Method for generating personalized learning scheme based on large model architecture

Abstract

The invention relates to the technical field of wisdom education, and discloses a method for generating a personalized learning scheme based on a large model architecture, which is used for improving the pertinence and effectiveness of learning of learners. The personalized learning scheme generation method comprises the steps of collecting learning and social behavior data of a learner by obtaining a learning platform, calculating learning path variability to classify learning types and evaluate learning effects, calculating transition probability between knowledge nodes, generating a personalized learning path in combination with learning classification and effects, and generating a personalized learning scheme according to the personalized learning path. A target learning scheme is customized for each learner by analyzing learning motivation and social productivity and processing the data by using a large model architecture.

Description

Technical Field

[0001] This invention relates to the field of smart education technology, and in particular to a method for generating personalized learning programs based on a large model architecture. Background Technology

[0002] In existing technologies, the learning behavior data of learners on online learning platforms is generally collected and analyzed to identify learners' learning patterns, and then a learning plan is generated based on the identified learning patterns.

[0003] However, this technical solution mainly focuses on static learning behavior data, ignoring the dynamic changes in the learning process. In addition, the data considered by this technical solution is relatively singular. Therefore, the final learning plan may not fully reflect the learner's personalized needs and preferences, nor can it accurately predict and adjust the learning pace to maximize learning effectiveness. Summary of the Invention

[0004] This invention provides a method for generating personalized learning schemes based on a large model architecture, in order to solve the problem that the data considered in the prior art is relatively singular, resulting in poor learning performance.

[0005] The first aspect of this invention provides a method for generating personalized learning schemes based on a large model architecture, comprising: acquiring learning behavior data and social behavior data of multiple learners on a learning platform; calculating the learning path variability of each learner based on the learning behavior data of each learner; determining the learning type corresponding to each learner according to the learning path variability, and calculating the learning effect of the multiple learners to obtain the learning classification and effect of each learner; calculating the transfer probability between knowledge nodes on the learning path of each learner; generating personalized learning paths corresponding to each learner based on the learning classification and effect of each learner and the corresponding transfer probability; analyzing the learning motivation and social productivity of each learner according to the personalized learning paths and the social behavior data, and processing the learning motivation and social productivity of each learner based on the large model architecture to obtain the target learning scheme of each learner.

[0006] In one feasible implementation, the step of calculating the learning path variability of each learner based on their learning behavior data includes: organizing the learning behavior data of each learner into a personal time series for each learner, wherein the personal time series reflects the actual behavioral changes of the learner during the learning process; calculating the DTW distance and decay factor between the personal time series and a preset target time series, wherein the target time series refers to a time series representing a typical or expected learning behavior pattern constructed based on a large amount of historical data of learners or standards set by experts; and calculating the learning path variability of each learner based on the DTW distance and the decay factor.

[0007] In one feasible implementation, the step of determining the learning type of each learner based on the learning path variation and calculating the learning effect of each learner to obtain the learning classification and effect of the learners includes: performing cluster analysis on the learning path variation data of all learners to obtain a target clustering result, wherein the target clustering result is used to indicate the learning type corresponding to each learner; calculating the learning effect evaluation index value of each learner based on the learning behavior data of each learner, and determining the learning effect of each learner based on the learning effect evaluation index value of each learner.

[0008] In one feasible implementation, the step of performing cluster analysis on the learning path variation data of all learners to obtain the target clustering result includes: extracting features that have an important impact on the classification of learning types from the learning path variation data of all learners to obtain feature data; determining initial cluster centers based on the feature data; and assigning each data point to the nearest initial cluster center through iterative calculation until the clustering result is stable to obtain the target clustering result.

[0009] In one feasible implementation, generating personalized learning paths for each learner based on their learning classification, performance, and corresponding transition probabilities includes: selecting the knowledge point with the highest transition probability as the next learning content using an ant colony optimization algorithm, based on the learner's learning classification, performance, and corresponding transition probabilities; repeating the selection process until all knowledge points have been traversed, thus completing the learning path planning and generating personalized learning paths for each learner. In one feasible implementation, based on the learning classification and effectiveness of each learner and the corresponding transition probability, an ant colony optimization algorithm is used to select the knowledge point with the highest transition probability as the next learning content. This includes: setting up a virtual ant colony for each learner, with each ant representing a possible learning path explorer; allocating initial pheromone amounts to knowledge nodes according to each learner's learning classification and effectiveness, with the pheromone amount being proportional to the learner's mastery and interest; in each iteration step, the ants select the knowledge point with the highest transition probability as the next learning content based on the current node's transition probability, combined with pheromone concentration and heuristic information, until all ants have completed a full path exploration.

[0010] In one feasible implementation, the step of analyzing each learner's learning motivation and social productivity based on the personalized learning path and the social behavior data, and processing each learner's learning motivation and social productivity based on a large model architecture to obtain each learner's target learning plan, includes: analyzing each learner's learning motivation and social productivity based on the personalized learning path and the social behavior data; performing feature fusion based on each learner's learning motivation and social productivity to obtain a comprehensive feature vector for each learner; inputting each learner's comprehensive feature vector into the feature extraction layer of a large model based on a Transformer architecture, and outputting the fused learner feature representations after feature extraction and encoding; inputting the fused learner feature representations into the prediction layer of the Transformer model to generate a personalized learning rhythm, wherein the personalized learning rhythm includes learning stages, learning intensity, and learning intervals; and combining the personalized learning rhythm with the generated personalized learning path to obtain each learner's target learning plan, wherein the target learning plan includes specific learning tasks, time planning, and learning resource links.

[0011] A second aspect of the present invention provides an apparatus for generating personalized learning schemes, comprising: an acquisition module for acquiring learning behavior data and social behavior data of multiple learners on a learning platform; a first calculation module for calculating the learning path variability of each learner based on the learning behavior data of each learner; a first processing module for determining the learning type corresponding to each learner based on the learning path variability, and calculating the learning effect of the multiple learners to obtain the learning classification and effect of each learner; a second calculation module for calculating the transfer probability between knowledge nodes on the learning path of each learner; a generation module for generating a personalized learning path corresponding to each learner based on the learning classification and effect of each learner and the corresponding transfer probability; and a second processing module for analyzing the learning motivation and social productivity of each learner based on the personalized learning path and the social behavior data, and processing the learning motivation and social productivity of each learner based on a large model architecture to obtain the target learning scheme of each learner.

[0012] In one feasible implementation, the first calculation module is specifically used to: organize the learning behavior data of each learner into a personal time series for each learner, wherein the personal time series reflects the actual behavioral changes of the learner during the learning process; calculate the DTW distance and decay factor between the personal time series and a preset target time series, wherein the target time series refers to a time series that represents a typical or expected learning behavior pattern, constructed based on a large amount of historical data of learners or standards set by experts; and calculate the learning path variability of each learner based on the DTW distance and the decay factor.

[0013] In one feasible implementation, the first processing module includes: an analysis unit, configured to perform cluster analysis on the learning path variation data of all learners to obtain a target clustering result, wherein the target clustering result is used to indicate the learning type corresponding to each learner; and a determination unit, configured to calculate the learning effect evaluation index value of each learner based on the learning behavior data of each learner, and determine the learning effect of each learner based on the learning effect evaluation index value of each learner.

[0014] In one feasible implementation, the analysis unit is specifically used to: extract features that have an important impact on the classification of learning types from the learning path variation data of all learners, and obtain feature data; determine the initial cluster centers based on the feature data; and assign each data point to the nearest initial cluster center through iterative calculation until the clustering result is stable, thereby obtaining the target clustering result.

[0015] In one feasible implementation, the generation module includes: a selection unit, used to select the knowledge point with the highest knowledge node transfer probability as the next learning content based on the learning classification and effect of each learner and the corresponding transfer probability using an ant colony optimization algorithm; and a generation unit, used to repeat the selection process until all knowledge points have been traversed, complete the learning path planning, and generate a personalized learning path for each learner. In one feasible implementation, the selection unit is specifically used to: set up a virtual ant colony for each learner, with each ant representing a possible learning path explorer; allocate an initial pheromone amount to knowledge nodes based on each learner's learning classification and effect, the pheromone amount being proportional to the learner's mastery and interest; in each iteration step, the ants select the knowledge point with the highest transition probability as the next learning content based on the transition probability of the current node, combined with the pheromone concentration and heuristic information, until all ants complete the exploration of a complete path.

[0016] In one feasible implementation, the second processing module is specifically used for: analyzing the learning motivation and social productivity of each learner based on the personalized learning path and the social behavior data; performing feature fusion based on the learning motivation and social productivity of each learner to obtain a comprehensive feature vector for each learner; inputting the comprehensive feature vector of each learner into the feature extraction layer of a large model based on the Transformer architecture, and outputting the fused feature representation of each learner after feature extraction and encoding; inputting the fused feature representation of each learner into the prediction layer of the Transformer model to generate a personalized learning rhythm, wherein the personalized learning rhythm includes learning stages, learning intensity, and learning intervals; and combining the personalized learning rhythm with the generated personalized learning path to obtain the target learning plan for each learner, wherein the target learning plan includes specific learning tasks, time planning, and learning resource links.

[0017] A third aspect of the present invention provides an electronic device, comprising: a memory and at least one processor, wherein the memory stores instructions; the at least one processor invokes the instructions in the memory to cause the electronic device to execute the above-described method for generating a personalized learning scheme based on a large model architecture.

[0018] A fourth aspect of the present invention provides a computer-readable storage medium storing instructions that, when executed on a computer, cause the computer to perform the above-described method for generating a personalized learning scheme based on a large model architecture.

[0019] The technical solution provided by this invention involves acquiring learning behavior data and social behavior data of multiple learners on a learning platform; calculating the variability of each learner's learning path based on their learning behavior data; determining the learning type corresponding to each learner based on the learning path variability, and calculating the learning effects of the multiple learners to obtain each learner's learning classification and effect; calculating the transfer probability between knowledge nodes on each learner's learning path; generating a personalized learning path for each learner based on their learning classification, effect, and corresponding transfer probability; analyzing each learner's learning motivation and social productivity based on the personalized learning path and the social behavior data, and processing each learner's learning motivation and social productivity based on a large model architecture to obtain each learner's target learning plan. In this embodiment of the invention, by analyzing the learning behavior data and social behavior data of multiple learners on a learning platform, calculating the variability of learning paths to identify learning types and learning effects, and calculating the transfer probability between knowledge nodes, a learning path that meets the personalized needs of each learner is generated. Furthermore, by combining social behavior data to analyze the learner's learning motivation and social productivity, and by using a large model architecture to tailor a target learning plan for each learner, the targetedness and effectiveness of learning are effectively improved. Attached Figure Description

[0020] Figure 1 This is a schematic diagram of an embodiment of the method for generating personalized learning schemes based on a large model architecture in this invention. Figure 2 This is a schematic diagram of another embodiment of the method for generating personalized learning schemes based on a large model architecture in this invention. Figure 3 This is a schematic diagram of another embodiment of the method for generating personalized learning schemes based on a large model architecture in this invention. Figure 4 This is a schematic diagram of an embodiment of the device for generating personalized learning schemes based on a large model architecture in this invention. Figure 5 This is a schematic diagram of another embodiment of the device for generating personalized learning schemes based on a large model architecture in this invention. Figure 6 This is a schematic diagram of one embodiment of the electronic device in this invention. Detailed Implementation

[0021] This invention provides a method for generating personalized learning schemes based on a large model architecture. By comprehensively analyzing learning behavior and social behavior data, it formulates personalized learning paths and paces, achieving efficient and targeted learning goal planning.

[0022] The terms "first," "second," "third," "fourth," etc. (if present) in the specification, claims, and accompanying drawings of this invention are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" or "having" and any variations thereof are intended to cover a non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0023] It is understood that the executing entity of this invention can be a device for generating personalized learning schemes based on a large model architecture, or it can be a terminal or a server; the specific implementation is not limited here. This embodiment of the invention will be described using a server as an example.

[0024] For ease of understanding, the specific process of the embodiments of the present invention is described below. Please refer to [link / reference]. Figure 1 One embodiment of the method for generating personalized learning schemes based on large model architecture in this invention includes: 101. Obtain learning behavior data and social behavior data from multiple learners on the learning platform; In this embodiment, all information acquisition must strictly comply with data protection regulations and ethical guidelines to ensure that all information collection activities are conducted within a legal and compliant framework. Learning behavior data records learners' access to course resources, completed learning tasks, active participation in learning activities, specific scores on online tests, duration of study time, and continuous learning progress. This data reveals learners' learning habits, personal preferences, and overall ability levels. Simultaneously, social behavior data is collected, including learners' posting and replying on forums, likes and interactions, established friendships, and participation in community activities. This information helps understand learners' social network characteristics, communication tendencies, and potential learning needs.

[0025] 102. Calculate the variability of each learner's learning path based on their learning behavior data; Based on learning behavior data, the order and time of each learner's access to different knowledge nodes are recorded to construct each learner's learning path. Statistical methods are used to calculate the degree of deviation of these learning paths from the average learning path or the ideal learning path. This degree of deviation can be used as a quantitative indicator of learning path variability. The greater the degree of deviation, the higher the variability of the learner's learning path, and vice versa.

[0026] 103. Based on the variability of learning paths, determine the learning type corresponding to each learner, calculate the learning effect of multiple learners, and obtain the learning classification and effect of each learner. Learners with high learning path variability may exhibit greater learning flexibility and exploratory nature, while learners with low learning path variability may be more inclined to follow fixed learning paths.

[0027] Based on the variability of learning paths, cluster analysis or machine learning algorithms can be used to classify learners. For example, learners with high variability can be categorized as exploratory learners, who prefer free exploration and are not bound by fixed learning frameworks; learners with low variability can be categorized as structured learners, who tend to learn according to predetermined learning steps and plans. Furthermore, other learning behavior characteristics, such as learning duration and task completion rate, can be combined to further refine the classification of learning types.

[0028] After determining the learner's learning type, a learning effectiveness evaluation model is used to calculate the learner's learning effectiveness. By comprehensively evaluating multiple dimensions such as the learner's online test scores, task completion, and learning progress, and comparing the learner's mastery of different knowledge points and achievement of learning goals, a quantitative learning effectiveness indicator can be obtained.

[0029] 104. Calculate the transition probability between knowledge nodes on each learner's learning path; Based on learning behavior data, the order and frequency of each learner's access to different knowledge nodes on the learning platform are statistically analyzed to form a learning path dataset. For each learner, the frequency of their transfer from one knowledge node to another is calculated, which is the number of times the learner transfers between two nodes divided by the total number of transfers from the starting node. The transfer frequencies of all learners are summarized and averaged to obtain the transfer probability matrix between knowledge nodes. This matrix reflects the probability that a learner will transfer from one knowledge node to another on the learning path.

[0030] 105. Based on each learner's learning classification, effectiveness, and corresponding transition probability, generate a personalized learning path for each learner; Based on each learner's learning category and learning outcomes, their learning needs and goals are determined. Using a transition probability matrix between knowledge nodes, potential points of interest and difficulties along the learning path are analyzed. Personalized learning paths are designed based on learners' learning preferences and ability levels. For exploratory learners, more opportunities for independent exploration and discovery are provided; for structured learners, a more ordered and systematic learning framework is offered. Simultaneously, the order and connections of knowledge nodes in the learning path are adjusted according to the transition probabilities to ensure learners can smoothly transition from one knowledge point to another, improving learning efficiency and effectiveness. Ultimately, the generated learning paths not only meet the learners' individual needs but also effectively utilize the transition probability information from their learning behavior data, providing learners with a customized learning experience.

[0031] 106. Based on personalized learning paths and social behavior data, analyze the learning motivation and social productivity of each learner, and process the learning motivation and social productivity of each learner based on a large model architecture to obtain the target learning plan for each learner.

[0032] Based on information such as the access order, dwell time, and revisit frequency of knowledge nodes in a personalized learning path, a learning motivation diagnostic model is used to identify learners' intrinsic interests, extrinsic drives, and self-efficacy. Simultaneously, by analyzing interaction frequency, community roles, and collaborative outcomes in social behavior data, social network analysis is employed to assess learners' social influence, teamwork ability, and willingness to share knowledge—indicators of social productivity. The analysis results of learning motivation and social productivity are input into a pre-built large-scale model architecture. This architecture integrates the deep feature extraction capabilities of deep learning, the semantic understanding advantages of natural language processing, and the generalization performance of large-scale data training, thus possessing data parsing and pattern recognition capabilities. This large-scale model architecture can not only process and analyze complex data in learners' personalized learning paths, such as access patterns and dwell time distributions of knowledge nodes, but also delve into the potential connections in social behavior data, such as interactions... The model examines the correlation between frequency and community roles, and how collaborative outcomes reflect teamwork capabilities. This large-scale model architecture allows for a deeper understanding of learners' motivations, including their intrinsic interests, sources of extrinsic drive, and levels of self-efficacy. It also enables a more accurate assessment of their social productivity, encompassing the breadth of their social influence, the depth of their teamwork, and their enthusiasm for knowledge sharing. Furthermore, pre-built algorithms, such as collaborative filtering and deep learning recommendation systems, intelligently match the most suitable learning resources and paths, ensuring that learning content aligns with learners' interests and needs while effectively promoting their learning outcomes. Additionally, the large-scale model architecture possesses predictive capabilities, forecasting learners' potential performance and needs in future learning based on pre-set historical data, providing strong support for developing forward-looking learning plans. Therefore, this large-scale model architecture can automatically generate customized learning programs that both match learners' learning motivations and fully leverage their social productivity.

[0033] In this embodiment of the invention, by comprehensively collecting learning behavior data and social behavior data from multiple learners on a learning platform, the variability of learning paths is calculated to determine the learning type and evaluate the learning effect. Then, the transfer probability between knowledge nodes is calculated to generate a personalized learning path for each learner. At the same time, by combining the analysis of learning motivation and social productivity, and by leveraging the deep processing capabilities of a large model architecture, a more tailored target learning plan that fits each learner's personality characteristics and social environment can be customized, significantly improving the relevance and effectiveness of learners' learning, thereby enhancing learning outcomes.

[0034] Please see Figure 2 Another embodiment of the method for generating personalized learning schemes based on large model architecture in this invention includes: 201. Obtain learning behavior data and social behavior data from multiple learners on the learning platform; The execution process of steps 201 and 101 is similar and will not be described again here. 202. Calculate the variability of each learner's learning path based on their learning behavior data; Learning behavior data is organized into individual time series for each learner, which reflects the actual behavioral changes of learners in the learning path. The DTW distance and decay factor between the individual time series and the preset target time series are calculated. The target time series refers to a time series that represents a typical or expected learning path, constructed based on a large amount of historical data of learners or standards set by experts. The variability of each learner's learning path is obtained by calculating the DTW distance and decay factor.

[0035] Organizing learning behavior data into individual time series for each learner refers to recording and organizing each learner's behavioral data on the learning platform in chronological order to form a time series reflecting the actual changes in their behavior during the learning process. Specifically, firstly, it is necessary to collect various behavioral data of learners during the learning process, such as login time, online learning duration, number of course visits, homework completion status, test scores, forum interaction records, etc. Then, these data are arranged in chronological order to ensure that the behavioral data corresponding to each time point is accurate. Finally, each learner's individual time series will clearly show the trend of their behavioral changes during the learning process.

[0036] Define a distance metric between two time series, such as Euclidean distance. Then, create a distance matrix by calculating the distance between any two points in the individual's time series and the target time series. Next, starting from the top left corner of the distance matrix, calculate the minimum distance to each point along the shortest path, forming a cumulative distance matrix. The final value in this cumulative distance matrix is ​​the DTW distance. The decay factor is typically used to consider the influence of time factors in the time series. Its calculation method can be determined based on the specific problem and algorithm implementation requirements; for example, it can be calculated using an exponential decay function. In calculating the DTW distance, the decay factor can be used to weight the distance to reflect the impact of time factors on data similarity. Based on the weighted DTW distance, assess the variability of each learner's learning path; a larger distance indicates higher variability, meaning a greater difference between the learner's behavioral pattern and the expected pattern.

[0037] The variation in learners' learning paths can be calculated using the following formula:

[0038] in, This indicates variations in learners' learning paths. This represents the DTW distance between two time series. This is the attenuation factor.

[0039] The formula for calculating the attenuation factor α is:

[0040] in, This represents the diagonal length of the optimal path. This represents the number of nodes along the optimal path.

[0041] DTW distance between two time series The calculation formula is:

[0042] in, The distance between the two time series is represented by DTW, s represents the time series position in the learner's learning path R1, and t represents the time series position in the expected learning path R2.

[0043] 203. Perform cluster analysis on the learning path variation data of all learners to obtain the target clustering results, which are used to indicate the learning type of each learner; Features that significantly influence the classification of learning types are extracted from the learning path variation data of all learners to obtain feature data; initial cluster centers are determined based on the feature data; each data point is assigned to the nearest initial cluster center through iterative calculation until the clustering results stabilize, and the target clustering result is obtained.

[0044] By utilizing statistical analysis and domain knowledge, features closely related to learning behavior patterns, such as learning duration, learning frequency, and task completion rate, are selected to construct a feature dataset. Then, the K-means++ algorithm is used to initialize cluster centers to ensure a more uniform distribution of the initial centers. Next, through iterative calculation, the allocation of each data point to its nearest cluster center is continuously updated, and the cluster centers are recalculated until the clustering results converge. Clustering results convergence is indicated by the cluster centers no longer changing or the change being less than a preset threshold, ultimately yielding a stable target clustering result, thereby achieving an effective classification of learners' learning types.

[0045] 204. Based on the learning behavior data of each learner, calculate the learning effect evaluation index value of each learner, and determine the learning effect of each learner according to the learning effect evaluation index value. Calculate learners' learning effectiveness data, i.e., learning outcomes. The calculation formula is:

[0046] in, T and T represent the duration of knowledge point h and the total course duration, respectively. This indicates the degree to which learner a has mastered knowledge point h.

[0047] level of mastery The calculation formula is:

[0048] in, This indicates learner a's mastery of knowledge point h. This indicates the interaction between learner a and knowledge point h.

[0049] Get to know the situation The calculation formula is:

[0050] in, This indicates the correlation between h and u; This represents learner a's score on test item u.

[0051] Interaction status The calculation formula is:

[0052] in, , , Represented as weighting factor, This represents the total interaction between learner a and knowledge point h. This represents the student interaction between learner a and knowledge point h. Teacher-student interaction between learner a and knowledge point h.

[0053] 205. Calculate the transition probability between knowledge nodes on each learner's learning path; 206. Based on each learner's learning classification, effectiveness, and corresponding transition probability, generate a personalized learning path for each learner; Steps 204-206 are similar to steps 103-105 above, and will not be described again here.

[0054] 207. Based on personalized learning paths and social behavior data, analyze the learning motivation and social productivity of each learner, and process the learning motivation and social productivity of each learner based on a large model architecture to obtain the target learning plan for each learner.

[0055] Based on personalized learning paths and social behavior data, the learning motivation and social productivity of each learner are analyzed. Features are fused based on each learner's motivation and productivity to obtain a comprehensive feature vector for each learner. This comprehensive feature vector is then input into the feature extraction layer of a large model based on the Transformer architecture. After feature extraction and encoding, the fused learner feature representations are output. These fused learner feature representations are then input into the prediction layer of the Transformer model to generate a personalized learning rhythm, which includes learning stages, learning intensity, and learning intervals. Finally, the personalized learning rhythm is combined with the generated personalized learning path to obtain the target learning plan for each learner, which includes specific learning tasks, time planning, and links to learning resources.

[0056] Pattern recognition is performed on the data in the personalized learning path to identify the learning motivation of each learner. Based on the social behavior data, the interaction frequency, influence, and cooperation tendency of each learner in the community are analyzed to assess the social productivity of each learner. Specifically, for the data in the personalized learning path, machine learning algorithms are used for pattern recognition. Through techniques such as cluster analysis and association rule mining, the behavioral characteristics of learners in the learning process, such as access frequency, learning duration, and preferred content, are analyzed to identify the intrinsic learning motivation of each learner. Secondly, based on the social behavior data, social network analysis technology is used to quantify the interaction frequency, influence, and cooperation tendency of learners in the community. By calculating indicators such as node centrality and PageRank, the social productivity of learners in the social environment is comprehensively evaluated.

[0057] The features related to learning motivation and social productivity are organized and standardized. Features related to learning motivation include, but are not limited to, learning preferences, goal orientation, and interests; features related to social productivity include, but are not limited to, interaction frequency, community influence, and collaborative ability. Feature fusion techniques are then used to organically combine these features. These techniques include, but are not limited to, concatenation, weighted summation, or attention mechanisms. Specifically, a fusion model can be constructed, taking learning motivation features and social productivity features as input. Through the model's internal computational logic, such as nonlinear transformations and interactions in neural network layers, a comprehensive feature vector is output. This comprehensive feature vector not only contains the learner's intrinsic learning motivation information but also incorporates their productivity performance in a social environment.

[0058] The Transformer model, through its unique self-attention and multi-head self-attention mechanisms, processes the combined feature vectors of each learner in parallel. In the feature extraction layer, the model can capture long-term dependencies and multi-faceted feature information in the combined feature vectors, such as the learner's semantic and syntactic features. At the same time, positional encoding technology is also used to encode the positional information in the combined feature vectors into the model, helping the model to better understand the sequential relationships between features. After this feature extraction and encoding process, the Transformer model can finally output the fused feature representations of each learner. These representations not only contain the information of the original feature vectors but also incorporate the model's understanding and extraction of features.

[0059] The prediction layer receives feature representations as input and uses the sequence modeling capabilities of the Transformer model to analyze and understand these features. Through the internal computation and inference processes of the model, the prediction layer can identify the learner's learning preferences, ability level, and optimal learning mode. Based on this information, the model generates a personalized learning rhythm, including: 1. Determining the appropriate learning stage for the learner, such as beginner, intermediate, and advanced; 2. Setting a reasonable learning intensity, such as daily study time and task difficulty; 3. Planning appropriate learning intervals, such as rest time and review cycles.

[0060] Based on the pre-set learning objectives and content framework in the personalized learning path, a series of specific learning tasks are customized for each learner. These tasks cover different knowledge points and skills, and the learning stages, learning intensity, and learning breaks in the personalized learning rhythm are integrated into the time planning. Reasonable time nodes and durations are allocated to each learning task to ensure that learners can gradually progress at their own pace. At the same time, in order to support learners' self-directed learning, corresponding learning resource links are provided for each learning task, including video tutorials, online courses, e-books, etc., so that learners can access the learning materials they need at any time.

[0061] In this embodiment of the invention, by comprehensively collecting learning and social behavior data of learners on the learning platform, calculating the variability of learning paths and performing cluster analysis, the learning types of learners are identified. At the same time, by combining the evaluation of learning effectiveness and the calculation of knowledge node transfer probability, a personalized learning path is tailored for each learner. By analyzing learning motivation and social productivity, and relying on the deep processing capabilities of the large model architecture, a target learning plan that is highly tailored to the characteristics and needs of learners is generated, which effectively improves the personalization, efficiency and practicality of learning and helps learners achieve better learning outcomes.

[0062] Please see Figure 3Another embodiment of the method for generating personalized learning schemes based on large model architecture in this invention includes: 301. Obtain learning behavior data and social behavior data from multiple learners on the learning platform; 302. Calculate the variability of each learner's learning path based on their learning behavior data; 303. Based on the variability of learning paths, determine the learning type corresponding to each learner, calculate the learning effect of multiple learners, and obtain the learning classification and effect of each learner; 304. Calculate the transition probability between knowledge nodes on each learner's learning path; Steps 301-304 are similar to steps 101-104 above, and will not be described again here.

[0063] 305. Based on the learning classification and effect of each learner and the corresponding transition probability, the ant colony optimization algorithm is used to select the knowledge point with the highest transition probability as the next learning content. A virtual ant colony is set up for each learner, with each ant representing a potential learning path explorer. Based on each learner's learning classification and effectiveness, an initial pheromone amount is allocated to knowledge nodes, with the pheromone amount being proportional to the learner's mastery and interest. In each iteration step, the ants select the knowledge point with the highest transition probability as the next learning content based on the transition probability of the current node, combined with pheromone concentration and heuristic information, until all ants have completed the exploration of a complete path.

[0064] Based on the learner's learning classification and effectiveness, an initial pheromone level is allocated to each knowledge node. The amount of pheromone directly reflects the learner's mastery and interest in the knowledge point; the higher the mastery and the greater the interest, the more initial pheromone is allocated. Subsequently, the algorithm enters an iterative process. Each ant, based on the transition probability of its current node, combined with the node's pheromone concentration and heuristic information such as the relevance and importance of knowledge points, selects the next knowledge point to visit. The calculation of the transition probability considers not only the direct connections between knowledge points but also incorporates the learner's personalized characteristics, ensuring that the selected learning path aligns with the learner's interests and abilities while covering key knowledge points. This selection process continues until all ants have completed a full path exploration from the starting point to the end point. In this way, the ant colony optimization algorithm can gradually converge to the optimal or near-optimal learning path, providing learners with personalized suggestions for the next learning content.

[0065] The probability of knowledge transfer between knowledge nodes on the learning path The specific calculation formula is as follows:

[0066] in, Indicates the initial pheromone concentration. Representing heuristic information, and Heuristic information and pheromone control parameters are used to intervene in the learning path from knowledge point c to d, so as to influence the transfer of knowledge point c to d.

[0067] Initial pheromone concentration The calculation formula is:

[0068] Where c and d both represent knowledge points, This represents the number of learning paths from knowledge point c to knowledge point d, that is, the number of times a student learns knowledge point c first and then learns knowledge point d.

[0069] Computational heuristic information The specific formula is:

[0070] in, This indicates the learning effectiveness of the learning path from knowledge point c to d.

[0071] 306. Repeat the selection process until all knowledge points have been traversed, complete the learning path planning, and generate personalized learning paths for each learner. When using ant colony optimization (ACO) for personalized learning path planning, the repeated selection process is implemented iteratively. Specifically, in each iteration, ants in the virtual ant colony select the next knowledge point based on the transition probability and pheromone concentration of the current node, until all ants have completed a full path exploration from the starting point to the destination. This process is repeated continuously, while the pheromone level is dynamically adjusted based on the learner's learning classification and performance to ensure that the ants consider the learner's personalized characteristics when selecting a path. As the number of iterations increases, more and more knowledge points are traversed until all knowledge points have been traversed at least once. At this point, the algorithm generates personalized learning paths for each learner based on the ants' exploration results.

[0072] 307. Based on personalized learning paths and social behavior data, analyze the learning motivation and social productivity of each learner, and process the learning motivation and social productivity of each learner based on a large model architecture to obtain the target learning plan for each learner.

[0073] Step 307 is similar to the execution process of step 106 above, and will not be described again here.

[0074] In this embodiment of the invention, by using the ant colony optimization algorithm combined with learners' learning classification, effects, and social behavior data, a personalized learning path is precisely planned, effectively mining learners' learning motivation and assessing their social productivity, thereby tailoring a learning pace to ensure that each learner can master knowledge in the most suitable way and at the most suitable speed, greatly improving the pertinence and efficiency of learning.

[0075] The method for generating personalized learning schemes based on large model architecture in embodiments of the present invention has been described above. The apparatus for generating personalized learning schemes based on large model architecture in embodiments of the present invention will be described below. Please refer to [link to relevant documentation]. Figure 4 One embodiment of the device for generating personalized learning schemes based on a large model architecture in this invention includes: Module 401 is used to acquire learning behavior data and social behavior data of multiple learners on the learning platform; The first calculation module 402 is used to calculate the variability of each learner's learning path based on each learner's learning behavior data; The first processing module 403 is used to determine the learning type corresponding to each learner based on the variability of the learning path, and to calculate the learning effect of multiple learners to obtain the learning classification and effect of each learner. The second calculation module 404 is used to calculate the transition probability between knowledge nodes on each learner's learning path; The generation module 405 is used to generate personalized learning paths for each learner based on their learning classification, performance, and corresponding transition probabilities. The second processing module 406 is used to analyze the learning motivation and social productivity of each learner based on personalized learning paths and social behavior data, and to process the learning motivation and social productivity of each learner based on the large model architecture to obtain the target learning plan for each learner.

[0076] In this embodiment of the invention, by comprehensively collecting learning behavior data and social behavior data from multiple learners on a learning platform, the variability of learning paths is calculated to determine the learning type and evaluate the learning effect. Then, the transfer probability between knowledge nodes is calculated to generate a personalized learning path for each learner. At the same time, by combining the analysis of learning motivation and social productivity, and by leveraging the deep processing capabilities of a large model architecture, a more tailored target learning plan that fits each learner's personality characteristics and social environment can be customized, significantly improving the relevance and effectiveness of learners' learning, thereby enhancing learning outcomes.

[0077] Please see Figure 5 Another embodiment of the device for generating personalized learning schemes based on a large model architecture in this invention includes: Module 401 is used to acquire learning behavior data and social behavior data of multiple learners on the learning platform; The first calculation module 402 is used to calculate the variability of each learner's learning path based on each learner's learning behavior data; The first processing module 403 is used to determine the learning type corresponding to each learner based on the variability of the learning path, and to calculate the learning effect of multiple learners to obtain the learning classification and effect of each learner. The second calculation module 404 is used to calculate the transition probability between knowledge nodes on each learner's learning path; The generation module 405 is used to generate personalized learning paths for each learner based on their learning classification, performance, and corresponding transition probabilities. The second processing module 406 is used to analyze the learning motivation and social productivity of each learner based on personalized learning paths and social behavior data, and to process the learning motivation and social productivity of each learner based on the large model architecture to obtain the target learning plan for each learner.

[0078] Optionally, the first calculation module 402 can be specifically used for: Learning behavior data is organized into individual time series for each learner, which reflects the actual behavioral changes of learners during the learning process. The DTW distance and decay factor between the individual time series and the preset target time series are calculated. The target time series refers to a time series that represents a typical or expected learning behavior pattern, constructed based on a large amount of historical data of learners or standards set by experts. The variability of each learner's learning path is obtained by calculating the DTW distance and decay factor.

[0079] Optionally, the first processing module 403 includes: Analysis unit 4031 is used to perform cluster analysis on the learning path variation data of all learners to obtain the target clustering result, which is used to indicate the learning type of each learner. The determination unit 4032 is used to calculate the learning effect evaluation index value of each learner based on the learning behavior data of each learner, and to determine the learning effect of each learner based on the learning effect evaluation index value of each learner.

[0080] Optionally, the analysis unit 4031 can be specifically used for: Features that significantly influence the classification of learning types are extracted from the learning path variation data of all learners to obtain feature data; initial cluster centers are determined based on the feature data; each data point is assigned to the nearest initial cluster center through iterative calculation until the clustering results stabilize, and the target clustering result is obtained.

[0081] Optionally, the generation module 405 is specifically used for: Selection unit 4051 is used to select the knowledge point with the highest knowledge node transfer probability as the next learning content based on the learning classification and effect of each learner and the corresponding transfer probability, using the ant colony optimization algorithm. The generation unit 4052 is used to repeat the selection process until all knowledge points have been traversed, complete the learning path planning, and generate personalized learning paths for each learner. Optionally, selection unit 4051 can be specifically used for: A virtual ant colony is set up for each learner, with each ant representing a potential learning path explorer. Based on each learner's learning classification and effectiveness, an initial pheromone amount is allocated to knowledge nodes, with the pheromone amount being proportional to the learner's mastery and interest. In each iteration step, the ants select the knowledge point with the highest transition probability as the next learning content based on the transition probability of the current node, combined with pheromone concentration and heuristic information, until all ants have completed the exploration of a complete path.

[0082] Optionally, the second processing module can be specifically used for: Based on personalized learning paths and social behavior data, the learning motivation and social productivity of each learner are analyzed. Features are fused based on each learner's motivation and productivity to obtain a comprehensive feature vector for each learner. This comprehensive feature vector is then input into the feature extraction layer of a large model based on the Transformer architecture. After feature extraction and encoding, the fused learner feature representations are output. These fused learner feature representations are then input into the prediction layer of the Transformer model to generate a personalized learning rhythm, which includes learning stages, learning intensity, and learning intervals. Finally, the personalized learning rhythm is combined with the generated personalized learning path to obtain the target learning plan for each learner, which includes specific learning tasks, time planning, and links to learning resources.

[0083] In this embodiment of the invention, by analyzing learners' learning and social behavior data, the variability of learning paths is calculated, learning types are classified and learning effects are evaluated. At the same time, ant colony optimization algorithm is used to plan personalized learning paths, and learners' learning motivations are further identified and their social productivity is evaluated. Based on this, personalized learning rhythms are customized, and with the deep processing capabilities of the large model architecture, a target learning plan that is more in line with each learner's personality characteristics and social environment is customized, which greatly improves the pertinence and efficiency of learning.

[0084] above Figure 4 and Figure 5The device for generating personalized learning schemes based on large model architecture in this embodiment of the invention will be described in detail from the perspective of modular functional entities. The electronic device in this embodiment of the invention will be described in detail from the perspective of hardware processing.

[0085] See Figure 6 As shown, the electronic device includes a processor 600 and a memory 601. The memory 601 stores machine-executable instructions that can be executed by the processor 600. The processor 600 executes the machine-executable instructions to implement the above-mentioned method for generating personalized learning schemes.

[0086] Furthermore, Figure 6 The electronic device shown also includes a bus 602 and a communication interface 603. The processor 600, the communication interface 603 and the memory 601 are connected via the bus 602.

[0087] The memory 601 may include high-speed random access memory (RAM) and may also include non-volatile memory, such as at least one disk storage device. Communication between this system network element and at least one other network element is achieved through at least one communication interface 603 (which can be wired or wireless), such as the Internet, wide area network, local area network, metropolitan area network, etc. The bus 602 may be an ISA bus, PCI bus, or EISA bus, etc. The bus can be divided into address bus, data bus, control bus, etc. For ease of representation, Figure 6 The symbol is represented by a single double-headed arrow, but this does not mean that there is only one bus or one type of bus.

[0088] The processor 600 may be an integrated circuit chip with signal processing capabilities. In implementation, each step of the above method can be completed by the integrated logic circuitry in the hardware of the processor 600 or by instructions in software form. The processor 600 may be a general-purpose processor, including a central processing unit (CPU), a network processor (NP), etc.; it may also be a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components. It can implement or execute the methods, steps, and logic block diagrams disclosed in the embodiments of this disclosure. The general-purpose processor may be a microprocessor or any conventional processor. The steps of the methods disclosed in the embodiments of this disclosure can be directly embodied in the execution of a hardware decoding processor, or executed by a combination of hardware and software modules in the decoding processor. The software modules may reside in random access memory, flash memory, read-only memory, programmable read-only memory, electrically erasable programmable memory, registers, or other mature storage media in the art. The storage medium is located in memory 601. Processor 600 reads the information in memory 601 and, in conjunction with its hardware, completes the method steps of the aforementioned embodiment.

[0089] The present invention also provides an electronic device, the computer device including a memory and a processor, wherein the memory stores computer-readable instructions, and when the computer-readable instructions are executed by the processor, the processor performs the steps of the method for generating the personalized learning scheme in the above embodiments. The present invention also provides a computer-readable storage medium, which can be a non-volatile computer-readable storage medium or a volatile computer-readable storage medium, wherein the computer-readable storage medium stores instructions, and when the instructions are executed on a computer, the computer performs the steps of the method for generating the personalized learning scheme based on a large model architecture.

[0090] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0091] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present 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.

[0092] The above-described embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for generating personalized learning schemes based on a large model architecture, characterized in that, The method for generating personalized learning schemes based on large model architecture includes: Acquire learning behavior data and social behavior data from multiple learners on the learning platform; Calculate the variability of each learner's learning path based on each learner's learning behavior data; Based on the variability of the learning path, the learning type corresponding to each learner is determined, and the learning effect of the multiple learners is calculated to obtain the learning classification and effect of each learner. Calculate the transition probability between knowledge nodes on each learner's learning path; Based on each learner's learning classification, effectiveness, and corresponding transition probability, a personalized learning path is generated for each learner. Based on the personalized learning path and the social behavior data, the learning motivation and social productivity of each learner are analyzed, and the learning motivation and social productivity of each learner are processed based on the large model architecture to obtain the target learning plan for each learner.

2. The method for generating personalized learning schemes based on large model architecture according to claim 1, characterized in that, The calculation of learning path variability for each learner based on their learning behavior data includes: The learning behavior data of each learner is compiled into a personal time series for each learner, which reflects the actual behavioral changes of the learner during the learning process; Calculate the DTW distance and decay factor between the individual time series and the preset target time series. The target time series refers to a time series that represents a typical or expected learning behavior pattern, constructed based on a large amount of historical data from learners or standards set by experts. The learning path variability of each learner is calculated based on the DTW distance and the decay factor.

3. The method for generating personalized learning schemes based on a large model architecture according to claim 1, characterized in that, The process of determining the learning type of each learner based on the variability of the learning path and calculating the learning effect of each learner to obtain the learner's learning classification and effect includes: Cluster analysis is performed on the learning path variation data of all learners to obtain the target clustering result, which is used to indicate the learning type corresponding to each learner; Based on the learning behavior data of each learner, the learning effectiveness evaluation index value of each learner is calculated, and the learning effectiveness of each learner is determined according to the learning effectiveness evaluation index value.

4. The method for generating personalized learning schemes based on large model architecture according to claim 3, characterized in that, The clustering analysis of all learners' learning path variation data to obtain the target clustering results includes: Feature data is obtained by extracting features that have a significant impact on the classification of learning types from the learning path variation data of all learners. Initial cluster centers are determined based on the aforementioned feature data; Each data point is assigned to the nearest initial cluster center through iterative calculation until the clustering results stabilize, thus obtaining the target clustering result.

5. The method for generating personalized learning schemes based on large model architecture as described in claim 1, characterized in that, The process of generating personalized learning paths for each learner based on their learning classification, performance, and corresponding transition probabilities includes: Based on the learning classification and effect of each learner and the corresponding transition probability, the ant colony optimization algorithm is used to select the knowledge point with the highest transition probability as the next learning content. Repeat the selection process until all knowledge points have been traversed, complete the learning path planning, and generate personalized learning paths for each learner.

6. The method for generating personalized learning schemes based on large model architecture according to claim 5, characterized in that, Based on the learning classification and effectiveness of each learner and the corresponding transition probability, the ant colony optimization algorithm is used to select the knowledge point with the highest transition probability as the next learning content, including: Each learner is given a virtual ant colony, with each ant representing a potential learner exploring a learning path. Based on each learner's learning category and results, an initial pheromone level is allocated to each knowledge node. The pheromone level is proportional to the learner's mastery and interest. In each iteration, the ants select the knowledge point with the highest transition probability as the next learning content based on the transition probability of the current node, combined with pheromone concentration and heuristic information, until all ants have completed the exploration of a complete path.

7. The method for generating personalized learning schemes based on large model architecture according to claim 1, characterized in that, Based on the personalized learning paths and social behavior data, the learning motivation and social productivity of each learner are analyzed. Then, based on a large model architecture, the learning motivation and social productivity of each learner are processed to obtain the target learning plan for each learner, including: Based on the personalized learning paths and the social behavior data, the learning motivation and social productivity of each learner are analyzed. Based on the learning motivation and social productivity of each learner, a comprehensive feature vector of each learner is obtained by fusing features. The combined feature vectors of each learner are input into the feature extraction layer of a large model based on the Transformer architecture. After feature extraction and encoding, the fused feature representations of each learner are output. The fused learner feature representations are input into the prediction layer of the Transformer model to generate a personalized learning rhythm, which includes learning phases, learning intensity, and learning intervals. By combining personalized learning pace with the generated personalized learning path, a target learning plan for each learner is obtained. The target learning plan includes specific learning tasks, time planning, and links to learning resources.

8. A device for generating personalized learning schemes based on a large model architecture, characterized in that, The device for generating the personalized learning scheme includes: The acquisition module is used to acquire learning behavior data and social behavior data from multiple learners on the learning platform; The first calculation module is used to calculate the variability of each learner's learning path based on each learner's learning behavior data; The first processing module is used to determine the learning type corresponding to each learner based on the variability of the learning path, and to calculate the learning effect of the multiple learners to obtain the learning classification and effect of each learner. The second calculation module is used to calculate the transition probability between knowledge nodes on each learner's learning path; The generation module is used to generate personalized learning paths for each learner based on their learning classification, performance, and corresponding transition probabilities. The second processing module is used to analyze the learning motivation and social productivity of each learner based on the personalized learning path and the social behavior data, and to process the learning motivation and social productivity of each learner based on the large model architecture to obtain the target learning plan of each learner.

9. An electronic device, characterized in that, The electronic device includes: a memory and at least one processor, wherein the memory stores instructions; The at least one processor invokes the instructions in the memory to cause the electronic device to execute the method for generating a personalized learning scheme based on a large model architecture as described in any one of claims 1-7.

10. A computer-readable storage medium storing instructions thereon, characterized in that, When the instructions are executed by the processor, they implement the method for generating personalized learning schemes based on large model architecture as described in any one of claims 1-7.

Images