A new energy learning content recommendation method and system based on an adaptive strategy

Abstract

The application provides a new energy learning content recommendation method and system based on an adaptive strategy, relates to the technical field of data processing, and comprises the following steps: obtaining learning resources about new energy; performing dimension reduction on the learning resources in combination with a self-encoder; performing self-clustering on the dimension-reduced learning resources in combination with a soft allocation algorithm and a k-means clustering algorithm to obtain multiple types of sub-learning resources; generating topic labels of the multiple types of sub-learning resources; obtaining multiple learning behaviors of a target user, wherein the learning behaviors include learning content, learning content completion degree, learning time and learning duration of each learning; calculating interest degrees of the target user for different topic labels based on the multiple learning behaviors; calculating recommendation proportions of the multiple types of sub-learning resources in a positive correlation manner with a Nash equilibrium factor according to the interest degrees; and recommending each sub-learning resource to the target user according to the corresponding recommendation proportion. The method ensures recommendation diversity and accuracy and improves user learning efficiency.

Description

Technical Field

[0001] This invention relates to the field of data processing technology, and in particular to a method and system for recommending new energy learning content based on an adaptive strategy. Background Technology

[0002] New energy learning content refers to knowledge resources surrounding new energy technologies, applications, and related industries, including technical theories, engineering applications, policies and regulations, market analysis, and learning materials for solar energy, wind energy, energy storage technologies, electric vehicles, and hydrogen energy, providing learners with systematic professional learning support.

[0003] Because new energy technologies are complex and rapidly evolving, learners face a vast amount of information and struggle to sift through high-quality resources. Therefore, recommending new energy learning content is crucial. Intelligent recommendations can accurately match user interests and needs, improving learning efficiency, helping users quickly grasp core knowledge, and guiding them to focus on key areas, thereby promoting the popularization of new energy technologies and industrial development.

[0004] However, existing new energy learning content often fails to select learning materials that meet user needs from a vast amount of available information. Furthermore, the recommended content is too limited, leading to insufficient accuracy and negatively impacting the user's learning outcomes. Summary of the Invention

[0005] To address the technical problems of existing technologies that often fail to select suitable learning content from a vast amount of new energy learning materials, and that the recommended content is too limited, leading to insufficient accuracy and negatively impacting user learning outcomes, this invention provides a new energy learning content recommendation method and system based on an adaptive strategy.

[0006] The technical solutions provided by the embodiments of the present invention are as follows:

[0007] First aspect

[0008] This invention provides a method for recommending new energy learning content based on an adaptive strategy, comprising:

[0009] S1: Obtain learning resources about new energy;

[0010] S2: Dimensionality reduction of learning resources by combining autoencoders;

[0011] S3: Combine the soft allocation algorithm and the k-means clustering algorithm to perform self-clustering on the dimensionality-reduced learning resources to obtain multiple sub-learning resources;

[0012] S4: Generate topic tags for various sub-learning resources;

[0013] S5: Obtain multiple learning behaviors of the target user, including the learning content, completion rate, learning time, and learning duration of each learning session;

[0014] S6: Calculate the target user's interest in different topic tags based on each learning behavior;

[0015] S7: Calculate the recommendation ratio of each type of sub-learning resource based on the degree of interest in a positively correlated manner with the Nash equilibrium factor;

[0016] S8: Recommend each sub-learning resource to the target user according to the corresponding recommendation ratio.

[0017] Second aspect

[0018] This invention provides a new energy learning content recommendation system based on an adaptive strategy, comprising:

[0019] processor;

[0020] A memory storing computer-readable instructions, which, when executed by the processor, implement the new energy learning content recommendation method based on an adaptive strategy as described in the first aspect.

[0021] Third aspect

[0022] The present invention provides a computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements the new energy learning content recommendation method based on an adaptive strategy as described in the first aspect.

[0023] The beneficial effects of the technical solutions provided in the embodiments of the present invention include at least the following:

[0024] In this embodiment of the invention, an autoencoder is used to reduce the dimensionality of learning resources, thereby reducing redundancy and preserving core semantic information. Then, a soft allocation algorithm and k-means clustering algorithm are combined to perform self-clustering on the dimensionality-reduced learning resources. This improves the discriminative power of different clusters, making resource classification more accurate. Simultaneously, dynamically allocating probabilities strengthens the influence of high-confidence samples, avoiding the limitations of hard clustering. This allows for more accurate segmentation of learning content topics, helping the recommendation system capture users' genuine interest in various new energy technologies, improving the relevance and diversity of recommended content, and meeting personalized learning needs. Next, topic tags for each sub-learning resource obtained from clustering are generated. Based on the target user's multiple learning behaviors, the target user's interest in different topic tags (i.e., sub-learning resources) is calculated. Then, a positive correlation method with a Nash equilibrium factor is used to calculate the recommendation ratio of each type of sub-learning resource. Finally, based on the recommendation ratio, different categories of sub-learning resource content are extracted and recommended to the target user. This ensures that highly interesting content is prioritized while preserving diversity, ultimately accurately matching user learning needs. This method can more effectively capture users' genuine interest in new energy technologies, improve the relevance and exploratory nature of recommendations, and dynamically adapt to changes in user interests, significantly improving user learning efficiency and experience. Attached Figure Description

[0025] 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.

[0026] Figure 1 A flowchart illustrating a new energy learning content recommendation method based on an adaptive strategy, provided in an embodiment of the present invention;

[0027] Figure 2 This is a schematic diagram of a new energy learning content recommendation system based on an adaptive strategy, provided in an embodiment of the present invention. Detailed Implementation

[0028] The technical solution of the present invention will now be described with reference to the accompanying drawings.

[0029] 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.

[0030] 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.

[0031] Reference manual attached Figure 1 The diagram illustrates a flowchart of a new energy learning content recommendation method based on an adaptive strategy, provided by an embodiment of the present invention.

[0032] This invention provides a method for recommending new energy learning content based on an adaptive strategy. This method can be implemented by a new energy learning content recommendation device based on an adaptive strategy, which can be a terminal or a server. The processing flow of the new energy learning content recommendation method based on an adaptive strategy may include the following steps:

[0033] S1: Obtain learning resources about new energy.

[0034] It should be noted that by acquiring various forms of new energy learning resources (such as research papers, video courses, industry reports, etc.), a diverse and comprehensive data foundation is provided for the recommendation system, covering new energy technologies, industries, and policy fields, ensuring the richness and applicability of the recommended content.

[0035] In one possible implementation, learning resources include e-books, research papers, academic journals, industry reports, case studies, and lab tutorials.

[0036] Understandably, the learning resources encompass ebooks, research papers, academic journals, industry reports, case studies, and experimental tutorials, providing a comprehensive and diverse range of knowledge formats. This approach ensures coverage of different learning needs, satisfying users' diverse learning requirements in theory, application, practice, and case studies, thereby enhancing the applicability and practical value of the recommended content.

[0037] S2: Combine autoencoders to reduce the dimensionality of learning resources.

[0038] Autoencoders are artificial neural network models used for unsupervised learning. They compress input data into a low-dimensional latent space (encoding) and reconstruct it back into a high-dimensional space (decoding), enabling the model to learn the core features and key patterns of the data and reduce data redundancy. Using autoencoders to reduce the dimensionality of new energy learning resources can reduce redundant information, extract core semantic features, retain key content, and reduce data complexity. This provides more accurate input for subsequent clustering and recommendation processes, thereby improving the relevance of recommended content and user learning efficiency.

[0039] In one possible implementation, S2 specifically includes:

[0040] S201: Convert the text content in the learning resources into a low-dimensional continuous vector using a weighted average.

[0041]

[0042] Among them, v text V represents a low-dimensional continuous vector representing each piece of text content, W represents the set of words in the text content, and v w Let represent the pre-trained word vector of word w, p(w) represent the recurrence frequency of word w in the corresponding text content, and a represent the smoothing factor used to control the weight of high-frequency words.

[0043] The pre-trained word vectors can be from Word2Vec or GloVe. A smoothing factor is a parameter used to dynamically adjust the weight of high-frequency words in the text representation. Although high-frequency words appear frequently in text, they usually contain less semantic information, so their influence needs to be appropriately reduced. Such a smoothing factor can effectively prevent high-frequency words from dominating the text representation while fully utilizing the semantic value of low-frequency words. For short texts, 0.001 or 0.01 is commonly used to ensure that low-frequency words have a greater influence. For long texts, 0.1 or 1 is commonly used to appropriately reduce the weight of high-frequency words.

[0044] It's worth noting that the weighted averaging method dynamically adjusts word importance, balancing the contributions of high-frequency and low-frequency words to text representation. The resulting low-dimensional continuous vectors are both semantically accurate and highly efficient and compact. This method can significantly improve the model's semantic understanding capabilities in various natural language processing tasks.

[0045] Understandably, representing text content as dense, low-dimensional continuous vectors effectively solves the sparsity problem that may exist in text content, while preserving core semantic information, making short texts easier to perform subsequent clustering or classification tasks and improving the final clustering accuracy.

[0046] S202: Remove the first principal component of all low-dimensional continuous vectors from the low-dimensional continuous vectors of each text content to obtain optimized low-dimensional continuous vectors:

[0047] v final =v text -(v text ·u)·u

[0048] Among them, v final This represents the optimized low-dimensional continuous vector of each text content obtained after removing the first principal component u from all low-dimensional continuous vectors.

[0049] Specifically, the first principal component *u* among all low-dimensional continuous vectors is obtained by performing Principal Component Analysis (PCA) on all low-dimensional continuous vectors together. After removing the principal components from the text embeddings, the resulting vectors are more focused on the unique semantic features of the text, rather than being influenced by frequently occurring global features. This improves the semantic discriminative power of the text embedding representation, making it more suitable for subsequent clustering or classification tasks.

[0050] It should be noted that by using weighted averaging and principal component removal, the core semantic information of the text is preserved while reducing feature redundancy.

[0051] S203: The encoder of the autoencoder maps the optimized low-dimensional continuous vectors to the latent representation, thus completing the dimensionality reduction of the learning resources.

[0052] Specifically, the encoder of an autoencoder uses a multi-layer neural network structure to compress the optimized low-dimensional continuous vector of the input into a more compact latent space, capturing the core semantic features of the data. Specifically, each layer of the encoder performs a linear transformation (weight matrix and bias term) on the input features and combines it with a non-linear activation function to progressively extract abstract features from higher layers, thereby generating the final latent representations. These latent representations condense the key information of the learning resources.

[0053] It should be noted that by employing weighted averaging, autoencoders, and PCA dimensionality reduction techniques, this scheme reduces redundancy in new energy learning resources, extracts their core semantic features, and makes the representation more compact and efficient. Simultaneously, principal component removal avoids interference from frequent features, ensuring the uniqueness of the latent representation. This process not only improves the semantic accuracy of the learning resources but also provides higher-quality input data for subsequent clustering and recommendation, significantly enhancing the relevance of recommended content and user learning efficiency.

[0054] In one possible implementation, the autoencoder is trained as follows:

[0055] The autoencoder is trained with the objective of minimizing the reconstruction error of all low-dimensional continuous vectors after removing the first principal component. The specific reconstruction error is as follows:

[0056]

[0057] Among them, v i Let represent the optimized low-dimensional continuous vector of the i-th text content, i = 1, 2, ..., N, where N represents the total number of text contents. The decoder representing the autoencoder pairs v i The reconstructed optimized low-dimensional continuous vector, L recon This indicates the reconstruction error.

[0058] It's important to note that during training, the autoencoder's network weights are continuously optimized to minimize reconstruction error. This ensures that the latent representation mapped by the encoder retains the semantic features of the original embeddings, i.e., optimizing the semantic features of the low-dimensional continuous vectors, thereby improving text classification accuracy. By minimizing the autoencoder's reconstruction error, the semantic features of the optimized low-dimensional continuous vectors are fully preserved, optimizing the latent representation. This allows the model to more accurately capture the core features of new energy learning content, thus improving the accuracy of recommended content and classification efficiency, meeting users' needs for diverse learning content.

[0059] S3: Combine the soft allocation algorithm and the k-means clustering algorithm to perform self-clustering on the dimensionality-reduced learning resources to obtain multiple sub-learning resources.

[0060] Among them, soft assignment algorithm is a clustering method that, when classifying data, does not assign each sample to a fixed cluster, but instead calculates the probability of each sample belonging to different clusters, generating flexible classification results that better reflect the uncertainty of the data. K-means clustering is a commonly used hard clustering algorithm that, through iterative optimization, divides the data into clusters, minimizing the Euclidean distance between each data point in a cluster and its cluster center, thus achieving efficient data classification. Sub-learning resources refer to a set of categorized resources divided from the original learning resources through clustering methods. Each sub-resource represents content from a specific topic or domain, such as photovoltaic technology or wind energy storage. By using soft assignment algorithm and K-means clustering to perform self-clustering on the dimensionality-reduced learning resources, the uncertainty of resources can be flexibly handled and the classification accuracy optimized, improving the distinguishability between different sub-learning resources. This approach makes the classification of new energy learning content more accurate, effectively capturing users' true interests in different topics, ensuring the diversity and relevance of recommended content, and meeting users' personalized learning needs.

[0061] In one possible implementation, S3 specifically includes:

[0062] S301: The k-means clustering algorithm is used to cluster each potential representation once to obtain multiple category clusters.

[0063] The objective function for a single clustering operation is as follows:

[0064]

[0065] Among them, z i c represents the latent representation of the i-th text content. i Indicate z i The label of the cluster to which it belongs. Indicates cluster label c i The corresponding cluster center, L k-means This represents the objective function value, i.e., the first-order clustering error. Indicates the calculation of z i and The square of the Euclidean distance between them.

[0066] Understandably, the objective function of a single clustering operation is to minimize the sum of the squared Euclidean distances between the latent representations of all text content and their corresponding cluster centers, thereby optimizing the distribution of cluster centers. This makes the text content within each cluster more compact, improving the accuracy and representativeness of the clustering results, and laying a good foundation for subsequent classification and recommendation.

[0067] S302: Calculate the probability that each latent representation belongs to a different class cluster using a soft assignment algorithm:

[0068]

[0069] Where, q ij Let μ represent the probability that the latent representation of the i-th text content belongs to the j-th category cluster, j′ represent the cluster index of all category clusters, and μ represent the probability that the latent representation of the i-th text content belongs to the j-th category cluster. j and μ j′ Let represent the center points of the j-th and j′-th category clusters, respectively.

[0070] It should be noted that by calculating the distance between each latent representation and the centers of all clusters, a probability distribution belonging to each cluster is generated. A higher probability indicates a stronger similarity between the latent representation and the corresponding cluster. This method avoids the rigid limitations of hard clustering, better reflects the uncertainty of the data, and makes the clustering results more flexible and accurate, thereby improving the detail of new energy learning resource classification and the flexibility of the recommendation system.

[0071] S303: Generate the expected target probability with respect to probability:

[0072]

[0073] Where, p ij Let represent the expected target probability that the latent representation of the i-th text content belongs to the j-th category cluster.

[0074] in, The current allocation q for cluster j ij Strengthen (square) and normalize to the cluster level. Iterate through all clusters, normalize the numerators, and ensure that the total probability of all clusters is 1. The expected target probability p... ij More inclined to strengthen those soft-assignment probabilities q ijLarger clusters increase the influence of high-confidence samples on the model. The expected target probability transforms the soft assignment of data points into a more rigorous distribution, making the optimization process more stable and the clustering results more concentrated, focusing on high-confidence cluster assignments. It is a self-supervised signal that drives continuous optimization of the model (such as encoder parameters and cluster centers) by minimizing the KL divergence with the soft assignment. The target probability can avoid the interference of large clusters on the overall optimization, improving cluster purity.

[0075] More specifically, the expected target probability can strengthen high-confidence first-order clustering results and weaken low-confidence first-order clustering results. For example, if a first-order clustering q ij If the result is 0.9, then the expected target probability will be increased to 0.95. If the clustering result is 0.1, then the expected target probability will be reduced to 0.01.

[0076] S304: Update the center position of the class cluster with the objective of minimizing the KL divergence between the probability and the expected target probability.

[0077]

[0078] Among them, L KL p ij With q ij The KL divergence between the groups, where K represents the total number of clusters and log represents the logarithmic function.

[0079] KL divergence measures the difference between the soft assignment probability and the expected target probability. The goal is to minimize this difference, thereby optimizing the center positions of the class clusters. By iteratively adjusting the cluster centers, the actual distribution gradually converges with the target distribution, strengthening the influence of high-confidence samples and improving the accuracy and stability of the clustering results. This optimization process ensures the rationality and robustness of the classification of new energy learning content, providing a foundation for accurate recommendations.

[0080] It should be noted that by using the results of the initial clustering and combining them with a soft assignment method to generate a probability distribution, and then optimizing the model based on this distribution, the final clustering results become more accurate and robust. This is a dynamic, self-supervised optimization strategy that improves clustering quality while avoiding the risks of premature convergence or model collapse.

[0081] In practical applications, encoder parameters and the center point position of the category cluster can be updated simultaneously:

[0082] The encoder update formula is as follows:

[0083]

[0084] Here, θ represents the encoder parameters, which include a weight matrix that determines how each layer extracts features from the output of the previous layer, and bias terms used to adjust the output value after the input passes through the weight matrix. ∝ represents the partial derivative, and ∝ represents a direct proportional relationship.

[0085] The specific formula for updating the center position of the category cluster, i.e., the cluster center, is as follows:

[0086]

[0087] Specifically, this joint update scheme for encoder parameters and cluster center positions dynamically adjusts feature extraction and cluster centers by minimizing KL divergence, enabling the encoder to extract more accurate latent representations while optimizing the cluster structure. For new energy learning content recommendation, this scheme can more clearly distinguish different topics in the clustering results, improve classification accuracy and the model's adaptability to dynamic interests, thereby providing more relevant and efficient recommended content to meet users' personalized learning needs.

[0088] S305: Using the updated center position of each category cluster as the cluster center point, perform secondary clustering on each latent representation using the k-means clustering algorithm to obtain multi-class sub-learning resources.

[0089] It is understandable that after clustering the latent representations, the learning resources can be divided into multiple sub-learning resources based on the clustering results.

[0090] In practical applications, combining primary clustering and soft allocation algorithms enhances the influence of high-confidence samples and dynamically optimizes cluster center positions, avoiding interference from large clusters and premature convergence, thereby improving the purity and robustness of clustering. Secondary clustering further optimizes the accuracy of resource classification, ensuring that new energy learning content can be accurately classified and allocated according to user interests, effectively improving the relevance of recommendations and user learning efficiency, while also enhancing the diversity and adaptability of recommendations to meet personalized user needs.

[0091] S4: Generate topic tags for various sub-learning resources.

[0092] It should be noted that by generating topic tags for various sub-learning resources, the core content of each type of learning resource can be clearly identified, improving the semantic expression ability of the resources. This enables the recommendation system to more accurately match user interests, while also making it easier for users to quickly understand and select new energy topics that interest them, thereby improving learning efficiency and the personalized experience of recommendations.

[0093] In one possible implementation, S4 specifically refers to:

[0094] Topic tags for various sub-learning resources are generated through topic models, including LDA topic models and NMF topic models.

[0095] It should be noted that generating topic tags for various sub-learning resources using LDA and NMF topic models allows for the extraction of core semantic features for each resource category. This enables the recommendation system to more accurately match user interests while providing users with intuitive topic information for quick understanding and selection. This approach effectively improves the accuracy of new energy learning content recommendations, the clarity of classification, and user learning efficiency, meeting the diverse learning needs of complex domains.

[0096] S5: Obtain multiple learning behaviors of the target user.

[0097] The learning behavior includes the learning content, completion rate, learning time, and learning duration for each learning session.

[0098] Learning behavior refers to the specific data showing how users interact with learning resources during the learning process. By acquiring multiple learning behaviors of users, we can comprehensively capture the dynamics of users' interests in different learning resources. Combined with information such as learning content, completion rate, and duration, we can more accurately characterize users' learning needs, providing data support for subsequent interest calculation and personalized recommendations, thereby improving the accuracy of recommended content and user satisfaction.

[0099] It should be noted that those skilled in the art can set the number of times the learning behavior is collected according to actual needs, and this invention does not limit this. The learning behavior obtained in step S5 is multiple learning behaviors obtained based on the current time, i.e., recent learning behaviors.

[0100] S6: Calculate the target user's interest in different topic tags based on each learning behavior.

[0101] It should be noted that by analyzing users' multiple learning behaviors and dynamically calculating users' interest in different topic tags, it is possible to accurately capture users' real needs and changes in interest in new energy learning content, providing targeted data support for the recommendation system, effectively improving the relevance and personalization of recommended content, and ensuring that recommendations can be dynamically adjusted according to users' interests, thereby enhancing the learning experience and efficiency.

[0102] In one possible implementation, S6 specifically includes:

[0103] S601: Extract learning content topic tags for each learning session using topic modeling.

[0104] S602: Calculate the cosine similarity between the topic tags of each learning content and the topic tags of each type of sub-learning resource.

[0105] The specific method for calculating cosine similarity is as follows:

[0106]

[0107] Where S represents the topic tag embedding vector V of the learning content for each learning session. a The embedding vector V of any topic tag in the topic tags of the learning resources b The cosine similarity is calculated, where the embedding vector can be generated by a pre-trained word vector model Word2Vec or GloVe.

[0108] Understandably, this approach effectively captures the semantic relationships between learning content and learning resources, ensuring that recommended resources are highly matched with users' learning content needs, thereby improving the accuracy and relevance of recommendations.

[0109] S603: Retain the topic tags of the sub-learning resources that have the highest cosine similarity to the topic tags of each learning content.

[0110] S604: Calculate the target user's interest in each retained topic tag by combining the completion rate of learning content, learning time, and learning duration.

[0111]

[0112] Among them, I r C represents the target user's interest in the retained r-th topic tag. r,k T represents the completion rate of the target user's learning of the sub-learning resources corresponding to the retained r-th topic tag in the k-th learning behavior. r,k This represents the learning time of the target user on the sub-learning resource corresponding to the retained r-th topic tag in the k-th learning behavior, where K represents the total number of learning behaviors acquired, and t represents the learning duration. now t represents the current timestamp. k The timestamp represents the k-th learning action, ε represents a constant, and M represents the time stamp. r P represents the number of times the target user selected the r-th topic tag in their historical learning behavior, M represents the total number of historical learning behaviors, and P represents the total number of historical learning behaviors. r This represents the target user's preference value for the r-th topic tag, where α and β represent the first weight and the second weight, respectively.

[0113] The criterion for determining whether a learning behavior is a historical learning behavior is based on the current moment; that is, all learning behaviors generated before the current moment are considered historical learning behaviors. User learning content tags are extracted using a topic model, and the topic tags most relevant to the sub-learning resources are retained. Simultaneously, time decay weights are used to dynamically calculate interest levels, taking into account both recent learning behaviors and long-term interest preferences. This method ensures that recommended content accurately matches user needs, improving the relevance of new energy learning content recommendations and enhancing the system's adaptability to dynamic changes in user interests, thereby improving the personalization of recommendations and learning efficiency.

[0114] Specifically, a time decay weight is introduced to calculate the target user's interest in each retained topic tag, ensuring that recent learning behavior has a greater impact on interest. ε specifically represents a small constant to avoid a zero denominator. The introduction of Pr ensures that the user's long-term interest preferences are considered in the interest calculation process.

[0115] In practical applications, the weights used to measure recent and long-term learning behaviors can be determined based on the completion rate of each learning content in the user's recent and historical learning behaviors, namely the first weight and the second weight.

[0116]

[0117] in, Represents the Laplace transform, C recent (t) and C historical (t) represents the correlation function between the completion degree of the learning content of the user's recent learning behavior and the user's historical learning behavior and time t, respectively. That is, the completion degree of the learning content of the learning behavior corresponding to the timestamp t is a time series data of the completion degree of the learning content. α(s) and β(s) represent the first weight and the second weight under the Laplace complex frequency variable s, respectively.

[0118] In this model, the Laplace complex frequency variable is the independent variable in the Laplace transform. By introducing the Laplace complex frequency variable to calculate the weights, the temporal trends of both recent and historical behaviors can be considered simultaneously, and the weight proportions can be dynamically allocated. This method can not only capture rapid changes in user behavior but also smoothly handle the impact of long-term behaviors through frequency domain analysis, thereby improving the model's responsiveness to changes in interests and the accuracy of weight allocation.

[0119] It's important to note that the Laplace transform maps the distribution of learning behavior in the time domain to the frequency domain, capturing both long-term trends and short-term fluctuations in user learning behavior. In the frequency domain, the balance between recent and historical weights can be adjusted more appropriately to more accurately measure the importance of recent and historical learning behaviors, resulting in recommendations that better align with the target user's preferences. Optionally, when there isn't sufficient learning behavior data, α and β can be set to specific values ​​of 0.5 and 0.5 respectively to balance the weights of recent and long-term learning behaviors.

[0120] S7: Calculate the recommendation ratio of each type of sub-learning resource based on interest level in a positive correlation with the Nash equilibrium factor.

[0121] The Nash equilibrium factor is a weighting adjustment factor introduced from a game theory perspective. It is used to optimize the recommendation ratio, enabling the recommendation system to achieve a dynamic balance among different topics of user interest. The weight of the recommendation ratio is adjusted by calculating the deviation of each topic's interest level from the average interest level. If the interest level of a topic is significantly higher or lower than the average, its corresponding Nash equilibrium factor will increase, strengthening the recommendation ratio and causing users to focus more on highly interesting content or be guided to less interesting but important content.

[0122] It's worth noting that the proportional recommendation method dynamically adjusts the proportion of various sub-learning resources to meet users' diverse interests on different topics, avoiding a decline in the learning experience caused by overly singular recommendations. By broadening the recommendation scope, the system not only covers areas of potential user interest but also guides users to discover new knowledge points, enhancing the comprehensiveness and exploratory nature of learning. Simultaneously, it effectively improves the diversity and generalization of recommendations, thereby increasing user engagement and satisfaction.

[0123] In one possible implementation, the recommended ratio is calculated as follows:

[0124]

[0125] Among them, G r This represents the recommendation ratio of sub-learning resources corresponding to the r-th topic tag. Δ represents the average interest level of the target user in all retained topic tags. r I represents the target user's interest in the retained r-th topic tag. r and The difference between them, where R represents the total number of hashtags, and θ r Let represent the Nash equilibrium factor for the r-th topic tag.

[0126] It's important to note that by introducing a Nash equilibrium factor to dynamically adjust the recommendation ratio, a balance is achieved based on the difference between users' interest in different topic tags and the average interest level. This ensures that highly interested topics receive higher recommendation weights, while simultaneously guiding users to focus on less interesting but potentially important topics. This method guarantees the dynamism and balance of recommendations, satisfying users' core learning needs while expanding their exploration areas, thus improving the diversity, accuracy, and learning efficiency of recommendations.

[0127] S8: Recommend each sub-learning resource to the target user according to the corresponding recommendation ratio.

[0128] Specifically, by combining the calculated user interest in each topic tag with the Nash equilibrium factor, the recommendation ratio of various sub-learning resources is dynamically adjusted. Then, based on the recommendation ratio, sub-learning resources for the topic tags with the corresponding area or database proportion are displayed on each recommendation page. The extraction method for sub-learning resources can be random. This ensures that the recommended content prioritizes matching the user's high-interest topics while preserving the diversity and exploratory nature of the recommended content, ultimately leading to accurate recommendations to the user. For example, when recommending new energy learning resources, assuming the user has a high interest in "photovoltaic technology" (recommendation ratio of 50%), and "energy storage technology" and "electric vehicle technology" (recommendation ratios of 30% and 20% respectively), the recommendation page displays 5 photovoltaic-related content items, 3 energy storage content items, and 2 electric vehicle content items according to the ratios, with the specific sub-learning resources displayed being randomly selected from each category. This approach prioritizes meeting the user's learning needs for photovoltaic technology while preserving content related to energy storage and electric vehicles, enhancing the diversity and exploratory nature of the recommendations.

[0129] In practical applications, this approach acquires comprehensive and diverse new energy learning resources, combines autoencoder dimensionality reduction to minimize redundancy and retain core features, and uses soft allocation algorithms and K-means clustering to accurately segment learning resources, generating clear topic tags to enhance semantic expression. By dynamically calculating user learning behavior and optimizing the recommendation ratio using Nash equilibrium factors, a dynamic balance is achieved, ultimately recommending content proportionally. This improves the accuracy, diversity, and learning efficiency of recommendations, fully meeting users' personalized needs. It effectively avoids the problem of current recommendation schemes often failing to clearly distinguish the categories of learning content, leading to significant biases in recommended content. This approach captures user interests and recommends learning content based on these interests, rather than relying on pre-defined patterns, reducing the difficulty of information acquisition for target users and improving learning efficiency.

[0130] The beneficial effects of the technical solutions provided in the embodiments of the present invention include at least the following:

[0131] In this embodiment of the invention, an autoencoder is used to reduce the dimensionality of learning resources, thereby reducing redundancy and preserving core semantic information. Then, a soft allocation algorithm and k-means clustering algorithm are combined to perform self-clustering on the dimensionality-reduced learning resources. This improves the discriminative power of different clusters, making resource classification more accurate. Simultaneously, dynamically allocating probabilities strengthens the influence of high-confidence samples, avoiding the limitations of hard clustering. This allows for more accurate segmentation of learning content topics, helping the recommendation system capture users' genuine interest in various new energy technologies, improving the relevance and diversity of recommended content, and meeting personalized learning needs. Next, topic tags for each sub-learning resource obtained from clustering are generated. Based on the target user's multiple learning behaviors, the target user's interest in different topic tags (i.e., sub-learning resources) is calculated. Then, a positive correlation method with a Nash equilibrium factor is used to calculate the recommendation ratio of each type of sub-learning resource. Finally, based on the recommendation ratio, different categories of sub-learning resource content are extracted and recommended to the target user. This ensures that highly interesting content is prioritized while preserving diversity, ultimately accurately matching user learning needs. This method can more effectively capture users' genuine interest in new energy technologies, improve the relevance and exploratory nature of recommendations, and dynamically adapt to changes in user interests, significantly improving user learning efficiency and experience.

[0132] Reference manual attached Figure 2 The diagram shows a schematic representation of a new energy learning content recommendation system based on an adaptive strategy provided by the present invention.

[0133] This invention also provides a new energy learning content recommendation system 20 based on an adaptive strategy, applied to the aforementioned new energy learning content recommendation method based on an adaptive strategy, comprising:

[0134] Processor 201.

[0135] The memory 202 stores computer-readable instructions. When the computer-readable instructions are executed by the processor 201, they implement the new energy learning content recommendation method based on an adaptive strategy as described in the method embodiment.

[0136] The new energy learning content recommendation system 20 based on adaptive strategy provided by the present invention can execute the above-mentioned new energy learning content recommendation method based on adaptive strategy and achieve the same or similar technical effects. To avoid repetition, the present invention will not elaborate further.

[0137] The beneficial effects of the technical solutions provided in the embodiments of the present invention include at least the following:

[0138] In this embodiment of the invention, an autoencoder is used to reduce the dimensionality of learning resources, thereby reducing redundancy and preserving core semantic information. Then, a soft allocation algorithm and k-means clustering algorithm are combined to perform self-clustering on the dimensionality-reduced learning resources. This improves the discriminative power of different clusters, making resource classification more accurate. Simultaneously, dynamically allocating probabilities strengthens the influence of high-confidence samples, avoiding the limitations of hard clustering. This allows for more accurate segmentation of learning content topics, helping the recommendation system capture users' genuine interest in various new energy technologies, improving the relevance and diversity of recommended content, and meeting personalized learning needs. Next, topic tags for each sub-learning resource obtained from clustering are generated. Based on the target user's multiple learning behaviors, the target user's interest in different topic tags (i.e., sub-learning resources) is calculated. Then, a positive correlation method with a Nash equilibrium factor is used to calculate the recommendation ratio of each type of sub-learning resource. Finally, based on the recommendation ratio, different categories of sub-learning resource content are extracted and recommended to the target user. This ensures that highly interesting content is prioritized while preserving diversity, ultimately accurately matching user learning needs. This method can more effectively capture users' genuine interest in new energy technologies, improve the relevance and exploratory nature of recommendations, and dynamically adapt to changes in user interests, significantly improving user learning efficiency and experience.

[0139] 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.

[0140] 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).

[0141] 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.

[0142] 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.

[0143] 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.

[0144] 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.

[0145] 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.

[0146] 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.

[0147] 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.

[0148] 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.

[0149] 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.

[0150] 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 portion 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.

[0151] This invention provides a computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements the new energy learning content recommendation method based on an adaptive strategy as described in the method embodiment.

[0152] The present invention provides a computer-readable storage medium that can implement the steps and effects of the adaptive strategy-based new energy learning content recommendation method in the above-described method embodiments. To avoid repetition, the present invention will not elaborate further.

[0153] The beneficial effects of the technical solutions provided in the embodiments of the present invention include at least the following:

[0154] In this embodiment of the invention, an autoencoder is used to reduce the dimensionality of learning resources, thereby reducing redundancy and preserving core semantic information. Then, a soft allocation algorithm and k-means clustering algorithm are combined to perform self-clustering on the dimensionality-reduced learning resources. This improves the discriminative power of different clusters, making resource classification more accurate. Simultaneously, dynamically allocating probabilities strengthens the influence of high-confidence samples, avoiding the limitations of hard clustering. This allows for more accurate segmentation of learning content topics, helping the recommendation system capture users' genuine interest in various new energy technologies, improving the relevance and diversity of recommended content, and meeting personalized learning needs. Next, topic tags for each sub-learning resource obtained from clustering are generated. Based on the target user's multiple learning behaviors, the target user's interest in different topic tags (i.e., sub-learning resources) is calculated. Then, a positive correlation method with a Nash equilibrium factor is used to calculate the recommendation ratio of each type of sub-learning resource. Finally, based on the recommendation ratio, different categories of sub-learning resource content are extracted and recommended to the target user. This ensures that highly interesting content is prioritized while preserving diversity, ultimately accurately matching user learning needs. This method can more effectively capture users' genuine interest in new energy technologies, improve the relevance and exploratory nature of recommendations, and dynamically adapt to changes in user interests, significantly improving user learning efficiency and experience.

[0155] 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.

[0156] The following points need to be explained:

[0157] (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.

[0158] (2) For clarity, the thickness of layers or regions is enlarged or reduced in the drawings used to describe embodiments of the present invention; that is, these drawings are not drawn to actual 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.

[0159] (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.

[0160] 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 new energy learning content recommendation method based on an adaptive strategy, characterized in that, The methods include: S1: Obtain learning resources about new energy; S2: Dimensionality reduction of the learning resources is performed using an autoencoder; S3: Combine the soft allocation algorithm and the k-means clustering algorithm to perform self-clustering on the dimensionality-reduced learning resources to obtain multiple sub-learning resources; S4: Generate topic tags for various sub-learning resources; S5: Obtain multiple learning behaviors of the target user, wherein the learning behaviors include the learning content, completion rate of the learning content, learning time, and learning duration of each learning session; S6: Calculate the target user's interest in different topic tags based on each learning behavior; S7: Calculate the recommendation ratio of each type of sub-learning resource based on the interest level in a positive correlation manner with the Nash equilibrium factor; S8: Recommend each sub-learning resource to the target user according to the corresponding recommendation ratio; Specifically, S6 includes: S601: Extract learning content topic tags for each learning session using topic modeling; S602: Calculate the cosine similarity between the topic tags of each learning content and the topic tags of each type of sub-learning resource; S603: Retain the topic tags of the sub-learning resources that have the highest cosine similarity to the topic tags of each learning content; S604: Combining the completion rate of the learning content, the learning time, and the learning duration, calculate the target user's interest in each retained topic tag: ； in, This represents the target user's interest in the retained r-th topic tag. This represents the completion rate of the target user's learning of the sub-learning resources corresponding to the retained r-th topic tag in the k-th learning behavior. This represents the learning time of the target user on the sub-learning resource corresponding to the retained r-th topic tag in the k-th learning behavior, where K represents the total number of learning behaviors acquired. Indicates the current timestamp. The timestamp representing the k-th learning action. Represents a constant. This represents the number of times the target user selected the r-th topic tag in their historical learning behavior, and M represents the total number of historical learning behaviors. This represents the target user's preference value for the r-th topic tag, where α and β represent the first weight and the second weight, respectively. The specific method for calculating the recommended ratio is as follows: ； in, This represents the recommendation ratio of sub-learning resources corresponding to the r-th topic tag. This represents the average level of interest of the target user in all retained topic tags. This represents the target user's interest in the retained r-th topic tag. and The difference between them, where R represents the total number of topic tags. Let represent the Nash equilibrium factor for the r-th topic tag.

2. The method for recommending new energy learning content based on an adaptive strategy according to claim 1, characterized in that, The learning resources include e-books, research papers, academic journals, industry reports, case studies, and experimental tutorials.

3. The method for recommending new energy learning content based on an adaptive strategy according to claim 1, characterized in that, S2 specifically includes: S201: Convert the text content in the learning resources into a low-dimensional continuous vector using a weighted average. ； in, Let W represent a low-dimensional continuous vector representing each piece of text content, and let W represent the set of words in the text content. This represents the pre-trained word vector of word w. The frequency of word w in the text content is represented by , and a is a smoothing factor used to control the weight of high-frequency words. S202: Remove the first principal component of all low-dimensional continuous vectors from the low-dimensional continuous vectors of each text content to obtain optimized low-dimensional continuous vectors: ； in, This represents the optimized low-dimensional continuous vector of each text content obtained after removing the first principal component u from all low-dimensional continuous vectors. S203: The optimized low-dimensional continuous vector is mapped to a latent representation by the encoder of the autoencoder, thereby completing the dimensionality reduction of the learning resource.

4. The method for recommending new energy learning content based on an adaptive strategy according to claim 1, characterized in that, The training method for the autoencoder is as follows: The autoencoder is trained with the objective of minimizing the reconstruction error of all low-dimensional continuous vectors after removing the first principal component. The specific reconstruction error is as follows: ； in, Let i represent the optimized low-dimensional continuous vector of the i-th text content. N represents the total amount of text content. Decoder pair representing an autoencoder The reconstructed optimized low-dimensional continuous vector This indicates the reconstruction error.

5. The new energy learning content recommendation method based on adaptive strategy according to claim 3, characterized in that, S3 specifically includes: S301: The k-means clustering algorithm is used to cluster each potential representation once to obtain multiple category clusters; S302: Calculate the probability that each latent representation belongs to a different class cluster using a soft assignment algorithm: ； in, Let represent the probability that the latent representation of the i-th text content belongs to the j-th category cluster. The cluster index representing all categories of clusters. and Represent the j-th category cluster and the j-th category cluster, respectively. The center point of each category cluster This represents the latent representation of the i-th text content; S303: Generate the desired target probability with respect to the stated probability: ； in, The latent representation of the i-th text content represents the expected target probability of belonging to the j-th category cluster; S304: Update the center position of the class cluster with the objective of minimizing the KL divergence between the probability and the expected target probability. ； in, express and The KL divergence between them, where K represents the total number of clusters and log represents the logarithmic function; S305: Using the updated center position of each category cluster as the cluster center point, perform secondary clustering on each potential representation using the k-means clustering algorithm to obtain multi-class sub-learning resources.

6. The method for recommending new energy learning content based on an adaptive strategy according to claim 1, characterized in that, Specifically, S4 is: Topic tags for various sub-learning resources are generated through topic models, including LDA topic models and NMF topic models.

7. A new energy learning content recommendation system based on an adaptive strategy, characterized in that, include: processor; A memory storing computer-readable instructions, which, when executed by the processor, implement the new energy learning content recommendation method based on an adaptive strategy as described in any one of claims 1 to 6.

8. A computer-readable storage medium having a computer program stored thereon, characterized in that, When executed by the processor, the program implements the new energy learning content recommendation method based on an adaptive strategy as described in any one of claims 1 to 6.

Images