Big data-based physical exercise potential discovery method

Abstract

The invention relates to the technical field of data processing, in particular to a sports potential discovery method based on big data. The method comprises the steps of obtaining and preprocessing multi-dimensional data of a user to obtain preprocessed multi-dimensional data; performing feature extraction and optimization on the preprocessed multi-dimensional data by using a dynamically constrained nonlinear feature interaction optimization algorithm to obtain multi-dimensional comprehensive feature data; performing comprehensive potential evaluation on the multi-dimensional comprehensive feature data by using a multi-modal hierarchical dynamic entropy constraint potential analysis algorithm to obtain a comprehensive potential evaluation result; and providing a customized exercise item recommendation scheme based on a comprehensive potential evaluation result in combination with user interests. The problems of inaccurate user data analysis and incomplete exercise potential evaluation are solved.

Description

Technical Field

[0001] The present invention relates to the field of data processing technology, and in particular to a sports potential discovery method based on big data. Background Art

[0002] In the field of sports, accurately discovering and tapping sports potential is a key link in cultivating excellent athletes. However, most traditional sports potential assessment methods rely on single-dimensional indicators or expert experience judgment. These methods are often based on physical physiological characteristics (such as height, weight, and muscle ratio), ignoring the influence of psychological characteristics (such as stress resistance and adaptability), historical data (such as competition results and training records), and environmental factors (such as climate and geographical location). In addition, traditional sports potential assessment methods usually use static analysis models, which cannot dynamically capture the complex interactive relationships between features, resulting in insufficient comprehensiveness and accuracy of potential assessment results. Therefore, a sports potential discovery method based on big data came into being. Summary of the invention

[0003] The present invention provides a method for discovering sports potential based on big data to solve the problems of inaccurate analysis of user data and incomplete evaluation of sports potential.

[0004] A method for discovering sports potential based on big data, comprising the following steps: S1. Acquire and preprocess the multi-dimensional data of the user to obtain pre-processed multi-dimensional data; perform feature extraction and optimization on the pre-processed multi-dimensional data using a nonlinear feature interactive optimization algorithm with dynamic constraints to obtain multi-dimensional comprehensive feature data; S2. Use the multimodal hierarchical dynamic entropy constraint potential analysis algorithm to conduct comprehensive potential evaluation on multi-dimensional comprehensive feature data to obtain comprehensive potential evaluation results; based on the comprehensive potential evaluation results and combined with user interests, provide customized sports project recommendation plans.

[0005] Preferably, the S1 specifically includes: In the process of implementing the dynamic constraint nonlinear feature interaction optimization algorithm, the preliminary feature data is extracted from the preprocessed multidimensional data, and the preliminary feature data is standardized to obtain the standardized feature data; based on the standardized feature data, the dynamic constraint nonlinear interaction matrix is ​​constructed. , ,in Indicates Features and The nonlinear interaction strength of the features; the specific calculation formula is: , in, is the number of eigenvalues ​​of the feature; is the parameter that controls the strength of nonlinear attenuation; It is the standardized The first feature eigenvalues; It is the standardized The first feature eigenvalues; is the power parameter; is a constant used to avoid zero denominator.

[0006] Preferably, the S1 specifically includes: In the process of implementing the dynamic constraint nonlinear feature interaction optimization algorithm, based on the nonlinear interaction strength between features, the feature weights are optimized by introducing the redundant suppression function and the gradient descent method to obtain the optimized feature weights; the redundant suppression function is defined as follows: , in, is the redundancy suppression coefficient, which is used to control the strength of the redundant suppression term; is the nonlinear coefficient of redundancy suppression; Indicates The weight of each feature; Indicates The weight of each feature; is the noise suppression coefficient, which is used to control the strength of the noise suppression term; is a redundancy suppression term used to measure the similarity between feature weights; It is a noise suppression term, which is used to prevent noise from interfering with feature weight optimization; Based on the optimized feature weights, multi-dimensional comprehensive feature data is generated.

[0007] Preferably, the S2 specifically includes: The multimodal hierarchical dynamic entropy-constrained potential analysis algorithm constructs a high-order tensor to represent the multi-dimensional comprehensive feature data, then uses the characteristics of entropy constraints to analyze the potential distribution layer by layer, and obtains the comprehensive potential evaluation result through multi-step dynamic feedback.

[0008] Preferably, the S2 specifically includes: In the process of implementing the multimodal hierarchical dynamic entropy constraint potential analysis algorithm, the multi-dimensional comprehensive feature data is standardized to obtain the standardized comprehensive feature data. Then, the dynamic modal hierarchical method is used to divide the standardized comprehensive feature data into different modal groups based on the similarity measurement between features. The specific calculation formula is as follows: , in It is the standardized Comprehensive feature data and the standardized Comprehensive feature data Similarity score of Represents the standardized The comprehensive characteristics of eigenvalues; Represents the standardized The comprehensive characteristics of feature values.

[0009] Preferably, the S2 specifically includes: In the implementation process of the multi-modal hierarchical dynamic entropy constraint potential analysis algorithm, after completing the mode grouping, an extended feature matrix is ​​constructed for each mode; the expansion process is implemented by the following nonlinear mapping function: , Represents the standardized The extended features of the comprehensive features are obtained; and an extended feature matrix is ​​generated based on the standardized extended features of the comprehensive features.

[0010] Preferably, the S2 specifically includes: In the implementation process of the multimodal hierarchical dynamic entropy constrained potential parsing algorithm, the extended feature matrix of each modality is mapped to the tensor space to generate a high-order tensor representation.

[0011] Preferably, the S2 specifically includes: In the implementation process of the multimodal hierarchical dynamic entropy constraint potential analysis algorithm, the high-order tensor is used as the input of the entropy constraint optimization. By calculating the information entropy of each mode and introducing the modal interaction intensity matrix, the entropy constraint optimization objective function is constructed. The formula is as follows: , in, is the entropy constrained optimization objective function; represents the number of modes generated; It is The weight of each mode; is modal Information entropy of is the redundancy suppression coefficient; Indicates The mode and The interaction strength of the modalities.

[0012] Preferably, the S2 specifically includes: In the implementation process of the multimodal hierarchical dynamic entropy constrained potential analysis algorithm, based on the entropy constraint optimization objective function, the modal weights are iteratively optimized by the gradient descent method to obtain the optimized modal weights; based on the optimized modal weights and tensor representation as well as the interaction strength between modalities, the comprehensive potential evaluation results are calculated.

[0013] Preferably, the S2 specifically includes: The comprehensive potential evaluation results are standardized, and the user interests are feature extracted and quantified to obtain a normalized interest representation; a comprehensive recommendation score is calculated based on the standardized comprehensive potential evaluation results and the normalized interest representation, and a recommendation list is obtained based on the comprehensive recommendation score; based on the recommendation list, a recommended plan based on the user's interests and potential is obtained, including the name of the recommended sports project and the comprehensive recommendation score.

[0014] The beneficial effects of the technical solution of the present invention are: 1. A nonlinear feature interaction optimization algorithm with dynamic constraints is adopted. By constructing a nonlinear interaction matrix with dynamic constraints and a nonlinear iterative optimization framework, the complex interaction relationship between features is comprehensively analyzed. During the optimization process, the feature weights are dynamically adjusted to effectively improve the robustness and stability of feature expression. At the same time, by introducing a redundant suppression function, the redundancy and noise interference between feature weights are greatly reduced, making the multi-dimensional comprehensive feature data generated in the end more accurate and reliable. The above optimization algorithm shows strong adaptability and processing capabilities in high-dimensional complex data space.

[0015] 2. Through the multimodal hierarchical dynamic entropy constraint potential analysis algorithm, high-order tensor quantization of multimodal comprehensive feature data is realized, and the multimodal comprehensive feature data is mapped to the multimodal tensor space to capture the potential relationship between features. The complexity and potential richness of the tensor slices of each modality are measured by information entropy, and the synergy and redundancy between different modalities are measured by the modal interaction intensity matrix. This combination of high-order tensor quantization and entropy constraint optimization can not only accurately characterize the user's potential distribution, but also dynamically adjust the importance weight of the modality, thereby ensuring the scientificity and personalization of potential evaluation. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 This is a flow chart of a method for discovering sports potential based on big data described in the present invention. DETAILED DESCRIPTION

[0017] In order to further explain the technical means and effects adopted by the present invention to achieve the predetermined invention purpose, the technical scheme in the embodiment of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiment of the present invention. Obviously, the described embodiment is only a part of the embodiment of the present invention, not all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0018] Unless defined otherwise, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention belongs.

[0019] The following is a detailed description of a specific scheme of a sports potential discovery method based on big data provided by the present invention in conjunction with the accompanying drawings.

[0020] See attached Figure 1 , which shows a flow chart of a method for discovering sports potential based on big data provided by an embodiment of the present invention, the method comprising the following steps: S1. Acquire and preprocess the multi-dimensional data of the user to obtain pre-processed multi-dimensional data; perform feature extraction and optimization on the pre-processed multi-dimensional data using a nonlinear feature interactive optimization algorithm with dynamic constraints to obtain multi-dimensional comprehensive feature data; Acquire the user's multi-dimensional data, including physical physiological data (such as height, weight, body fat percentage, etc.), psychological test data (such as personality analysis, stress resistance), sports history data (such as training records, competition results) and environmental factors (such as climate, living area), etc.; perform preprocessing such as data cleaning, data standardization, and data noise reduction on the multi-dimensional data to obtain preprocessed multi-dimensional data, including preprocessed physical physiological data, preprocessed psychological test data, preprocessed sports history data and preprocessed environmental factors; in the preprocessing process, technical means well known in the art are used, which will not be elaborated here.

[0021] Furthermore, the preprocessed multi-dimensional data is subjected to feature extraction and optimization using a nonlinear feature interaction optimization algorithm with dynamic constraints. The nonlinear feature interaction optimization algorithm with dynamic constraints constructs a nonlinear interaction matrix with dynamic constraints, comprehensively analyzes the complex interaction relationship between features, and then uses a nonlinear iterative optimization framework to gradually adjust feature weights and feature values ​​to generate a comprehensive optimized feature representation, i.e., multi-dimensional comprehensive feature data. The specific implementation process is as follows: First, the existing feature engineering technology is used to extract preliminary feature data from the preprocessed multi-dimensional data, such as extracting preliminary physical physiological feature data from the preprocessed physical physiological data, such as height, weight, body fat rate, basal metabolic rate, heart rate, heart rate variability, blood pressure, blood sugar, etc.; extracting preliminary psychological feature data from the preprocessed psychological test data, such as personality characteristics, personality type characteristics, stress resistance characteristics, etc.; extracting preliminary sports feature data from the preprocessed sports history data, such as training records, training progress, historical performance correction, physical condition characteristics, etc.; extracting preliminary environmental feature data from the preprocessed environmental factors, such as climate condition characteristics, living environment characteristics, social environment characteristics, etc.; merging the extracted preliminary physical physiological feature data, preliminary psychological feature data, preliminary sports feature data and preliminary environmental feature data to obtain preliminary feature data. , denoted as ,in Indicates The first feature eigenvalues, is the total number of features, is the number of eigenvalues ​​of the feature. In order to eliminate the scale influence between different features, the existing standardization method is used to standardize the preliminary feature data to obtain the standardized feature data It ensures that the mean of each feature in the data set is zero and the standard deviation is 1. In the formula, It is the standardized The first feature The purpose is to eliminate unit differences and improve the stability of subsequent modeling.

[0022] Furthermore, the nonlinear interaction matrix of dynamic constraints is constructed , ,in Indicates Features and The nonlinear interaction strength of features. The specific calculation formula is: , in, It is a parameter that controls the strength of nonlinear attenuation. The larger the value, the smaller the impact of eigenvalue differences. It is the standardized The first feature eigenvalues; is the power parameter used to adjust the sensitivity of differences; is a constant used to avoid zero denominator; It is the nonlinear effect of amplifying the eigenvalue differences using exponential functions and power forms; is the interaction strength that amplifies positively or negatively correlated features; It describes the nonlinear relationship between the sum of squares of eigenvalues ​​and their absolute values, and is used to measure the balance and symmetry of eigenvalue distribution.

[0023] The nonlinear interaction matrix of dynamic constraints constructed above is used for subsequent feature weight optimization.

[0024] In the feature weight optimization process, dynamically adjust the importance weight of the feature. Define the initial feature weight vector ,in Indicates that all features are initially weighted equally. In each iteration, the weight update formula is: , in, Indicates The feature in The weight of the iteration; It is The feature in The weight of the iteration; It is The feature in The weight of the iteration; It is The feature in The weight of the iteration; , , Respectively represent , , Standardized feature data; It is a power parameter, which is used to control the sensitivity of feature differences and is determined according to the specific scenario; is the learning rate, which is used to control the step size of weight adjustment in each iteration; Indicates the norm. Each iteration gradually optimizes the importance distribution of features by increasing the coupling between feature weights and the strength of nonlinear interactions between features.

[0025] Furthermore, in order to avoid the redundancy problem in the weight optimization process, a redundancy suppression function is introduced , which is defined as follows: , in, is the redundancy suppression coefficient, which is used to control the strength of the redundant suppression term; is the nonlinear coefficient of redundancy suppression, which is used to control the amplification effect of feature weight differences; Indicates The weight of each feature; Indicates The weight of each feature; is the noise suppression coefficient, which is used to control the strength of the noise suppression term; It is a redundancy suppression term, which is used to measure the similarity between feature weights and aims to reduce the influence of features with weight values ​​that are too close, that is, the influence of redundant features; It is a noise suppression term, which is used to suppress weights that are too small to prevent noise from interfering with feature weight optimization.

[0026] The gradient of the redundancy suppression function is: , The gradient descent method is used to optimize the feature weights. The specific formula is: , in, The optimized The weight of the feature represents the weight of the first feature after gradient descent update. The weight of the feature is The status of the iteration; is the learning rate of gradient descent, which is used to control the step size when the weight is updated. The value range is .

[0027] After completing the optimization of feature weights, multi-dimensional comprehensive feature data is finally generated The calculation formula is: , in, Indicates Comprehensive feature data.

[0028] S2. Use the multimodal hierarchical dynamic entropy constraint potential analysis algorithm to conduct comprehensive potential evaluation on multi-dimensional comprehensive feature data to obtain comprehensive potential evaluation results; based on the comprehensive potential evaluation results and combined with user interests, provide customized sports project recommendation plans.

[0029] The multimodal hierarchical dynamic entropy constraint potential analysis algorithm constructs high-order tensors to represent multi-dimensional comprehensive feature data to capture the multimodal relationship between features, and then uses the characteristics of entropy constraints to analyze the potential distribution layer by layer, and obtains the final potential evaluation through multi-step dynamic feedback. The specific implementation process is as follows: First, the multi-dimensional comprehensive feature data is standardized to obtain the standardized comprehensive feature data , Represents the standardized The comprehensive feature data is divided into different modal groups using the dynamic modal stratification method. The purpose of modal stratification is to identify the dynamic relationship between features. Modal division is based on the similarity measurement between features. The specific calculation formula is as follows: , in It is the standardized Comprehensive feature data and the standardized Comprehensive feature data The similarity score of , the smaller the value, the higher the similarity between the two features; Represents the standardized The comprehensive characteristics of eigenvalues; Represents the standardized The comprehensive characteristics of eigenvalues. The existing spectral clustering algorithm is used to group the features and generate modalities, each modality represents a set of features with high similarity.

[0030] After completing the modal grouping, for each modal Constructing the extended feature matrix The expansion process is achieved through the following nonlinear mapping function: , Represents the standardized The purpose of the nonlinear mapping function is to increase the nonlinear relationship of the features and expand the dimension and expression ability of the features. It is expressed as: , in is modal The number of features in .

[0031] After the modal expansion is completed, the expanded feature matrix of each modality is further mapped to the tensor space. Specifically, the tensor slice is constructed: , in, For modal Tensor slices of ; is the modal weight matrix, which is used to adjust the importance of the expanded features; is the expanded feature dimension; is the time series length; Represents transpose. Stack the tensor slices of all modes by mode index to generate the final high-order tensor representation: , At this point, the high-order tensor The dimension is ; Any of these elements can be used express.

[0032] Higher-order tensors It is the input of the subsequent entropy constrained optimization. To evaluate the potential complexity of the tensor slice of each modality, its information entropy is first calculated: , , in, is modal The information entropy of The complexity of the feature distribution in the model. The higher the entropy, the more diverse the feature distribution of the model is and the richer the potential is. The lower the entropy, the more uniform the feature distribution of the model is. is a tensor slice No. Row, No. Elements of the column, indicating the mode Values ​​on specific feature dimensions and time series; is a tensor slice No. Row, No. Elements of a column; is a tensor slice No. Row, No. Normalized probability values ​​for the elements of the column.

[0033] Furthermore, in order to optimize the collaborative relationship between modes, the modal interaction strength matrix is ​​introduced: , Interaction strength between modalities Weighing the modal and modal relevance; is modal Tensor slices of No. Row, No. Elements of a column.

[0034] In order to avoid information redundancy caused by high-intensity interactions, it is necessary to perform control processing during the optimization process, that is, to combine the information entropy of the modalities and the interaction intensity between the modalities to construct the entropy constrained optimization objective function: , in, is the entropy constrained optimization objective function, which is used to balance the modal complexity and inter-modal redundancy; It is The weight of the modality, indicating the The importance of each mode in the optimization objective is determined based on expert experience; is the redundancy suppression coefficient, which is used to control the influence weight of the interaction intensity between modes on the optimization target. The larger the value, the more attention the optimization process pays to reducing modal redundancy.

[0035] Furthermore, the modal weights are iteratively optimized by the gradient descent method: , in, It is The mode in The weight after iterations; It is The mode in The weight after iterations; is the learning rate, which is used to control the step size of each update and is determined according to the specific scenario; is the entropy constrained optimization objective function Weight of modal The partial derivative of indicates the gradient direction of the current weight contribution to the objective function.

[0036] Finally, based on the optimized modal weights and tensor representations and the strength of interaction between modalities, computing potential evaluation scalars : , in, It is a modal slice norm, which represents the overall strength of the slice of the modal tensor; It is the weighted coefficient of the interaction strength between modalities, which is used to measure the impact of modal interaction on the potential score. The potential evaluation scalar, that is, the comprehensive potential evaluation result, directly quantifies the potential characteristics of the multi-dimensional comprehensive feature data.

[0037] Based on the potential assessment scalar and user interests, customized sports project recommendation solutions are provided, including: First, the potential evaluation scalar is standardized to obtain the standardized potential evaluation scalar. At the same time, the user's interest is extracted and quantified using existing feature engineering technology to obtain the normalized interest expression. The comprehensive recommendation score is further calculated, that is, the user's interest in sports items. The recommendation value is obtained by weighted calculation based on the standardized potential evaluation scalar and the normalized interest expression, and the comprehensive recommendation scores of all sports items are calculated and sorted in descending order to obtain a recommendation list. The elements in the recommendation list are compared with a threshold preset according to the expert experience method, and sports items with a value less than the preset threshold are filtered out. Finally, a plan recommended according to the user's interests and potential is obtained, including the recommended sports item name and the comprehensive recommendation score.

[0038] In summary, a method for discovering sports potential based on big data has been completed.

[0039] The order of the embodiments of the invention is for description only and does not represent the advantages and disadvantages of the embodiments. The processes depicted in the drawings do not necessarily require the specific order or continuous order shown to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0040] The various embodiments in this specification are described in a progressive manner, and the same or similar parts between the various embodiments can be referenced to each other, and each embodiment focuses on the differences from other embodiments.

[0041] The above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that the technical solutions described in the above embodiments may still be modified, or some of the technical features may be replaced by equivalents. Such modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention, and should be included in the protection scope of the present invention.

Claims

1. A method for discovering sports potential based on big data, characterized in that: The following steps are involved: S1. Acquire and preprocess the user's multi-dimensional data to obtain the preprocessed multi-dimensional data; The preprocessed multi-dimensional data is subjected to feature extraction and optimization using a nonlinear feature interactive optimization algorithm with dynamic constraints to obtain multi-dimensional comprehensive feature data; S2. Use the multimodal hierarchical dynamic entropy constraint potential analysis algorithm to conduct comprehensive potential evaluation on multi-dimensional comprehensive feature data to obtain comprehensive potential evaluation results; based on the comprehensive potential evaluation results and combined with user interests, provide customized sports project recommendation plans.

2. The method for discovering sports potential based on big data according to claim 1, characterized in that: The S1 specifically includes: In the process of implementing the dynamic constraint nonlinear feature interaction optimization algorithm, the preliminary feature data is extracted from the preprocessed multidimensional data, and the preliminary feature data is standardized to obtain the standardized feature data; based on the standardized feature data, the dynamic constraint nonlinear interaction matrix is ​​constructed. , ,in Indicates Features and The nonlinear interaction strength of the features; the specific calculation formula is: ， in, is the number of eigenvalues ​​of the feature; is the parameter that controls the strength of nonlinear attenuation; It is the standardized The feature of eigenvalues; It is the standardized The feature of eigenvalues; is the power parameter; is a constant used to avoid zero denominator.

3. The method for discovering sports potential based on big data according to claim 2, characterized in that: The S1 specifically includes: In the process of implementing the dynamic constraint nonlinear feature interaction optimization algorithm, based on the nonlinear interaction strength between features, the feature weights are optimized by introducing the redundant suppression function and the gradient descent method to obtain the optimized feature weights; the redundant suppression function is defined as follows: ， in, is the redundancy suppression coefficient, which is used to control the strength of the redundant suppression term; is the nonlinear coefficient of redundancy suppression; Indicates The weight of each feature; Indicates The weight of each feature; is the noise suppression coefficient, which is used to control the strength of the noise suppression term; is a redundancy suppression term used to measure the similarity between feature weights; It is a noise suppression term, which is used to prevent noise from interfering with feature weight optimization; Based on the optimized feature weights, multi-dimensional comprehensive feature data is generated.

4. The method for discovering sports potential based on big data according to claim 1, characterized in that: The S2 specifically includes: The multimodal hierarchical dynamic entropy-constrained potential analysis algorithm constructs a high-order tensor to represent the multi-dimensional comprehensive feature data, then uses the characteristics of entropy constraints to analyze the potential distribution layer by layer, and obtains the comprehensive potential evaluation result through multi-step dynamic feedback.

5. The method for discovering sports potential based on big data according to claim 4, characterized in that: The S2 specifically includes: In the process of implementing the multimodal hierarchical dynamic entropy constraint potential analysis algorithm, the multi-dimensional comprehensive feature data is standardized to obtain the standardized comprehensive feature data. Then, the dynamic modal hierarchical method is used to divide the standardized comprehensive feature data into different modal groups based on the similarity measurement between features. The specific calculation formula is as follows: ， in It is the standardized Comprehensive feature data and the standardized Comprehensive feature data Similarity score of Represents the standardized The comprehensive characteristics of eigenvalues; Represents the standardized The comprehensive characteristics of feature values.

6. The method for discovering sports potential based on big data according to claim 5, characterized in that: The S2 specifically includes: In the implementation process of the multi-modal hierarchical dynamic entropy constraint potential analysis algorithm, after completing the mode grouping, an extended feature matrix is ​​constructed for each mode; the expansion process is implemented by the following nonlinear mapping function: ， Represents the standardized The extended features of the comprehensive features are obtained; and an extended feature matrix is ​​generated based on the standardized extended features of the comprehensive features.

7. The method for discovering sports potential based on big data according to claim 6, characterized in that: The S2 specifically includes: In the implementation process of the multimodal hierarchical dynamic entropy constrained potential parsing algorithm, the extended feature matrix of each modality is mapped to the tensor space to generate a high-order tensor representation.

8. The method for discovering sports potential based on big data according to claim 7, characterized in that: The S2 specifically includes: In the implementation process of the multimodal hierarchical dynamic entropy constraint potential analysis algorithm, the high-order tensor is used as the input of the entropy constraint optimization. By calculating the information entropy of each mode and introducing the modal interaction intensity matrix, the entropy constraint optimization objective function is constructed. The formula is as follows: ， in, is the entropy constrained optimization objective function; represents the number of modes generated; It is The weight of each mode; is modal Information entropy of is the redundancy suppression coefficient; Indicates The mode and The interaction strength of the modalities.

9. The method for discovering sports potential based on big data according to claim 8, characterized in that: The S2 specifically includes: In the implementation process of the multimodal hierarchical dynamic entropy constrained potential analysis algorithm, based on the entropy constraint optimization objective function, the modal weights are iteratively optimized by the gradient descent method to obtain the optimized modal weights; based on the optimized modal weights and tensor representation as well as the interaction strength between modalities, the comprehensive potential evaluation results are calculated.

10. The method for discovering sports potential based on big data according to claim 9, characterized in that: The S2 specifically includes: The comprehensive potential evaluation results are standardized, and the user interests are feature extracted and quantified to obtain a normalized interest representation; a comprehensive recommendation score is calculated based on the standardized comprehensive potential evaluation results and the normalized interest representation, and a recommendation list is obtained based on the comprehensive recommendation score; based on the recommendation list, a recommended plan based on the user's interests and potential is obtained, including the name of the recommended sports project and the comprehensive recommendation score.

Images