example, but not limitation, many forms of random access memory (RAM) are available, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate synchronous DRAM (DDR SDRAM), enhanced synchronous DRAM (ESDRAM), synchronous linked DRAM (SLDRAM), and direct rambus RAM (DR RAM).

[0087] The above embodiments can be implemented, in whole or in part, by software, hardware (such as circuits), firmware, or any other combination thereof. When implemented using software, the above embodiments can be implemented, in whole or in part, as a computer program product. The computer program product includes one or more computer instructions or computer programs. When the computer instructions or computer programs are loaded or executed on a computer, all or part of the processes or functions described in the embodiments of the present invention are generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium can be any available medium that a computer can access or a data storage device such as a server or data center that includes one or more sets of available media. The available medium can be a magnetic medium (e.g., floppy disk, hard disk, magnetic tape), an optical medium (e.g., DVD), or a semiconductor medium. A semiconductor medium can be a solid-state drive.

[0088] It should be understood that the term "and / or" in this article is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A existing alone, A and B existing simultaneously, or B existing alone. A and B can be singular or plural. Additionally, the character " / " in this article generally indicates an "or" relationship between the preceding and following related objects, but it can also represent an "and / or" relationship. Please refer to the context for a more accurate understanding.

[0089] In this invention, "at least one" means one or more, and "more than one" means two or more. "At least one of the following" or similar expressions refer to any combination of these items, including any combination of a single item or a plurality of items. For example, at least one of a, b, or c can represent: a, b, c, ab, ac, bc, or abc, where a, b, and c can be a single item or multiple items.

[0090] It should be understood that, in various embodiments of the present invention, the order of the above-mentioned process numbers does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present invention.

[0091] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementations should not be considered beyond the scope of this invention.

[0092] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the devices, apparatuses, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0093] In the several embodiments provided by this invention, it should be understood that the disclosed devices, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another device, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between devices or units may be electrical, mechanical, or other forms.

[0094] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0095] In addition, the functional units in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit.

[0096] If the aforementioned functions are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this invention, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0097] This invention provides a computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements the training course recommendation method based on big data analysis as described in the method embodiment.

[0098] The present invention provides a computer-readable storage medium that can implement the steps and effects of the training course recommendation method based on big data analysis in the above-described method embodiments. To avoid repetition, the present invention will not repeat the details.

[0099] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

[0100] The following points need to be explained: (1) The accompanying drawings of the embodiments of the present invention only involve the structures involved in the embodiments of the present invention. Other structures can refer to the general design.

[0101] (2) For clarity, the thickness of layers or regions is enlarged or reduced in the drawings used to describe embodiments of the invention, i.e., these drawings are not drawn to scale. It is understood that when an element such as a layer, film, region or substrate is referred to as being “above” or “below” another element, the element may be “directly” located “above” or “below” the other element or there may be intermediate elements.

[0102] (3) Where there is no conflict, the embodiments of the present invention and the features in the embodiments can be combined with each other to obtain new embodiments.

[0103] The above are merely specific embodiments of the present invention, but the scope of protection of the present invention is not limited thereto. The scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A method for recommending practical training courses based on big data analysis, characterized in that, include: S1: Collect historical interaction data of multiple users on different practical training courses; S2: Generate a sparse matrix based on the historical interaction data; S3: Construct a matrix singular value decomposition function constrained by Young's inequality for the sparse matrix to suppress random noise that is positively correlated with the scale of historical interaction data; S4: Combine the sparse matrix filling rate to update the suppression strength of the random noise in the singular value decomposition function of the matrix; S5: Solve the updated singular value decomposition function of the matrix and output the user latent factor matrix, the training course latent factor matrix and the combination matrix; S6: Based on the latent factor matrix of the training courses and the combination matrix, calculate the training course similarity matrix describing the degree of similarity between different categories of training courses; S7: Based on the similarity matrix of the training courses, calculate the predicted interest score of the target user for each of the training courses, and recommend the training courses to the target user according to the predicted interest score.

2. The method for recommending practical training courses based on big data analysis according to claim 1, characterized in that, The historical interaction data specifically refers to the user's rating of the training course.

3. The method for recommending practical training courses based on big data analysis according to claim 2, characterized in that, The element values ​​of the sparse matrix are specifically the ratings of the users for the training courses. The number of rows in the sparse matrix is ​​the same as the number of users, and the number of columns in the sparse matrix is ​​the same as the number of training course categories.

4. The method for recommending practical training courses based on big data analysis according to claim 1, characterized in that, The singular value decomposition function of the matrix includes a reconstruction error term, a regularization term, and a constraint term; S3 specifically includes: S301: With the goal of minimizing the sparse matrix reconstruction error, establish a reconstruction term for the sparse matrix; S302: To avoid overfitting between the user latent factor matrix and the training course latent factor matrix, the regularization term is established; S303: To suppress random noise that is positively correlated with the scale of the historical interaction data, establish constraint terms based on Young's inequality; S304: Combine the reconstruction term, the regularization term, and the constraint term to obtain the singular value decomposition function of the matrix.

5. The method for recommending practical training courses based on big data analysis according to claim 1, characterized in that, S4 specifically includes: S401: Obtain the total number of non-zero elements in the sparse matrix; S402: Take the quotient of the total number of non-zero elements and the total number of elements in the sparse matrix to obtain the sparse matrix fill rate; S403: Based on the sparse matrix filling rate, update the suppression intensity according to the principle that it is positively correlated with the sparse matrix filling rate.

6. The method for recommending practical training courses based on big data analysis according to claim 1, characterized in that, S5 specifically includes: S501: Fix the implicit factor matrix of the training course; S502: Calculate the gradient of the user's latent factor matrix at the current iteration number; S503: Update the user latent factor matrix along the gradient descent direction; S504: Project the updated user latent factor matrix onto the constraint terms based on Young's inequality, and truncate the combined matrix composed of the updated user latent factor matrix and the training course latent factor matrix so that the singular values ​​and values ​​of the combined matrix satisfy the constraint terms. S505: Fix the user latent factor matrix obtained after truncation, and symmetrically update the training course latent factor matrix; S506: Determine whether the current iteration count is greater than the preset iteration count. If yes, output the updated user latent factor matrix, training course latent factor matrix, and the combined matrix. Otherwise, increment the iteration count by 1 and return to step S501.

7. The method for recommending practical training courses based on big data analysis according to claim 1, characterized in that, S6 specifically includes: S601: Obtain the singular values ​​of the combined matrix; S602: Sort the singular values ​​in descending order to construct a diagonal matrix; S603: The similarity matrix of the training courses is calculated using the latent factor matrix and the diagonal matrix of the training courses.

8. The method for recommending practical training courses based on big data analysis according to claim 1, characterized in that, Specifically, S7 includes: S701: Obtain a set of training courses that have historical interaction data with the target user; S702: Combining the training course similarity matrix and the training course set, calculate the target user's predicted interest score for each training course for which there is no historical interaction data; S703: Recommend a preset number of the training courses to the target user in descending order of the predicted interest scores.

9. A practical training course recommendation system based on big data analysis, characterized in that, include: processor; A memory storing computer-readable instructions, which, when executed by the processor, implement the training course recommendation method based on big data analysis as described in any one of claims 1 to 8.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the program is executed by the processor, it implements the training course recommendation method based on big data analysis as described in any one of claims 1 to 8.

Images